Introduction
Bergeron’s Hook-Components and Coefficient-Length conjectures [4 Conjectures 2 and 3] describe how the symmetric-group isotypes of multivariate diagonal harmonics fit together. The first recovers every hook-label character from the alternating character by elementary skewing. The second gives the exact number of alphabets needed to retain every Schur coefficient in an arbitrary label isotype. We prove both conjectures for the Vandermonde polarization module in all ranks and for full diagonal harmonics, with derivative maps and finite-alphabet reconstruction.
We also answer Church–Ellenberg–Farb’s eventual-character and dimension-polynomial questions [11 Problems 1.10 and 1.12] , construct natural Schur presentations and resolutions for tensor coinvariants of arbitrary augmented algebras, and prove uniform Vandermonde generation ranges. An idempotent specialization identifies every first unstable character correction. The hook transfer gives the fully graded Theta sign component.
The hook-component and coefficient-length conjectures
Let K K K be a field of characteristic zero, let n , k ≥ 1 n,k \ge1 n , k ≥ 1 , and put V = K k V=K^k V = K k . Define
P n , k = Sym ( V ⊗ K n ) = K [ x i ( a ) : i ∈ [ n ] , a ∈ [ k ] ] , R n , k = P n , k / ⟨ ( P n , k S n ) + ⟩ . P_{n,k}=\operatorname{Sym}(V\otimes K^n)=K[x_i^{(a)}:i\in[n],\ a\in[k]],\qquad R_{n,k}=P_{n,k}/\langle(P_{n,k}^{S_n})_+\rangle. P n , k = Sym ( V ⊗ K n ) = K [ x i ( a ) : i ∈ [ n ] , a ∈ [ k ]] , R n , k = P n , k / ⟨( P n , k S n ) + ⟩ .
The row action of GL ( V ) \operatorname{GL}(V) GL ( V ) commutes with the label action of S n S_n S n . We use λ \lambda λ for row partitions and μ \mu μ for label partitions, with Schur modules S λ V S_\lambda V S λ V and Specht modules S μ S^\mu S μ :
R n , k ≅ ⨁ μ ⊢ n ⨁ ℓ ( λ ) ≤ k c λ , μ ( n ) S λ V ⊗ S μ . (1) R_{n,k}\cong\bigoplus_{\mu\vdash n}\ \bigoplus_{\ell(\lambda)\le k} c_{\lambda,\mu}^{(n)}S_\lambda V\otimes S^\mu. \tag*{(1)} R n , k ≅ μ ⊢ n ⨁ ℓ ( λ ) ≤ k ⨁ c λ , μ ( n ) S λ V ⊗ S μ . (1)
Let H n , k \mathcal{H}_{n,k} H n , k be the apolar harmonic model of R n , k \mathcal{R}_{n,k} R n , k , and let P n , k \mathcal{P}_{n,k} P n , k be generated from the Vandermonde determinant Δ n ( x ( 1 ) ) \Delta_n(x^{(1)}) Δ n ( x ( 1 ) ) by all partial derivatives and generalized polarizations
E a , b ( r ) = ∑ i = 1 n x i ( a ) ∂ r ∂ x i ( b ) , r ≥ 1. E^{(r)}_{a,b}=\sum_{i=1}^{n}x_i^{(a)}\frac{\partial^r}{\partial x_i^{(b)}},\qquad r\geq1. E a , b ( r ) = i = 1 ∑ n x i ( a ) ∂ x i ( b ) ∂ r , r ≥ 1.
This is Bergeron’s module E n ⟨ k ⟩ E_n^{\langle k\rangle} E n ⟨ k ⟩ [4 Section 2] , in the general polarization framework of [7 Section 3.1] . Haiman’s operator theorem gives P n , 2 = H n , 2 \mathcal{P}_{n,2}=\mathcal{H}_{n,2} P n , 2 = H n , 2 [19 Theorem 4.2] . For M = H M=\mathcal{H} M = H or P \mathcal{P} P , let C μ M C^M_\mu C μ M be the stable row character of its label-μ \mu μ multiplicity and put A n M = C ( 1 n ) M A^M_n=C^M_{(1^n)} A n M = C ( 1 n ) M . Here stability refers to adjoining alphabets, with n n n fixed, and e j ⊥ e_j^\perp e j ⊥ is adjoint to multiplication by e j e_j e j for the Hall inner product.
Theorem A (Bergeron’s Hook-Components and Coefficient-Length conjectures). For every n ≥ 1 n\geq1 n ≥ 1 and M = H , P M=\mathcal{H},\mathcal{P} M = H , P ,
C ( j + 1 , 1 n − j − 1 ) M = e j ⊥ A n M ( 0 ≤ j < n ) , C^M_{(j+1,1^{n-j-1})}=e_j^\perp A^M_n \qquad(0\leq j<n), C ( j + 1 , 1 n − j − 1 ) M = e j ⊥ A n M ( 0 ≤ j < n ) , max { ℓ ( λ ) : [ s λ ] C μ M ≠ 0 } = n − μ 1 ( μ ⊢ n ) . \max\{\ell(\lambda):[s_\lambda]C^M_\mu\ne0\}=n-\mu_1 \qquad(\mu\vdash n). max { ℓ ( λ ) : [ s λ ] C μ M = 0 } = n − μ 1 ( μ ⊢ n ) . In k ≥ 1 k\geq1 k ≥ 1 alphabets the largest length is min { k , n − μ 1 } \min\{k,n-\mu_1\} min { k , n − μ 1 } . The empty row partition has length zero.
The length bound is attained by generated row-hook multiplicity spaces (Theorem 4.6 ). The hook identities are realized by exteriorization and derivative maps (Theorems 6.8 and 6.10 ).
The same calculation proves the nontrivial hook-label cases of Bergeron’s Conjecture 4, with the hook–hook formula (4.8). It uses the Fields character theorem [27 ] and the superharmonic operator theorem [29 ] .
The identities also make Bergeron’s finite-alphabet reconstruction unconditional: for n ≥ 3 n\geq3 n ≥ 3 , the hook-label characters in ⌊ ( n − 1 ) / 2 ⌋ \lfloor(n-1)/2\rfloor ⌊( n − 1 ) /2 ⌋ alphabets determine A n M A^M_n A n M (Theorem 6.17 ). The support bound λ 1 ′ + λ 2 ′ ≤ n − 1 \lambda'_1+\lambda'_2\leq n-1 λ 1 ′ + λ 2 ′ ≤ n − 1 gives alternating and hook-label generation for n ≤ 6 n\leq6 n ≤ 6 (Corollary 6.16).
For one bosonic and two fermionic alphabets, the transfer proves the fully graded sign component of the t = 0 t=0 t = 0 Theta conjecture [14 Conjecture 8.2] (Theorem 7.2). At q = 1 q=1 q = 1 its character is ( 1 + u + v ) n − 1 (1+u+v)^{n-1} ( 1 + u + v ) n − 1 and its dimension is 3 n − 1 3^{n-1} 3 n − 1 . The known prediction [20 Theorem 5.6] , recorded in [24 Theorem 8.3] , is realized by the hook identity and diagonal supersymmetry [25 ] .
For a nontrivial label μ = ( c , τ ) \mu=(c,\tau) μ = ( c , τ ) , the maximal-length row-hook multiplicities form S τ K c S_\tau K^c S τ K c , with principal grading shifted by one. Explicit harmonic operators realize its auxiliary G L c \mathrm{GL}_c GL c action and Lefschetz maps (Theorem 8.2 ).
The character questions of Church–Ellenberg–Farb
Problem 1.10 of [11 ] asks for the multigraded symmetric-group characters in every number of alphabets, at least for sufficiently large n n n . Problem 1.12 asks for explicit dimension polynomials whose existence follows from FI-module theory. Write X j ( σ ) X_j(\sigma) X j ( σ ) for the number of j j j -cycles of σ ∈ S n \sigma\in S_n σ ∈ S n .
Theorem B (Full characters and dimension polynomials). Let b = ( b 1 , … , b k ) ∈ Z ≥ 0 k \mathbf{b}=(b_1,\ldots,b_k)\in\mathbb{Z}_{\geq0}^k b = ( b 1 , … , b k ) ∈ Z ≥ 0 k have total degree D ≥ 3 D\geq3 D ≥ 3 . If n ≥ max { D , 2 D − 6 } n\geq\max\{D,2D-6\} n ≥ max { D , 2 D − 6 } , then
tr ( σ ∣ ( R n , k ) b ) = [ q b ] ∏ α ∈ Z ≥ 0 k 0 < ∣ α ∣ ≤ D ( 1 − q α ) ∏ j = 1 D ∏ a = 1 k ( 1 − q a j ) − X j ( σ ) . (2) \operatorname{tr}(\sigma\mid(\mathcal{R}_{n,k})_{\mathbf{b}})=[\mathbf{q}^{\mathbf{b}}]\prod_{\substack{\alpha\in\mathbb{Z}_{\geq0}^k\\0<|\alpha|\leq D}}\left(1-\mathbf{q}^{\alpha}\right)^{\prod_{j=1}^{D}\prod_{a=1}^{k}(1-q_a^j)^{-X_j(\sigma)}}. \tag*{(2)} tr ( σ ∣ ( R n , k ) b ) = [ q b ] α ∈ Z ≥ 0 k 0 < ∣ α ∣ ≤ D ∏ ( 1 − q α ) ∏ j = 1 D ∏ a = 1 k ( 1 − q a j ) − X j ( σ ) . (2) The moment Koszul complex is exact in positive homological degrees in this range. The formula also holds for D ≤ 1 D\leq1 D ≤ 1 , n ≥ 1 n\geq1 n ≥ 1 , and for D = 2 D=2 D = 2 , n ≥ 2 n\geq2 n ≥ 2 .
At the identity this gives the dimension polynomial in every multidegree, with a range uniform in the number of alphabets. Church–Ellenberg–Farb’s theory establishes eventual polynomiality and representation stability; subsequent quantitative bounds include [2 , 8 ] . This establishes Bergeron’s proposed low-degree Euler product [6 Section 4, equation (33)] in the displayed range.
The quadratic model also computes the first-occurrence degree and the next two degrees of every label in every rank, and the entire two-column row sector (Corollary 10.2). These degrees may lie beyond the uniform stable range.
Orthogonal Littlewood theory [31 ] and the Schur relation confine positive homology to boundary rows. Exactness in the first possible degree removes an entire boundary before any case calculation. Wang’s dimension polynomial [36 Theorem 5.1] and a free cover give bivariate exactness; the Fields character treats row hooks. At n = 2 D − 6 n = 2D - 6 n = 2 D − 6 , one explicit nonzero cubic image finishes the proof, independently of operator generation. Supersymmetry transfers the formula to every bosonic–fermionic superspace [25 ] .
In every rank, two-column moment homology is acyclic at degree n n n and reflects degrees D > n D > n D > n to 2 n − D 2n - D 2 n − D by an explicit cycle map (Theorem 10.13 ). Under this reflection, multiplication becomes symplectic contraction. Thus all two-column first syzygies are generated from their degree-( n + 1 ) (n + 1) ( n + 1 ) layer (Corollary 10.16 ). Invariants and harmonics separate through the character range, and through degree n n n on row hooks and partitions of length at most two (Corollary 10.11).
Natural stable multiplicity spaces
The underlying presentation applies to every augmented commutative algebra B = K ⊕ m B = K \oplus\mathfrak{m} B = K ⊕ m in characteristic zero. The following stable consequence requires only a connected positive grading. Define
Q n ( B ) = B ⊗ n / ⟨ ( B ⊗ n ) + S n ⟩ . Q_n(B) = B^{\otimes n}\big/\left\langle(B^{\otimes n})_+^{S_n}\right\rangle. Q n ( B ) = B ⊗ n / ⟨ ( B ⊗ n ) + S n ⟩ .
For a partition τ ⊢ s \tau\vdash s τ ⊢ s , write τ [ n ] = ( n − s , τ ) \tau[n] = (n - s,\tau) τ [ n ] = ( n − s , τ ) when this is a partition, and set τ 1 = 0 \tau_1 = 0 τ 1 = 0 for the empty partition.
Theorem C. (Natural stable multiplicities). For D ≥ s D \ge s D ≥ s and n ≥ D + τ 1 n \ge D + \tau_1 n ≥ D + τ 1 , there is an isomorphism
Hom S n ( S τ [ n ] , Q n ( B ) D ) ≅ ( S τ m ) D . \operatorname{Hom}_{S_n}\left(S^{\tau[n]}, Q_n(B)_D\right) \cong(S_\tau\mathfrak{m})_D. Hom S n ( S τ [ n ] , Q n ( B ) D ) ≅ ( S τ m ) D . After fixing the one-dimensional Pieri branching line, the isomorphism is natural for graded augmented algebra maps and auxiliary symmetries preserving the grading and augmentation.
Multiplication enters these multiplicity spaces through the canonical relation map β μ , B \beta_{\mu,B} β μ , B , whose source for μ = ( t , τ ) \mu= (t,\tau) μ = ( t , τ ) is S ( t + 1 , τ 2 , … ) m S_{(t+1,\tau_2,\ldots)}\mathfrak{m} S ( t + 1 , τ 2 , … ) m . In a connected positive grading, its first possible contribution has degree n + 1 − τ 1 n + 1 - \tau_1 n + 1 − τ 1 . The map commutes with arbitrary augmented algebra maps and vanishes when m 2 = 0 \mathfrak{m}^2 = 0 m 2 = 0 (Theorem 9.3). For B = Sym V B = \operatorname{Sym} V B = Sym V , the multiplicity character is [ s τ [ ∑ j ≥ 1 h j ( V ) ] ] D [s_\tau[\sum_{j \ge1} h_j(V)]]_D [ s τ [ ∑ j ≥ 1 h j ( V )] ] D .
For nonempty τ \tau τ , the first correction is exact in every degree D > s D > s D > s . At n = D + τ 1 − 1 n = D + \tau_1 - 1 n = D + τ 1 − 1 , the kernel of the canonical Schur cover is
S ( D − s + τ 1 , τ 2 , … ) V . S_{(D-s+\tau_1,\tau_2,\ldots)}V. S ( D − s + τ 1 , τ 2 , … ) V .
It is nonzero whenever dim V ≥ ℓ ( τ ) \dim V \ge\ell(\tau) dim V ≥ ℓ ( τ ) , proving sharpness of the stable threshold in every such degree (Theorem 9.8 ). The proof specializes to orthogonal idempotents: their tensor moment complex contracts on every nonzero color. Naturality then detects the polynomial relation.
The relation extends to a natural filtered resolution with one Schur generator in each homological degree (Theorem 9.5 , proved in Appendix § 17 ). The idempotent argument gives injectivity at every first possible polynomial degree except the invariant first relation (Theorem 9.7 ).
The first parabolic algebra retains multiplication:
Q n ( B ) S n − 1 ≅ B / m n . Q_n(B)^{S_{n-1}} \cong B/\mathfrak{m}^n. Q n ( B ) S n − 1 ≅ B / m n .
Equivariant maps recover the augmented jet maps (Theorem 9.12 ), and compatible towers recover the formal completion. Thus these algebras retain the multiplication that stable multiplicities forget.
Constructive Vandermonde generation
Selected-label evaluation and disjoint products construct harmonic lifts. Write M n , k [ D ] μ \mathcal{M}_{n,k}[D]_{\mu} M n , k [ D ] μ for the label-μ \mu μ isotype in total degree D D D .
Theorem D (Uniform multivariate Vandermonde generation). In every number of alphabets, either condition
μ 1 ≥ D − 1 or D ≤ μ 1 + μ 3 + μ 5 + ⋯ \mu_1 \ge D-1 \qquad\text{or} \qquad D \le\mu_1+\mu_3+\mu_5+\cdots μ 1 ≥ D − 1 or D ≤ μ 1 + μ 3 + μ 5 + ⋯ implies P n , k [ D ] μ = H n , k [ D ] μ \mathcal{P}_{n,k}[D]_{\mu}=\mathcal{H}_{n,k}[D]_{\mu} P n , k [ D ] μ = H n , k [ D ] μ . For D ≥ 3 D \ge3 D ≥ 3 ,
n ≥ 2 D − 3 ⟹ P n , k [ d ] = H n , k [ d ] ( 0 ≤ d ≤ D ) . (3) n \ge2D-3 \quad\Longrightarrow\quad\mathcal{P}_{n,k}[d]=\mathcal{H}_{n,k}[d] \qquad(0 \le d \le D). \tag*{(3)} n ≥ 2 D − 3 ⟹ P n , k [ d ] = H n , k [ d ] ( 0 ≤ d ≤ D ) . (3) For every n , k n,k n , k , every label, and every strictly positive integral row-weight vector, the full label isotype at its first weighted occurrence is generated. The first occurrence means the least weighted degree in which the label appears.
Thus every degree through ⌊ ( n + 3 ) / 2 ⌋ \lfloor(n+3)/2 \rfloor ⌊( n + 3 ) /2 ⌋ is generated, uniformly in the number of alphabets. A colored determinant map turns the curvilinear Quot degeneration of Bérczi–Svendsen [3 ] into disjoint-label multiplication and every positive-weight first occurrence. Pair kernels save auxiliary labels, column alternation gives the odd-row criterion, and the hook identities leave one family, ( D − 2 , D − 2 , 1 ) (D-2,D-2,1) ( D − 2 , D − 2 , 1 ) , settled by an explicit rank-five lift.
Two-column Vandermonde generation reduces to the single degree n − 1 n-1 n − 1 . For n ≥ 6 n \ge6 n ≥ 6 , it suffices to test the remaining rows in n − 3 n-3 n − 3 alphabets; the cases n ≤ 5 n \le5 n ≤ 5 are already generated (Proposition 10.17 ).
Reading routes
Selected elimination, harmonic forms and the commuting model lead to the conjecture proofs in § 6 . The Theta application follows immediately; the auxiliary action and tensor-coinvariant results form the next independent branches. The canonical Schur presentation and its stable and first-unstable consequences occupy § 9 . The character proof in § 10 combines them with quadratic, bivariate and row-hook exactness, independently of operator generation. The finer rowwise trace arguments are kept in Appendix D. Disjoint products then feed the local lifts of § 14 .
Selected-label elimination
Let K K K be a field, let n ! n! n ! be invertible in K K K , and let ε : B → K \varepsilon: B \to K ε : B → K be an augmented commutative K K K -algebra, with m = ker ε \mathfrak{m} = \ker\varepsilon m = ker ε . Put
Q n ( B ) = B ⊗ n / ⟨ ( B ⊗ n ) S n ∩ ker ( ε ⊗ n ) ⟩ . Q_n(B) = B^{\otimes n}/\langle(B^{\otimes n})^{S_n} \cap\ker(\varepsilon^{\otimes n})\rangle. Q n ( B ) = B ⊗ n / ⟨( B ⊗ n ) S n ∩ ker ( ε ⊗ n )⟩ .
For b ∈ B b \in B b ∈ B , write b ( i ) b^{(i)} b ( i ) for its image in the i i i th tensor factor. The notation S n − r S_{n-r} S n − r below means the subgroup fixing the first r r r tensor factors. We call these r r r factors selected and the remaining factors auxiliary. When B B B is the coordinate algebra of a pointed affine scheme, Q n ( B ) Q_n(B) Q n ( B ) is its diagonal coinvariant algebra at the basepoint.
Traces on the auxiliary factors can be eliminated using the global invariant traces. The relations left after elimination are complete symmetric functions on the selected factors. The proof starts from the fundamental trace presentation [16 Theorem 1] ; its orbit-sum form makes both the resulting relations and the characteristic hypothesis transparent.
Theorem 2.1 (Exact selected-label presentation). For 1 ≤ r ≤ n 1 \le r \le n 1 ≤ r ≤ n , set m = n − r + 1 m = n-r+1 m = n − r + 1 . Inclusion of the first r r r tensor factors induces a natural algebra isomorphism
Q n ( B ) S n − r ≅ B ⊗ r ⟨ h m ( b ( 1 ) , … , b ( r ) ) : b ∈ m ⟩ . (4) Q_n(B)^{S_{n-r}} \cong\frac{B^{\otimes r}}{\langle h_m(b^{(1)},\ldots,b^{(r)}) : b \in\mathfrak{m}\rangle}. \tag*{(4)} Q n ( B ) S n − r ≅ ⟨ h m ( b ( 1 ) , … , b ( r ) ) : b ∈ m ⟩ B ⊗ r . (4) Equivalently, one may use every multilinear polarization of the displayed relation, allowing arbitrary inputs from m \mathfrak{m} m . The isomorphism is equivariant for S r S_r S r and for every group acting on B B B by augmentation-preserving algebra automorphisms.
Proof. Let 0 ≤ k ≤ n 0 \le k \le n 0 ≤ k ≤ n and put T S k ( B ) = ( B ⊗ k ) S k T S^k(B) = (B^{\otimes k})^{S_k} T S k ( B ) = ( B ⊗ k ) S k . In the polynomial algebra Sym K ( m ) \operatorname{Sym}_K(\mathfrak{m}) Sym K ( m ) write T b T_b T b for the generator depending K K K -linearly on b ∈ m b \in\mathfrak{m} b ∈ m . For j ≥ 1 j \ge1 j ≥ 1 define
Ψ j ( b 1 , … , b j ) = ∑ π ∈ Π j ( − 1 ) j − ∣ π ∣ ∏ C ∈ π ( ∣ C ∣ − 1 ) ! ∏ C ∈ π T ∏ i ∈ C b i . \Psi_j(b_1,\ldots,b_j) = \sum_{\pi\in\Pi_j} (-1)^{j-|\pi|} \prod_{C \in\pi} (|C|-1)! \prod_{C \in\pi} T_{\prod_{i \in C} b_i}. Ψ j ( b 1 , … , b j ) = π ∈ Π j ∑ ( − 1 ) j − ∣ π ∣ C ∈ π ∏ ( ∣ C ∣ − 1 )! C ∈ π ∏ T ∏ i ∈ C b i .
where Π j \Pi_j Π j is the set of set partitions of { 1 , … , j } \{1,\ldots,j\} { 1 , … , j } . Under T b ↦ ∑ i = 1 k b ( i ) T_b \mapsto\sum_{i=1}^{k} b^{(i)} T b ↦ ∑ i = 1 k b ( i ) , inclusion–exclusion on the partition lattice identifies Ψ j \Psi_j Ψ j with
∑ i 1 , … , i j ∈ { 1 , … , k } pairwise distinct b 1 ( i 1 ) ⋯ b j ( i j ) . \sum_{\substack{i_1,\ldots,i_j \in\{1,\ldots,k\}\\ \text{pairwise distinct}}} b_1^{(i_1)} \cdots b_j^{(i_j)}. i 1 , … , i j ∈ { 1 , … , k } pairwise distinct ∑ b 1 ( i 1 ) ⋯ b j ( i j ) .
In particular Ψ k + 1 \Psi_{k+1} Ψ k + 1 maps to zero. Multiplication by one further trace gives the recursion
Ψ j + 1 ( b 1 , … , b j , c ) = T c Ψ j ( b 1 , … , b j ) − ∑ a = 1 j Ψ j ( b 1 , … , b a − 1 , b a c , b a + 1 , … , b j ) . (5) \Psi_{j+1}(b_1,\ldots,b_j,c) = T_c\Psi_j(b_1,\ldots,b_j) - \sum_{a=1}^{j} \Psi_j(b_1,\ldots,b_{a-1},b_ac,b_{a+1},\ldots,b_j). \tag*{(5)} Ψ j + 1 ( b 1 , … , b j , c ) = T c Ψ j ( b 1 , … , b j ) − a = 1 ∑ j Ψ j ( b 1 , … , b a − 1 , b a c , b a + 1 , … , b j ) . (5)
which shows that all Ψ j \Psi_j Ψ j with j > k j > k j > k follow from Ψ k + 1 \Psi_{k+1} Ψ k + 1 and that the quotient by these relations is spanned by Ψ j \Psi_j Ψ j , 0 ≤ j ≤ k 0 \le j \le k 0 ≤ j ≤ k (with Ψ 0 = 1 \Psi_0 = 1 Ψ 0 = 1 ).
Choose a K K K -basis of m \mathfrak{m} m and append 1 1 1 to obtain a basis of B B B . The images of these Ψ j \Psi_j Ψ j , indexed by multisets of j j j basis elements, are nonzero scalar multiples of the orbit basis of T S k ( B ) T S^k(B) T S k ( B ) . Their scalar factors are products of factorials at most k ! k! k ! , and hence are invertible. They are therefore linearly independent. This proves
T S k ( B ) ≅ Sym K ( m ) / ⟨ Ψ k + 1 ( b 1 , … , b k + 1 ) ⟩ . T S^k(B) \cong\operatorname{Sym}_K(\mathfrak{m})/\langle\Psi_{k+1}(b_1,\ldots,b_{k+1})\rangle. T S k ( B ) ≅ Sym K ( m ) / ⟨ Ψ k + 1 ( b 1 , … , b k + 1 )⟩ .
In particular, symmetric tensors are generated by the traces of elements of m \mathfrak{m} m . Now take k = n − r k = n-r k = n − r . Before imposing the full S n S_n S n -invariants, the S k S_k S k -invariant algebra is B ⊗ r ⊗ T S k ( B ) B^{\otimes r} \otimes T S^k(B) B ⊗ r ⊗ T S k ( B ) . Reynolds averaging shows that the intersection of the full invariant-generated ideal with this algebra is generated there by the same invariant traces. Thus the quotient imposes
T b = − p r ( b ) , p r ( b ) = ∑ i = 1 r b ( i ) ( b ∈ m ) . T_b = -p_r(b), \qquad p_r(b) = \sum_{i=1}^{r} b^{(i)} \qquad(b \in\mathfrak{m}). T b = − p r ( b ) , p r ( b ) = i = 1 ∑ r b ( i ) ( b ∈ m ) .
Under this substitution every set-partition term of Ψ m \Psi_m Ψ m has sign ( − 1 ) m (-1)^m ( − 1 ) m . Removing that common sign leaves
∑ π ∈ Π m ∏ C ∈ π ( ∣ C ∣ − 1 ) ! p r ( ∏ i ∈ C b i ) . \sum_{\pi\in\Pi_m}\prod_{C\in\pi} (|C|-1)! \, p_r\left(\prod_{i\in C} b_i\right). π ∈ Π m ∑ C ∈ π ∏ ( ∣ C ∣ − 1 )! p r ( i ∈ C ∏ b i ) .
The complete-symmetric generating identity identifies this with the coefficient of t 1 ⋯ t m t_1\cdots t_m t 1 ⋯ t m in
h m ( ∑ a = 1 m t a b a ( 1 ) , … , ∑ a = 1 m t a b a ( r ) ) . h_m\left(\sum_{a=1}^{m}t_a b_a^{(1)},\ldots,\sum_{a=1}^{m}t_a b_a^{(r)}\right). h m ( a = 1 ∑ m t a b a ( 1 ) , … , a = 1 ∑ m t a b a ( r ) ) .
Polarization is valid because m ! m! m ! is invertible. Elimination of all the T b T_b T b therefore gives exactly (2.1). Every construction commutes with augmented algebra maps and with the stated actions.
Harmonic models and row characters
For the remainder of the paper, except where a more general base is specified, K K K is a field of characteristic zero. We retain the polynomial row convention and the notation P n , k P_{n,k} P n , k and R n , k R_{n,k} R n , k from the introduction.
Harmonics and the operator module
For α = ( α 1 , … , α k ) ∈ N k \alpha=(\alpha_1,\ldots,\alpha_k)\in\mathbb{N}^k α = ( α 1 , … , α k ) ∈ N k , define
D α = ∑ i = 1 n ∏ a = 1 k ∂ x i a α a . D_\alpha=\sum_{i=1}^{n}\prod_{a=1}^{k}\partial_{x_i^{a}}^{\alpha_a}. D α = i = 1 ∑ n a = 1 ∏ k ∂ x i a α a .
The full diagonal harmonic space is
H n , k = ⋂ α ≠ 0 ker D α . (6) \mathcal{H}_{n,k}=\bigcap_{\alpha\ne0}\ker D_\alpha. \tag*{(6)} H n , k = α = 0 ⋂ ker D α . (6)
Polarized power sums generate P n , k S n P_{n,k}^{S_n} P n , k S n in characteristic zero. Their apolar equations identify (3.1) with the orthogonal complement of the Hilbert ideal for the Fischer pairing
⟨ f , g ⟩ = f ( ∂ ) g ∣ X = 0 . \langle f,g\rangle=f(\partial)g\big|_{X=0}. ⟨ f , g ⟩ = f ( ∂ ) g X = 0 .
Over the reals this pairing is positive definite on each homogeneous component. The quotient projection therefore restricts to an isomorphism
H n , k → ∼ R n , k . (7) \mathcal{H}_{n,k}\xrightarrow{\sim}R_{n,k}. \tag*{(7)} H n , k ∼ R n , k . (7)
The spaces and maps are defined over Q \mathbb{Q} Q , so the assertion holds over K K K by extension of scalars. The differential equations are stable under row changes; consequently the isomorphism is G L k × S n \mathrm{GL}_k\times S_n GL k × S n -equivariant with our polynomial row convention.
The ordinary Vandermonde is harmonic. Partial derivatives preserve (3.1), and
[ D α , E a , b ( s ) ] = α a D α − e a + s e b , (8) [D_\alpha,E_{a,b}^{(s)}]=\alpha_aD_{\alpha-e_a+se_b}, \tag*{(8)} [ D α , E a , b ( s ) ] = α a D α − e a + s e b , (8)
where a zero coefficient gives zero. This proves P n , k ⊆ H n , k P_{n,k}\subseteq\mathcal{H}_{n,k} P n , k ⊆ H n , k . The operators with s = 1 s=1 s = 1 give the row Lie algebra. As every homogeneous part is a polynomial representation, P n , k P_{n,k} P n , k is a G L k × S n \mathrm{GL}_k\times S_n GL k × S n -submodule. It is graded: the initial Vandermonde is homogeneous and all generating operators are homogeneous for total degree.
The linear harmonic equations say that every harmonic is translation-invariant in each row. Write
W n = { ( z 1 , … , z n ) ∈ K n : ∑ i z i = 0 } . W_n=\{(z_1,\ldots,z_n)\in K^n:\sum_i z_i=0\}. W n = {( z 1 , … , z n ) ∈ K n : i ∑ z i = 0 } .
Using the invariant dot product to identify W n W_n W n with its dual, we can regard H n , k \mathcal{H}_{n,k} H n , k and P n , k P_{n,k} P n , k as derivative-closed submodules of Sym ( K k ⊗ W n ) \operatorname{Sym}(K^k\otimes W_n) Sym ( K k ⊗ W n ) . In particular every occurring row partition has length at most n − 1 n-1 n − 1 by the Cauchy decomposition.
Stable row coefficients.
Lemma 3.1 (Trailing-zero stability). For ℓ ( λ ) ≤ k \ell(\lambda) \le k ℓ ( λ ) ≤ k , the highest-weight multiplicity space of row weight λ \lambda λ in H n , k \mathcal{H}_{n,k} H n , k is naturally the highest-weight multiplicity space of ( λ , 0 ) (\lambda,0) ( λ , 0 ) in H n , k + 1 \mathcal{H}_{n,k+1} H n , k + 1 .
Proof. A polynomial with last row degree zero does not use the last row. Every differential equation involving that row vanishes on it. The other equations are exactly the equations in k k k rows. The additional row raising operators also differentiate in the last row and hence vanish. The remaining highest-weight conditions are unchanged. □ \square □
Thus adjoining an unused row preserves every full harmonic multiplicity in (1) . The stable character is finite: its row lengths are bounded by n − 1 n-1 n − 1 , and R n , n − 1 R_{n,n-1} R n , n − 1 is finite-dimensional: the invariant ideal contains the ordinary invariant ideal in each row, so the quotient is a quotient of a tensor product of ordinary coinvariant rings. We write its label coefficients as symmetric functions
C μ ( n ) = C μ H = ∑ λ c λ , μ ( n ) s λ , A n = A n H = C ( 1 n ) ( n ) . C^{(n)}_{\mu}=C^{\mathcal{H}}_{\mu}=\sum_{\lambda}c^{(n)}_{\lambda,\mu}s_{\lambda},\qquad A_n=A^{\mathcal{H}}_n=C^{(n)}_{(1^n)}. C μ ( n ) = C μ H = λ ∑ c λ , μ ( n ) s λ , A n = A n H = C ( 1 n ) ( n ) .
Evaluation in k k k variables recovers the character in k k k row alphabets. For symmetric functions, f ⊥ f^{\perp} f ⊥ denotes the adjoint of multiplication by f f f for the Hall inner product, in which the Schur functions are orthonormal.
For P \mathcal{P} P , we use the direct system obtained by adjoining unused rows. For a fixed row partition its highest-weight spaces form an increasing sequence inside the fixed finite-dimensional full-harmonic space of Lemma 3.1; the sequence therefore stabilizes. This defines the stable characters
C μ P = ∑ λ c λ , μ P s λ , A n P = C ( 1 n ) P . C^{\mathcal{P}}_{\mu}=\sum_{\lambda}c^{\mathcal{P}}_{\lambda,\mu}s_{\lambda},\qquad A^{\mathcal{P}}_n=C^{\mathcal{P}}_{(1^n)}. C μ P = λ ∑ c λ , μ P s λ , A n P = C ( 1 n ) P .
The commuting normal form in Theorem 5.2 will identify these stable characters with their actual specialization in every finite number of rows.
A parabolic invariant quotient.
For a set of labels T ⊆ [ n ] T\subseteq[n] T ⊆ [ n ] , write S T S_T S T for the permutations supported on T T T . Set X i = ( x i ( 1 ) , … , x i ( k ) ) X_i=(x_i^{(1)},\ldots,x_i^{(k)}) X i = ( x i ( 1 ) , … , x i ( k ) ) . Passing to tail invariants leaves only the selected labels: every tail power sum becomes the negative of the selected power sum. This elementary elimination controls row length throughout the paper.
Lemma 3.2. For 0 ≤ r ≤ n 0\le r\le n 0 ≤ r ≤ n , the natural map
K [ X 1 , … , X r ] ⟶ R n , k S { r + 1 , … , n } K[X_1,\ldots,X_r]\longrightarrow R_{n,k}^{S_{\{r+1,\ldots,n\}}} K [ X 1 , … , X r ] ⟶ R n , k S { r + 1 , … , n } is surjective. Consequently every row partition occurring in the target has length at most r r r .
Proof. For r ≥ 1 r\ge1 r ≥ 1 , surjectivity is the polynomial case of Theorem 2.1; for r = 0 r=0 r = 0 it is R n , k S n = K R_{n,k}^{S_n}=K R n , k S n = K . Finally,
K [ X 1 , … , X r ] = Sym ( K k ⊗ K r ) ≅ ⨁ λ S λ K k ⊗ S λ K r , K[X_1,\ldots,X_r]=\operatorname{Sym}(K^k\otimes K^r)\cong\bigoplus_{\lambda}S_{\lambda}K^k\otimes S_{\lambda}K^r, K [ X 1 , … , X r ] = Sym ( K k ⊗ K r ) ≅ λ ⨁ S λ K k ⊗ S λ K r , and S λ K r = 0 S_{\lambda}K^r=0 S λ K r = 0 for ℓ ( λ ) > r \ell(\lambda)>r ℓ ( λ ) > r . □ \square □
Proposition 3.3 (The length upper bound). If c λ , μ ( n ) ≠ 0 c^{(n)}_{\lambda,\mu}\ne0 c λ , μ ( n ) = 0 , then
ℓ ( λ ) ≤ n − μ 1 . \ell(\lambda)\le n-\mu_1. ℓ ( λ ) ≤ n − μ 1 . The same bound holds in every P n , k \mathcal{P}_{n,k} P n , k .
Proof. The restriction of S μ S^{\mu} S μ to a subgroup S μ 1 S_{\mu_1} S μ 1 contains the trivial representation. By the branching rule, one may remove every box below the first row, in a sequence of removable corners, and reach the one-row partition ( μ 1 ) (\mu_1) ( μ 1 ) . Thus an occurrence of S λ K k ⊗ S μ S_{\lambda}K^k\otimes S^{\mu} S λ K k ⊗ S μ gives an occurrence of S λ K k S_{\lambda}K^k S λ K k in R n , k S μ 1 R_{n,k}^{S_{\mu_1}} R n , k S μ 1 . Apply Lemma 3.2 with r = n − μ 1 r=n-\mu_1 r = n − μ 1 . The assertion for P \mathcal{P} P follows from its inclusion in H ≅ R \mathcal{H}\cong R H ≅ R . □ \square □
Harmonic forms and the Coefficient-Length conjecture
The Euler splitting of harmonic forms
A row hook has an alternating block of one-box rows below its first row. Replacing that block by exterior variables turns its highest-weight condition into contraction by the Euler vector field. The de Rham identity then separates precisely the primitive forms corresponding to row hooks.
On K [ x 1 , … , x n ] ⊗ Λ ( θ 1 , … , θ n ) K[x_1,\ldots,x_n]\otimes\Lambda(\theta_1,\ldots,\theta_n) K [ x 1 , … , x n ] ⊗ Λ ( θ 1 , … , θ n ) , put
D r = ∑ i ∂ x i r , κ r = ∑ i ∂ x i r ∂ θ i , ι = ∑ i x i ∂ θ i , d d R = ∑ i θ i ∂ x i . D_r=\sum_i\partial_{x_i}^{r},\qquad\kappa_r=\sum_i\partial_{x_i}^{r}\partial_{\theta_i},\qquad\iota=\sum_i x_i\partial_{\theta_i},\qquad d_{\mathrm{dR}}=\sum_i\theta_i\partial_{x_i}. D r = i ∑ ∂ x i r , κ r = i ∑ ∂ x i r ∂ θ i , ι = i ∑ x i ∂ θ i , d dR = i ∑ θ i ∂ x i .
The superharmonic space S H n SH_n S H n is the joint kernel of D r D_r D r for r ≥ 1 r\ge1 r ≥ 1 and κ r \kappa_r κ r for r ≥ 0 r\ge0 r ≥ 0 . These are the apolar equations for the coinvariant algebra of differential forms. The invariant differential-form description goes back to Solomon [33 ] . Write S H n a , j SH_n^{a,j} S H n a , j for polynomial degree a a a and exterior degree j j j .
For a ≥ 1 a\ge1 a ≥ 1 , let K a , j K_{a,j} K a , j be the S n S_n S n -module of row highest-weight vectors of weight ( a , 1 j ) (a,1^j) ( a , 1 j ) in H n , j + 1 \mathcal{H}_{n,j+1} H n , j + 1 . For the boundary terms put K 0 , j = 0 K_{0,j}=0 K 0 , j = 0 for j > 0 j>0 j > 0 and K 0 , 0 = K K_{0,0}=K K 0 , 0 = K .
Lemma 4.1 (The row-hook bridge). There are S n S_n S n -equivariant identifications
K a , j = ker ι ∩ S H n a , j . (9) K_{a,j}=\ker\iota\cap SH_n^{a,j}. \tag*{(9)} K a , j = ker ι ∩ S H n a , j . (9) Away from bidegree ( 0 , 0 ) (0,0) ( 0 , 0 ) , there is a direct sum
S H n a , j = K a , j ⊕ d d R K a + 1 , j − 1 , (10) SH_n^{a,j}=K_{a,j}\oplus d_{\mathrm{dR}}K_{a+1,j-1}, \tag*{(10)} S H n a , j = K a , j ⊕ d dR K a + 1 , j − 1 , (10) with nonexistent exterior degrees interpreted as zero.
Proof. Replace an exterior monomial by a determinant in j j j auxiliary rows:
θ i 1 ⋯ θ i j ⟼ det ( y i s ( r ) ) 1 ≤ r , s ≤ j . (11) \theta_{i_1}\cdots\theta_{i_j}\longmapsto\det(y_{i_s}^{(r)})_{1\le r,s\le j}. \tag*{(11)} θ i 1 ⋯ θ i j ⟼ det ( y i s ( r ) ) 1 ≤ r , s ≤ j . (11)
This identifies exterior degree j j j with polynomials multilinear and alternating in the auxiliary rows. Raising operators between auxiliary rows vanish on these polynomials. Raising an auxiliary row into the x x x -row is precisely contraction by ι \iota ι . Conversely, the auxiliary highest-weight conditions in weight ( 1 , … , 1 ) (1,\ldots,1) ( 1 , … , 1 ) select the determinant representation, so every row-hook highest vector has this form.
The harmonic equations using only the x x x -row are D r D_r D r . Those using exactly one auxiliary row are κ r \kappa_r κ r . Any equation with two auxiliary derivatives at the same label vanishes on an alternating determinant, and an equation differentiating twice in an auxiliary row vanishes by multilinearity. These are all possible equations, proving (4.1).
Both ι \iota ι and d d R d_{\mathrm{dR}} d dR preserve S H n SH_n S H n , since
[ D r , d d R ] = 0 , [ D r , ι ] = r κ r − 1 , κ r d d R + d d R κ r = D r + 1 , κ r ι + ι κ r = 0. [D_r,d_{\mathrm{dR}}]=0,\qquad[D_r,\iota]=r\kappa_{r-1},\qquad\kappa_r d_{\mathrm{dR}}+d_{\mathrm{dR}}\kappa_r=D_{r+1},\qquad\kappa_r\iota+\iota\kappa_r=0. [ D r , d dR ] = 0 , [ D r , ι ] = r κ r − 1 , κ r d dR + d dR κ r = D r + 1 , κ r ι + ι κ r = 0.
Moreover ι 2 = d d R 2 = 0 \iota^2=d_{\mathrm{dR}}^2=0 ι 2 = d dR 2 = 0 and
ι d d R + d d R ι = ( a + j ) id on bidegree ( a , j ) . \iota d_{\mathrm{dR}}+d_{\mathrm{dR}}\iota=(a+j)\operatorname{id}\quad\text{on bidegree }(a,j). ι d dR + d dR ι = ( a + j ) id on bidegree ( a , j ) .
Thus every f ∈ S H n a , j f\in SH_n^{a,j} f ∈ S H n a , j with a + j > 0 a+j>0 a + j > 0 has the decomposition
f = ι d d R f a + j + d d R ( ι f a + j ) . f=\frac{\iota d_{\mathrm{dR}}f}{a+j}+d_{\mathrm{dR}}\left(\frac{\iota f}{a+j}\right). f = a + j ι d dR f + d dR ( a + j ι f ) .
The first term belongs to K a , j K_{a,j} K a , j , and the expression in parentheses belongs to K a + 1 , j − 1 K_{a+1,j-1} K a + 1 , j − 1 . On this latter primitive space, ι d d R = ( a + j ) id \iota d_{\mathrm{dR}}=(a+j)\operatorname{id} ι d dR = ( a + j ) id , so d d R d_{\mathrm{dR}} d dR is injective with inverse ι / ( a + j ) \iota/(a+j) ι / ( a + j ) on its image. The same identity shows that its image meets ker ι \ker\iota ker ι trivially. This proves (4.2). When a = 0 a=0 a = 0 and j > 0 j>0 j > 0 , d d R d_{\mathrm{dR}} d dR vanishes on the source; the same identity gives ker ι = 0 \ker\iota=0 ker ι = 0 , as required by the boundary convention.
A tableau formula and generation of all row hooks
For a standard tableau T T T , a descent is an integer i i i for which i + 1 i+1 i + 1 lies in a lower row than i i i . Let des ( T ) \operatorname{des}(T) des ( T ) be the number of descents and maj ( T ) \operatorname{maj}(T) maj ( T ) their sum.
Theorem 4.2 (Row-hook multiplicities). For every μ ⊢ n \mu\vdash n μ ⊢ n and j ≥ 0 j \ge0 j ≥ 0 ,
∑ a ≥ 1 [ K a , j : S μ ] q a = ∑ T ∈ SYT ( μ ) t = des ( T ) ≥ 1 q maj ( T ) e j ( q − 2 , q − 3 , … , q − t ) . (12) \sum_{a\geq1} [K_{a,j}:S^\mu]q^a
=
\sum_{\substack{T\in\operatorname{SYT}(\mu)\\ t=\operatorname{des}(T)\geq1}}
q^{\operatorname{maj}(T)}e_j(q^{-2},q^{-3},\ldots,q^{-t}). \tag*{(12)} a ≥ 1 ∑ [ K a , j : S μ ] q a = T ∈ SYT ( μ ) t = des ( T ) ≥ 1 ∑ q maj ( T ) e j ( q − 2 , q − 3 , … , q − t ) . (12) Equivalently, the coefficient indexed by ( a , 1 j ) (a,1^j) ( a , 1 j ) and μ \mu μ counts pairs ( T , J ) (T,J) ( T , J ) satisfying
J ⊆ { 2 , … , des ( T ) } , ∣ J ∣ = j , maj ( T ) − ∑ r ∈ J r = a . J\subseteq\{2,\ldots,\operatorname{des}(T)\},\qquad|J|=j,\qquad\operatorname{maj}(T)-\sum_{r\in J}r=a. J ⊆ { 2 , … , des ( T )} , ∣ J ∣ = j , maj ( T ) − r ∈ J ∑ r = a .
Proof. The Fields theorem of Murai, Rhoades and Wilson [27 Theorem 5.2 and equation (2.5)] gives
Frob ( S H n ; q , u ) = ∑ μ ⊢ n s μ [ Z ] ∑ T ∈ SYT ( μ ) q maj ( T ) ∏ r = 1 des ( T ) ( 1 + u q − r ) . (13) \operatorname{Frob}(SH_n;q,u)=\sum_{\mu\vdash n}s_\mu[Z]\sum_{T\in\operatorname{SYT}(\mu)}q^{\operatorname{maj}(T)}\prod_{r=1}^{\operatorname{des}(T)}(1+uq^{-r}). \tag*{(13)} Frob ( S H n ; q , u ) = μ ⊢ n ∑ s μ [ Z ] T ∈ SYT ( μ ) ∑ q maj ( T ) r = 1 ∏ des ( T ) ( 1 + u q − r ) . (13)
The generating series of (4.2) says that the nonconstant character is ( 1 + u / q ) (1+u/q) ( 1 + u / q ) times the generating series of the K a , j K_{a,j} K a , j . The unique tableau without descents is the one-row tableau, and contributes only the constant trivial representation. For every other tableau the factor 1 + u / q 1+u/q 1 + u / q cancels. Extracting u j u^j u j gives (4.4). All remaining powers of q q q are positive: if t = des ( T ) t=\operatorname{des}(T) t = des ( T ) , then maj ( T ) ≥ t ( t + 1 ) / 2 \operatorname{maj}(T)\geq t(t+1)/2 maj ( T ) ≥ t ( t + 1 ) /2 , whereas ∑ r = 2 t r = t ( t + 1 ) / 2 − 1 \sum_{r=2}^{t}r=t(t+1)/2-1 ∑ r = 2 t r = t ( t + 1 ) /2 − 1 .
The harmonic-form bridge extracts the row-hook multiplicities from the Fields character. For the alternating label, the resulting specialization also appears in [25 Corollary 4.6] . The row-hook coefficients after summing symmetric-group dimensions are studied in [12 ] .
Proposition 4.3 (Generation of the row-hook sector). Every row-hook isotypic component of H n , k \mathcal{H}_{n,k} H n , k is contained in P n , k \mathcal{P}_{n,k} P n , k .
Proof. For r ≥ 1 r\geq1 r ≥ 1 set Θ r = ∑ i ∂ i Θ x i \Theta_r=\sum_i\partial_i\Theta_{x_i} Θ r = ∑ i ∂ i Θ x i . The operator theorem of Rhoades and Wilson [29 Theorem 5.1] gives
S H n = ∑ J ⊆ [ n − 1 ] Θ J H n , 1 , (14) SH_n=\sum_{J\subseteq[n-1]}\Theta_J\mathcal{H}_{n,1}, \tag*{(14)} S H n = J ⊆ [ n − 1 ] ∑ Θ J H n , 1 , (14)
where the Θ r \Theta_r Θ r anticommute and a fixed order is used in each product. Every ordinary harmonic is a linear combination of partial derivatives of Δ n \Delta_n Δ n . Under (4.3), the form Θ r 1 ⋯ Θ r j h \Theta_{r_1}\cdots\Theta_{r_j}h Θ r 1 ⋯ Θ r j h becomes
∑ σ ∈ S j sgn ( σ ) E r 1 ( σ ( 1 ) ) ⋯ E r j ( σ ( j ) ) h , E s ( a ) = ∑ i y i ( a ) ∂ x i . (15) \sum_{\sigma\in S_j}\operatorname{sgn}(\sigma)E_{r_1}^{(\sigma(1))}\cdots E_{r_j}^{(\sigma(j))}h,
\qquad
E_s^{(a)}=\sum_i y_i^{(a)}\partial_{x_i}. \tag*{(15)} σ ∈ S j ∑ sgn ( σ ) E r 1 ( σ ( 1 )) ⋯ E r j ( σ ( j )) h , E s ( a ) = i ∑ y i ( a ) ∂ x i . (15)
The even operators in this formula commute. Expansion of both sides verifies the formula term by term. It belongs to P n , j + 1 \mathcal{P}_{n,j+1} P n , j + 1 . Thus the entire subspace ker ι ∩ S H n , j a \ker\iota\cap SH_{n,j}^{a} ker ι ∩ S H n , j a is represented inside P n , j + 1 \mathcal{P}_{n,j+1} P n , j + 1 . By Lemma 4.1, this is the full row-hook highest-weight multiplicity space. Closing under GL j + 1 \operatorname{GL}_{j+1} GL j + 1 , and then adjoining unused rows and closing under GL k \operatorname{GL}_k GL k , proves the proposition.
For a nontrivial hook label μ = ( a + 1 , 1 b ) \mu=(a+1,1^b) μ = ( a + 1 , 1 b ) , a + b = n − 1 a+b=n-1 a + b = n − 1 , b ≥ 1 b\geq1 b ≥ 1 , the tableau formula and row-hook generation give
P μ [ q ∣ u ] q + u = [ n − 1 b ] ∏ r = 2 b ( q r + u ) . (16) \frac{\mathcal{P}_{\mu}[q\mid u]}{q+u}
=
\begin{bmatrix} n-1 \\ b \end{bmatrix}
\prod_{r=2}^{b}(q^r+u). \tag*{(16)} q + u P μ [ q ∣ u ] = [ n − 1 b ] r = 2 ∏ b ( q r + u ) . (16)
Here the supersymmetric specialization is characterized by p r ↦ q r + ( − 1 ) r − 1 u r p_r\mapsto q^r+(-1)^{r-1}u^r p r ↦ q r + ( − 1 ) r − 1 u r , and the bracket is the Gaussian binomial coefficient. To verify the formula, every tableau of this hook has b b b descents, and their possible sets are all b b b -element subsets of [ n − 1 ] [n-1] [ n − 1 ] . Thus their major-index generating polynomial is q b ( b + 1 ) / 2 [ n − 1 b ] q q^{b(b+1)/2}\begin{bmatrix}n-1\\b\end{bmatrix}_q q b ( b + 1 ) /2 [ n − 1 b ] q . Now use s ( c , 1 j ) [ q ∣ u ] = ( q + u ) q c − 1 u j s_{(c,1^j)}[q\mid u]=(q+u)q^{c-1}u^j s ( c , 1 j ) [ q ∣ u ] = ( q + u ) q c − 1 u j and Theorem 4.2. All other nonempty row shapes vanish under this specialization. This proves the nontrivial hook-label cases of Bergeron’s Conjecture 4 [4 Section 4.2, equation (4.4)] .
Maximal row hooks and exact coefficient lengths
At the longest possible hook length, the tableau formula forces the maximum descent number. Transposing the tableaux turns this into a minimum-descent condition, which is counted by a finite principal specialization. This gives the whole multiplicity series at once.
Theorem 4.4 (The complete maximal-length row-hook series). Let μ = ( c , τ ) ⊢ n \mu=(c,\tau)\vdash n μ = ( c , τ ) ⊢ n and r = ∣ τ ∣ > 0 r=|\tau|>0 r = ∣ τ ∣ > 0 . Then, in both the harmonic and Vandermonde modules,
H μ ( q ) : = ∑ a ≥ 1 [ K a , r − 1 : S μ ] q a = q s τ ′ ( 1 , q , … , q c − 1 ) . H_{\mu}(q):=\sum_{a\geq1}[K_{a,r-1}:S^{\mu}]q^{a}=q\,s_{\tau'}(1,q,\ldots,q^{c-1}). H μ ( q ) := a ≥ 1 ∑ [ K a , r − 1 : S μ ] q a = q s τ ′ ( 1 , q , … , q c − 1 ) . Thus all maximal-length row-hook coefficients are given by one finite principal specialization of a Schur polynomial.
Proof. A tableau of shape μ \mu μ has at most r r r descents. In (12) , exterior degree r − 1 r-1 r − 1 therefore forces exactly r r r descents and selects the whole set { 2 , … , r } \{2,\ldots,r\} { 2 , … , r } . Transposition complements the descent set. If N = ( n 2 ) N=\binom{n}{2} N = ( 2 n ) , then
H μ ( q ) = q N − r ( r + 1 ) / 2 + 1 ∑ U ∈ SYT ( μ ′ ) des ( U ) = c − 1 q − maj ( U ) = q N − r ( r + 1 ) / 2 + 1 s μ ′ ( 1 , q − 1 , … , q − ( c − 1 ) ) . H_{\mu}(q)=q^{N-r(r+1)/2+1}\sum_{\substack{U\in\operatorname{SYT}(\mu')\\ \operatorname{des}(U)=c-1}}q^{-\operatorname{maj}(U)}
=q^{N-r(r+1)/2+1}s_{\mu'}(1,q^{-1},\ldots,q^{-(c-1)}). H μ ( q ) = q N − r ( r + 1 ) /2 + 1 U ∈ SYT ( μ ′ ) des ( U ) = c − 1 ∑ q − maj ( U ) = q N − r ( r + 1 ) /2 + 1 s μ ′ ( 1 , q − 1 , … , q − ( c − 1 ) ) .
For the second equality, use the standard principal-specialization identity on exactly c = ℓ ( μ ′ ) c=\ell(\mu') c = ℓ ( μ ′ ) variables. Standardization gives a sum of q maj ( U ) q^{\operatorname{maj}(U)} q maj ( U ) times Gaussian coefficients; only the minimum descent number c − 1 c-1 c − 1 survives, with Gaussian coefficient one. One may standardize using comajor index and use tableau evacuation to replace comajor by major index.
Now μ ′ = ( 1 c ) + τ ′ \mu'=(1^c)+\tau' μ ′ = ( 1 c ) + τ ′ , with τ ′ \tau' τ ′ padded by zeros. Factoring the determinant contributes q ( c 2 ) q^{\binom{c}{2}} q ( 2 c ) . Reverse the remaining c c c variables and use homogeneity of s τ ′ s_{\tau'} s τ ′ , of degree r r r , to obtain an additional factor q − r ( c − 1 ) q^{-r(c-1)} q − r ( c − 1 ) . Since
N − r ( r + 1 ) 2 − ( c 2 ) − r ( c − 1 ) = 0 , N-\frac{r(r+1)}{2}-\binom{c}{2}-r(c-1)=0, N − 2 r ( r + 1 ) − ( 2 c ) − r ( c − 1 ) = 0 ,
the total remaining factor is q q q , proving the formula. □
Corollary 4.5 (A positive contribution at every permitted hook length). With μ = ( c , τ ) \mu=(c,\tau) μ = ( c , τ ) and r = ∣ τ ∣ > 0 r=|\tau|>0 r = ∣ τ ∣ > 0 , fix 1 ≤ ℓ ≤ r 1\leq\ell\leq r 1 ≤ ℓ ≤ r . In the row-hook coefficient series of length ℓ \ell ℓ , the contribution of tableaux with exactly r r r descents is
q ( r − ℓ + 2 ) / 2 ( r − 1 ℓ − 1 ) q s τ ′ ( 1 , q , … , q c − 1 ) . q^{(r-\ell+2)/2}\binom{r-1}{\ell-1}_{q}s_{\tau'}(1,q,\ldots,q^{c-1}). q ( r − ℓ + 2 ) /2 ( ℓ − 1 r − 1 ) q s τ ′ ( 1 , q , … , q c − 1 ) . All other contributions have nonnegative coefficients. Hence, if b = ∑ i ( τ i 2 ) b=\sum_i\binom{\tau_i}{2} b = ∑ i ( 2 τ i ) , the row hook
( ( r − ℓ + 2 2 ) + b , ℓ ℓ − 1 ) \left(\binom{r-\ell+2}{2}+b,\ell^{\ell-1}\right) ( ( 2 r − ℓ + 2 ) + b , ℓ ℓ − 1 ) occurs. Taking ℓ = min ( k , r ) \ell=\min(k,r) ℓ = min ( k , r ) gives the sharp coefficient length in k k k alphabets directly.
Proof. The proof of Theorem 4.4 gives
∑ des ( T ) = r q maj ( T ) = q r ( r + 1 ) / 2 s τ ′ ( 1 , q , … , q c − 1 ) . \sum_{\operatorname{des}(T)=r}q^{\operatorname{maj}(T)}
=q^{r(r+1)/2}s_{\tau'}(1,q,\ldots,q^{c-1}). des ( T ) = r ∑ q maj ( T ) = q r ( r + 1 ) /2 s τ ′ ( 1 , q , … , q c − 1 ) .
Multiply by
e ℓ − 1 ( q − 2 , … , q − r ) = q − r ( ℓ − 1 ) + ( ℓ − 1 2 ) ( r − 1 ℓ − 1 ) q e_{\ell-1}(q^{-2},\ldots,q^{-r})
=q^{-r(\ell-1)+\binom{\ell-1}{2}}\binom{r-1}{\ell-1}_{q} e ℓ − 1 ( q − 2 , … , q − r ) = q − r ( ℓ − 1 ) + ( 2 ℓ − 1 ) ( ℓ − 1 r − 1 ) q
and simplify the exponent. Its least coefficient is positive. □
Theorem 4.6 (Exact coefficient lengths). For μ ⊢ n \mu\vdash n μ ⊢ n , put r = n − μ 1 r=n-\mu_1 r = n − μ 1 . The largest length of a row partition in the μ \mu μ -isotype of either H n , k H_{n,k} H n , k or P n , k P_{n,k} P n , k is min ( k , r ) \min(k,r) min ( k , r ) . The corresponding stable largest length is r r r .
Proof. Proposition 3.3 gives the upper bound. If μ = ( c , τ ) \mu=(c,\tau) μ = ( c , τ ) and r = ∣ τ ∣ > 0 r=|\tau|>0 r = ∣ τ ∣ > 0 , take ℓ = min ( k , r ) \ell=\min(k,r) ℓ = min ( k , r ) in Corollary 4.5. Its positive contribution contains a row hook of length ℓ \ell ℓ ; Proposition 4.3 places the whole component in P n , k P_{n,k} P n , k . If r = 0 r=0 r = 0 , exactness of taking S n S_n S n -invariants makes the trivial isotype of R n , k R_{n,k} R n , k equal to K K K , and constants belong to P n , k P_{n,k} P n , k . □
Corollary 4.7 (Palindromicity, unimodality, and the extremal hook). Put b = ∑ i ( τ i 2 ) b=\sum_i\binom{\tau_i}{2} b = ∑ i ( 2 τ i ) . The coefficients of H μ ( q ) H_\mu(q) H μ ( q ) are palindromic and unimodal on the interval
a min = 1 + b , a max = 1 + r ( c − 1 ) − b , a_{\min}=1+b,\qquad a_{\max}=1+r(c-1)-b, a m i n = 1 + b , a m a x = 1 + r ( c − 1 ) − b , with both endpoint coefficients equal to one. Their center is 1 + r ( c − 1 ) / 2 1+r(c-1)/2 1 + r ( c − 1 ) /2 . In particular the least-degree row hook of maximal possible length r r r is
λ μ = ( 1 + b , 1 r − 1 ) , c λ μ , μ ( n ) = 1. (17) \lambda_\mu=(1+b,1^{r-1}),\qquad c_{\lambda_\mu,\mu}^{(n)}=1. \tag*{(17)} λ μ = ( 1 + b , 1 r − 1 ) , c λ μ , μ ( n ) = 1. (17) Its total degree is r + b = ∑ i ( τ i + 1 2 ) r+b=\sum_i\binom{\tau_i+1}{2} r + b = ∑ i ( 2 τ i + 1 ) . The sum of all maximal-length row-hook multiplicities is
H μ ( 1 ) = dim S τ ′ K c = ∏ 1 ≤ i < j ≤ c τ i ′ − τ j ′ + j − i j − i . H_\mu(1)=\dim S_{\tau'}K^c=\prod_{1\le i<j\le c}\frac{\tau'_i-\tau'_j+j-i}{j-i}. H μ ( 1 ) = dim S τ ′ K c = 1 ≤ i < j ≤ c ∏ j − i τ i ′ − τ j ′ + j − i . An exact product form is
H μ ( q ) = q 1 + b ∏ ( i , j ) ∈ τ 1 − q c + i − j 1 − q h τ ( i , j ) . H_\mu(q)=q^{1+b}\prod_{(i,j)\in\tau}\frac{1-q^{c+i-j}}{1-q^{h_\tau(i,j)}}. H μ ( q ) = q 1 + b ( i , j ) ∈ τ ∏ 1 − q h τ ( i , j ) 1 − q c + i − j . Proof. The least principal-specialization exponent is n ( τ ′ ) = ∑ i ( τ i 2 ) = b n(\tau')=\sum_i\binom{\tau_i}{2}=b n ( τ ′ ) = ∑ i ( 2 τ i ) = b , with coefficient one; reversal of the c c c variables gives the greatest exponent r ( c − 1 ) − b r(c-1)-b r ( c − 1 ) − b , also with coefficient one. This proves the endpoints and palindromicity.
For unimodality, apply the standard s l 2 \mathfrak{sl}_2 sl 2 criterion [28 , 34 ] : regard K c K^c K c as the irreducible s l 2 \mathfrak{sl}_2 sl 2 -module Sym c − 1 K 2 \operatorname{Sym}^{c-1}K^2 Sym c − 1 K 2 . Then
t − r ( c − 1 ) s τ ′ ( 1 , t 2 , … , t 2 ( c − 1 ) ) t^{-r(c-1)}s_{\tau'}(1,t^2,\ldots,t^{2(c-1)}) t − r ( c − 1 ) s τ ′ ( 1 , t 2 , … , t 2 ( c − 1 ) ) is the s l 2 \mathfrak{sl}_2 sl 2 character of S τ ′ K c S_{\tau'}K^c S τ ′ K c . Its decomposition into irreducibles expresses its coefficients as a sum of indicator functions of centered weight intervals. The coefficients therefore increase weakly up to the center and decrease weakly thereafter. § 8 constructs the operators on harmonic multiplicities. The Weyl dimension and hook-content formulas give the last two identities. □
For μ = ( 3 , 3 ) \mu=(3,3) μ = ( 3 , 3 ) this hook is ( 4 , 1 , 1 ) (4,1,1) ( 4 , 1 , 1 ) ; for the alternating label it is ( 1 n − 1 ) (1^{n-1}) ( 1 n − 1 ) .
Corollary 4.8 (Small row support and operator recovery). The stable character C μ ( n ) C_\mu^{(n)} C μ ( n ) is determined by its specialization to n − μ 1 n-\mu_1 n − μ 1 row variables. If μ 1 ≥ n − 2 \mu_1\ge n-2 μ 1 ≥ n − 2 , then the μ \mu μ -isotypes of P n , k P_{n,k} P n , k and H n , k H_{n,k} H n , k agree for every k k k .
Proof. Schur functions of length at most r r r remain linearly independent in r r r variables. The length bound therefore proves the first assertion. For the second, Haiman’s bivariate operator theorem [19 Theorem 4.2] gives P n , 2 = H n , 2 P_{n,2}=H_{n,2} P n , 2 = H n , 2 . Every highest-weight vector in the specified labels uses at most two rows, by the length bound and Lemma 3.1 . It is therefore generated in those rows. The row group generates its full isotypic component in any larger number of rows. The one-row assertion is classical. □
Corollary 4.9. In three alphabets, a nonzero component of H n , 3 / P n , 3 H_{n,3}/P_{n,3} H n , 3 / P n , 3 can occur only at a pair ( λ , μ ) (\lambda,\mu) ( λ , μ ) satisfying
λ = ( a , b , c ) , a ≥ b ≥ 2 , c ≥ 1 , μ 1 ≤ n − 3. \lambda=(a,b,c),\qquad a\ge b\ge2,\qquad c\ge1,\qquad\mu_1\le n-3. λ = ( a , b , c ) , a ≥ b ≥ 2 , c ≥ 1 , μ 1 ≤ n − 3. Proof. All row partitions of length at most two are generated by the bivariate theorem and trailing-zero stability. The remaining row hooks are generated by Proposition 4.3. Corollary 4.8 excludes all label types with μ 1 ≥ n − 2 \mu_1\ge n-2 μ 1 ≥ n − 2 . □
A commuting operator model
All generalized polarizations act as vector fields on a single family of pointed jet determinants. An invertible source jet can be reparametrized to t t t ; the other jet coefficients then become commuting polarizations from that source row. Polynomial dependence extends this description across the singular jets. The resulting normal form makes specialization in the number of alphabets exact.
Lemma 5.1 (Normal ordering). Let V V V be a finite-dimensional row space over a field of characteristic zero. Suppose M ( V ) \mathcal{M}(V) M ( V ) is generated from alternating harmonic polynomials by partial derivatives and generalized polarizations. Let B ( V ) \mathcal{B}(V) B ( V ) be the span of the words using only generalized polarizations on those generators. Then
M ( V ) = K [ ∂ ] B ( V ) , M ( V ) sgn = B ( V ) . \mathcal{M}(V)=K[\partial]\mathcal{B}(V),\qquad\mathcal{M}(V)^{\operatorname{sgn}}=\mathcal{B}(V). M ( V ) = K [ ∂ ] B ( V ) , M ( V ) sgn = B ( V ) . In particular, the alternating part of P \mathcal{P} P is generated from the ordinary Vandermonde by generalized polarizations alone.
Proof. For E = ∑ i x i ( a ) ( ∂ x i ( b ) ) r E=\sum_i x_i^{(a)}(\partial_{x_i^{(b)}})^r E = ∑ i x i ( a ) ( ∂ x i ( b ) ) r ,
[ E , ∂ x j ( c ) ] = − δ a c ( ∂ x j ( b ) ) r . [E,\partial_{x_j^{(c)}}]=-\delta_{ac}(\partial_{x_j^{(b)}})^r. [ E , ∂ x j ( c ) ] = − δ a c ( ∂ x j ( b ) ) r .
More generally, the commutator of E E E with any constant coefficient differential operator is again such an operator. Commuting partial derivatives to the left, by induction on the number of polarization operators and on their positions in a word, proves the first identity. Every element of B ( V ) \mathcal{B}(V) B ( V ) is alternating, because the polarizations commute with S n S_n S n .
Apply the sign projector to an expression ∑ ℓ D ℓ g ℓ \sum_\ell D_\ell g_\ell ∑ ℓ D ℓ g ℓ , where g ℓ ∈ B ( V ) g_\ell\in\mathcal{B}(V) g ℓ ∈ B ( V ) . Since each g ℓ g_\ell g ℓ is alternating, the result is ∑ ℓ Av S n ( D ℓ ) g ℓ \sum_\ell\operatorname{Av}_{S_n}(D_\ell)g_\ell ∑ ℓ Av S n ( D ℓ ) g ℓ . The positive-degree invariant differential operators annihilate every diagonal harmonic. Only the constant term of each averaged differential operator remains. This proves the second identity. □ \square □
Put N = ( n 2 ) N=\binom{n}{2} N = ( 2 n ) , write x i = x i ( 1 ) x_i=x_i^{(1)} x i = x i ( 1 ) , and normalize
Δ n = det ( x i j ) 1 ≤ i ≤ n 0 ≤ j < n . \Delta_n=\det(x_i^j)_{\substack{1\le i\le n\\0\le j<n}}. Δ n = det ( x i j ) 1 ≤ i ≤ n 0 ≤ j < n .
All fields in this section have characteristic zero. For a ≥ 2 a\ge2 a ≥ 2 put Y a , s = ∑ i x i ( a ) ∂ x i s Y_{a,s}=\sum_i x_i^{(a)}\partial_{x_i}^{s} Y a , s = ∑ i x i ( a ) ∂ x i s . The operators Y a , s Y_{a,s} Y a , s and the individual ∂ x i \partial_{x_i} ∂ x i commute. Let B n , k \mathcal{B}_{n,k} B n , k be the span of all words in generalized polarizations applied to Δ n \Delta_n Δ n .
Theorem 5.2 (Commuting normal form). For every n , k ≥ 1 n,k\ge1 n , k ≥ 1 ,
B n , k = K [ Y a , s : 2 ≤ a ≤ k , 1 ≤ s < n ] Δ n , (18) \mathcal{B}_{n,k}=K[Y_{a,s}:2\le a\le k,\ 1\le s<n]\Delta_n, \tag*{(18)} B n , k = K [ Y a , s : 2 ≤ a ≤ k , 1 ≤ s < n ] Δ n , (18) P n , k = K [ ∂ x 1 , … , ∂ x n , Y a , s : 2 ≤ a ≤ k , 1 ≤ s < n ] Δ n . (19) \mathcal{P}_{n,k}=K[\partial_{x_1},\ldots,\partial_{x_n},Y_{a,s}:2\le a\le k,\ 1\le s<n]\Delta_n. \tag*{(19)} P n , k = K [ ∂ x 1 , … , ∂ x n , Y a , s : 2 ≤ a ≤ k , 1 ≤ s < n ] Δ n . (19) Moreover, P n , k sgn = B n , k \mathcal{P}_{n,k}^{\operatorname{sgn}}=\mathcal{B}_{n,k} P n , k sgn = B n , k . Thus one source row and a commuting family of operators suffice for the module defined using arbitrary generalized polarizations.
Proof. The assertion for n = 1 n=1 n = 1 is immediate, so assume n ≥ 2 n\ge2 n ≥ 2 . Let t = t K [ t ] / ( t n ) \mathfrak{t}=tK[t]/(t^n) t = t K [ t ] / ( t n ) , and for f = ( f 1 , … , f k ) ∈ t k f=(f_1,\ldots,f_k)\in\mathfrak{t}^k f = ( f 1 , … , f k ) ∈ t k define
G f ( X ) = ( ∏ j = 0 n − 1 j ! ) det ( [ t j ] exp ( ∑ a = 1 k x i ( a ) f a ( t ) ) ) i , j . (20) G_f(X)=\left(\prod_{j=0}^{n-1}j!\right)\det\left([t^j]\exp\left(\sum_{a=1}^{k}x_i^{(a)}f_a(t)\right)\right)_{i,j}. \tag*{(20)} G f ( X ) = ( j = 0 ∏ n − 1 j ! ) det ( [ t j ] exp ( a = 1 ∑ k x i ( a ) f a ( t ) ) ) i , j . (20)
Only coefficients of order less than n n n are used. Thus this is a polynomial in X X X and in the finitely many coefficients of f f f , of X X X -degree at most N N N . Let J J J be the span of its values as f f f varies.
The one-particle identity
x ( a ) ∂ x ( b ) e ∑ c x ( c ) f c ( t ) = x ( a ) f b ( t ) s e ∑ c x ( c ) f c ( t ) x^{(a)}\partial_{x^{(b)}}e^{\sum_c x^{(c)}f_c(t)}
=x^{(a)}f_b(t)^s e^{\sum_c x^{(c)}f_c(t)} x ( a ) ∂ x ( b ) e ∑ c x ( c ) f c ( t ) = x ( a ) f b ( t ) s e ∑ c x ( c ) f c ( t )
and differentiation of a determinant give
E a , b ( s ) G f = d d ϵ G f + ϵ e a f b s ∣ ϵ = 0 . (21) E_{a,b}^{(s)}G_f=\left.\frac{d}{d\epsilon}G_{f+\epsilon e_af_b^s}\right|_{\epsilon=0}. \tag*{(21)} E a , b ( s ) G f = d ϵ d G f + ϵ e a f b s ϵ = 0 . (21)
The product f b s f_b^s f b s is taken modulo t n t^n t n and has zero constant term. A derivative of a polynomial family lies in the span of its values, by polynomial interpolation over an infinite field. Consequently J J J is stable under every generalized polarization, including those with output in the source row. It contains G ( t , 0 , … , 0 ) = Δ n G_{(t,0,\ldots,0)}=\Delta_n G ( t , 0 , … , 0 ) = Δ n , and hence B n , k ⊆ J \mathcal{B}_{n,k}\subseteq J B n , k ⊆ J . Suppose that f 1 ′ ( 0 ) ≠ 0 f_1'(0) \ne0 f 1 ′ ( 0 ) = 0 , and let g ∈ t g \in\mathfrak{t} g ∈ t be its compositional inverse modulo t n t^n t n . Composition by g g g acts on coefficients of degrees 0 , … , n − 1 0,\ldots,n-1 0 , … , n − 1 through a triangular matrix with diagonal 1 , g ′ ( 0 ) , … , g ′ ( 0 ) n − 1 1,g'(0),\ldots,g'(0)^{n-1} 1 , g ′ ( 0 ) , … , g ′ ( 0 ) n − 1 . Therefore
G f ∘ g = g ′ ( 0 ) N G f . (22) G_{f\circ g}=g'(0)^N G_f. \tag*{(22)} G f ∘ g = g ′ ( 0 ) N G f . (22)
Since f 1 ∘ g = t f_1\circ g=t f 1 ∘ g = t , write f a ∘ g = ∑ s = 1 n − 1 c a , s t s f_a\circ g=\sum_{s=1}^{n-1}c_{a,s}t^s f a ∘ g = ∑ s = 1 n − 1 c a , s t s for a ≥ 2 a\ge2 a ≥ 2 . Expanding the commuting operator exponential gives
G f ∘ g = exp ( ∑ a = 2 k ∑ s = 1 n − 1 c a , s Y a , s ) Δ n . (23) G_{f\circ g}=\exp\left(\sum_{a=2}^{k}\sum_{s=1}^{n-1}c_{a,s}Y_{a,s}\right)\Delta_n. \tag*{(23)} G f ∘ g = exp ( a = 2 ∑ k s = 1 ∑ n − 1 c a , s Y a , s ) Δ n . (23)
The exponential is finite on Δ n \Delta_n Δ n , since every Y a , s Y_{a,s} Y a , s lowers its source-row degree by s s s . Thus G f G_f G f belongs to the right side of (5.1) whenever f 1 ′ ( 0 ) ≠ 0 f_1'(0)\ne0 f 1 ′ ( 0 ) = 0 . That condition defines a dense open subset of the jet parameter space. A polynomial family taking values in a fixed linear subspace on a dense subset takes values in that subspace everywhere: apply any linear functional vanishing on the subspace. It follows that J \mathcal{J} J is contained in the right side of (5.1). The reverse inclusion is immediate, which proves the first identity.
Lemma 5.1 gives P n , k = K [ ∂ x i ( a ) ] B n , k \mathcal{P}_{n,k}=K[\partial_{x_i^{(a)}}]\mathcal{B}_{n,k} P n , k = K [ ∂ x i ( a ) ] B n , k . Substitute (5.1) and move every derivative in an auxiliary row to the right, using
[ ∂ x i ( a ) , Y b , s ] = δ a b ∂ x i ( a ) ( a , b ≥ 2 ) . [\partial_{x_i^{(a)}},Y_{b,s}]=\delta_{ab}\partial_{x_i^{(a)}}\qquad(a,b\ge2). [ ∂ x i ( a ) , Y b , s ] = δ ab ∂ x i ( a ) ( a , b ≥ 2 ) .
Those derivatives kill Δ n \Delta_n Δ n when they reach it. The remaining operators are exactly those in (5.2).
The same normal-ordering lemma identifies the alternating part with B n , k \mathcal{B}_{n,k} B n , k . □
Positive-order polarizations preserve the pointed condition f a ( 0 ) = 0 f_a(0)=0 f a ( 0 ) = 0 in (5.4). Individual translations of the source variables encode partial derivatives and give the following finite generating family.
Corollary 5.3 (A finite generating determinant). The coefficient span, in the parameters u i u_i u i and c a , s c_{a,s} c a , s , of
( ∏ j = 0 n − 1 j ! ) det ( [ t j ] exp ( ( x i + u i ) t + ∑ a = 2 k x i ( a ) ∑ s = 1 n − 1 c a , s t s ) ) i , j (24) \left(\prod_{j=0}^{n-1}j!\right)\det\left([t^j]\exp\left((x_i+u_i)t+\sum_{a=2}^{k}x_i^{(a)}\sum_{s=1}^{n-1}c_{a,s}t^s\right)\right)_{i,j} \tag*{(24)} ( j = 0 ∏ n − 1 j ! ) det ( [ t j ] exp ( ( x i + u i ) t + a = 2 ∑ k x i ( a ) s = 1 ∑ n − 1 c a , s t s ) ) i , j (24) is exactly P n , k \mathcal{P}_{n,k} P n , k . Without the parameters u i u_i u i , the coefficient span is its alternating part. A spanning set consists of the words
∂ x α ∏ a , s Y a , s m a , s Δ n , ∣ α ∣ + ∑ a , s s m a , s ≤ N . \partial_x^\alpha\prod_{a,s}Y_{a,s}^{m_{a,s}}\Delta_n,\qquad|\alpha|+\sum_{a,s}sm_{a,s}\le N. ∂ x α a , s ∏ Y a , s m a , s Δ n , ∣ α ∣ + a , s ∑ s m a , s ≤ N .
Proof. The determinant is exp ( ∑ i u i ∂ x i + ∑ a , s c a , s Y a , s ) Δ n \exp(\sum_i u_i\partial_{x_i}+\sum_{a,s}c_{a,s}Y_{a,s})\Delta_n exp ( ∑ i u i ∂ x i + ∑ a , s c a , s Y a , s ) Δ n . Use Theorem 5.2 and compare coefficients. The displayed inequality is the available source-row degree. □
Corollary 5.4 (Exact specialization in the number of rows). If r ≥ k r\ge k r ≥ k , setting rows k + 1 , … , r k+1,\ldots,r k + 1 , … , r equal to zero maps P n , r \mathcal{P}_{n,r} P n , r onto P n , k \mathcal{P}_{n,k} P n , k . Moreover,
P n , r ∩ K [ x i ( a ) : a ≤ k ] = P n , k . \mathcal{P}_{n,r}\cap K[x_i^{(a)}:a\le k]=\mathcal{P}_{n,k}. P n , r ∩ K [ x i ( a ) : a ≤ k ] = P n , k . The stable row character specializes to the actual finite-row module for every k k k .
Proof. In the commuting normal form, the auxiliary-row degree of a word is exactly the number of its corresponding Y Y Y factors. Setting a row to zero kills precisely the words using a polarization into that row. All remaining words belong to P n , k \mathcal{P}_{n,k} P n , k , and every such word already occurs in P n , r \mathcal{P}_{n,r} P n , r . If a polynomial is independent of the discarded rows, restriction fixes it, proving the intersection assertion. For a polynomial row representation, the zero-weight subspace in the discarded rows is the usual specialization of its Schur character. □
Haiman’s bivariate operator theorem [19 Theorem 4.2] describes the full bivariate harmonic space using the commuting operators in (19) . Bergeron’s multivariate module [4 Section 2.1] is defined by closure under derivatives in all variables and polarizations between arbitrary pairs of rows. Theorem 5.2 identifies these operator descriptions in every number of rows and makes their finite-row specialization exact.
The Hook-Components conjecture and reconstruction
The sign representation turns an exterior degree into a hook label. To use this observation on coinvariants, we show that every positive odd trace already vanishes on the sign component once the even traces have been removed. The argument uses multiplication and augmentation alone, so we work at the level of an arbitrary pointed algebra.
This section works over any commutative unital ring K K K in which n ! n! n ! is invertible. Let A A A be a commutative unital K K K -algebra with an augmentation ε : A → K \varepsilon:A\to K ε : A → K , and let m = ker ε \mathfrak{m}=\ker\varepsilon m = ker ε . Set
B = A ⊗ n , ε B = ε ⊗ n . B=A^{\otimes n},\qquad\varepsilon_B=\varepsilon^{\otimes n}. B = A ⊗ n , ε B = ε ⊗ n .
The symmetric group permutes the tensor factors. For a ∈ A a\in A a ∈ A , let a ( i ) a^{(i)} a ( i ) be a a a in the i i i th tensor factor and 1 1 1 in all other factors. Let N n N_n N n be the permutation representation with basis θ 1 , … , θ n \theta_1,\ldots,\theta_n θ 1 , … , θ n , and set
S = B ⊗ Λ N n . S=B\otimes\Lambda N_n. S = B ⊗ Λ N n .
Its augmentation ε S \varepsilon_S ε S extends ε B \varepsilon_B ε B and sends every θ i \theta_i θ i to zero.
Definition 6.1. The ordinary and exterior coinvariant quotients at the augmentation are
I A = B ker ( ε B ∣ B S n ) , R A = B / I A , I_A=B\ker(\varepsilon_B|_{B^{S_n}}),\qquad R_A=B/I_A, I A = B ker ( ε B ∣ B S n ) , R A = B / I A ,
and
J A = S ker ( ε S ∣ S S n ) , Q A = S / J A . J_A=S\ker(\varepsilon_S|_{S^{S_n}}),\qquad Q_A=S/J_A. J A = S ker ( ε S ∣ S S n ) , Q A = S / J A .
If A A A is connected and graded, these are the usual quotients by invariants with zero constant term. The augmentation definition also makes sense for ungraded and nonreduced algebras. Put θ Σ = ∑ i θ i \theta_\Sigma=\sum_i\theta_i θ Σ = ∑ i θ i . As n n n is invertible, N n / K θ Σ N_n/K\theta_\Sigma N n / K θ Σ identifies canonically with the standard representation W n W_n W n .
Theorem 6.2 (Augmented sign transfer). The natural quotient induces an isomorphism
( R A ⊗ Λ W n ) s g n → ∼ Q A s g n (25) (R_A\otimes\Lambda W_n)^{\mathrm{sgn}}\xrightarrow{\sim}Q_A^{\mathrm{sgn}} \tag*{(25)} ( R A ⊗ Λ W n ) sgn ∼ Q A sgn (25) It preserves exterior degree, is natural under augmented algebra homomorphisms, commutes with arbitrary base change, and commutes with all augmented automorphisms of A A A . It also preserves any additional grading respected by the augmentation.
For a polynomial algebra in two generators, this is the sign-transfer theorem of González, Lentfer and Mularczyk [18 Theorem 4.21] . Their complementary-averaging and determinant argument [18 Lemmas 4.13–4.15] extends to the present setting because collisions multiply elements of the algebra and positive traces remain in its augmentation ideal.
Trace generators
For a ∈ A a\in A a ∈ A and T ⊆ [ n ] T\subseteq[n] T ⊆ [ n ] , write
t a ( T ) = ∑ i ∈ T a ( i ) , q a = ∑ i = 1 n θ i a ( i ) , t a = t a ( [ n ] ) . t_a(T)=\sum_{i\in T}a^{(i)},\qquad q_a=\sum_{i=1}^{n}\theta_i a^{(i)},\qquad t_a=t_a([n]). t a ( T ) = i ∈ T ∑ a ( i ) , q a = i = 1 ∑ n θ i a ( i ) , t a = t a ([ n ]) .
Lemma 6.3. The algebra B S n B^{S_n} B S n is generated by 1 1 1 and t a t_a t a for a ∈ m a\in\mathfrak{m} a ∈ m . The superalgebra S S n S^{S_n} S S n is generated by these even elements together with q a q_a q a for a ∈ A a\in A a ∈ A . Consequently
I A = ⟨ t a : a ∈ m ⟩ , J A = ⟨ t a : a ∈ m ; q a : a ∈ A ⟩ . (26) I_A=\langle t_a:a\in\mathfrak{m}\rangle,\qquad J_A=\langle t_a:a\in\mathfrak{m};\ q_a:a\in A\rangle. \tag*{(26)} I A = ⟨ t a : a ∈ m ⟩ , J A = ⟨ t a : a ∈ m ; q a : a ∈ A ⟩ . (26)
Proof. Use the split K K K -module decomposition A = K 1 ⊕ m A = K1 \oplus\mathfrak{m} A = K 1 ⊕ m . Pure tensors in B B B , together with exterior monomials, span S S S as a K K K -module; averaging spans its invariant submodule. An averaged tensor can be described by finitely many occupied labels, each carrying an even decoration from m \mathfrak{m} m or an odd decoration θ a \theta a θ a with a ∈ A a \in A a ∈ A .
Multiply the traces corresponding to those decorations. The terms with pairwise distinct labels give the desired orbit sum, up to the multiplicity factorials of repeated even decorations. These factorials divide n ! n! n ! . A repeated odd decoration has zero distinct-label antisymmetrization, so no division by a vanishing scalar is required. Every other term has a collision and uses fewer occupied labels. Two odd decorations at one label give zero. An even collision is multiplication in A A A , and a collision involving an even decoration from m \mathfrak{m} m stays in m \mathfrak{m} m . Induction on the number of occupied labels proves the two generation statements. All generators except 1 1 1 have zero augmentation, giving (6.2). □
For a set T T T of labels, let Av T \operatorname{Av}_T Av T and Alt T \operatorname{Alt}_T Alt T denote the normalized symmetric and alternating averaging operators over S T S_T S T . The averaging identity below proves sign transfer. A trace relation moves a factor from the complementary labels onto an alternating block, where the resulting determinant has a repeated row and vanishes.
Lemma 6.4 (Complementary averaging). Let D ⊆ [ n ] D \subseteq[n] D ⊆ [ n ] , let i ∉ D i \notin D i ∈ / D , and put C = D ∪ { i } C = D \cup\{i\} C = D ∪ { i } . Suppose f ∈ B f \in B f ∈ B is alternating on C C C and symmetric on C c C^c C c . For every a ∈ m a \in\mathfrak{m} a ∈ m ,
Av D c ( a ( i ) f ) ∈ I A . (27) \operatorname{Av}_{D^c}\left(a^{(i)}f\right) \in I_A. \tag*{(27)} Av D c ( a ( i ) f ) ∈ I A . (27) Proof. Separate the tensor factors on C C C and C c C^c C c . The symmetric and alternating averaging operators are idempotents on separate tensor factors. Their images are split direct summands; tensoring these splittings identifies the module with the indicated symmetries as the tensor product of the alternating module on C C C and the invariant module on C c C^c C c . No flatness is required. By Lemma 6.3, the latter is generated by the traces t b ( C c ) t_b(C^c) t b ( C c ) with b ∈ m b \in\mathfrak{m} b ∈ m . Modulo I A I_A I A ,
t b ( C c ) = − t b ( C ) . t_b(C^c) = -t_b(C). t b ( C c ) = − t b ( C ) . After making these substitutions, f f f is congruent to a sum of elements using only the tensor factors on C C C and alternating on C C C . Multiplication by a ( i ) a^{(i)} a ( i ) and averaging preserve congruence modulo I A I_A I A .
Write c 1 , … , c h c_1,\ldots,c_h c 1 , … , c h for the labels of C C C , with c h = i c_h=i c h = i . Alternating tensors on these factors are spanned as a K K K -module by determinants
F = det ( b j ( c ℓ ) ) 1 ≤ ℓ , j ≤ h , b j ∈ A . F = \det\left(b_j^{(c_\ell)}\right)_{1\leq\ell,j\leq h}, \qquad b_j \in A. F = det ( b j ( c ℓ ) ) 1 ≤ ℓ , j ≤ h , b j ∈ A . Indeed, the alternating projector sends a pure tensor to its determinant divided by h ! h! h ! . Thus it suffices to prove (6.3) for F F F . Let M j M_j M j be the cofactor of the entry in its last row and j j j th column. For r ∈ D c r \in D^c r ∈ D c , let F r F_r F r be the same determinant with its last row evaluated at label r r r . The cofactor M j M_j M j uses labels in D D D only. It follows that
∣ D c ∣ Av D c ( a ( i ) F ) = ∑ r ∈ D c a ( r ) F r |D^c| \operatorname{Av}_{D^c}(a^{(i)}F) = \sum_{r\in D^c} a^{(r)}F_r ∣ D c ∣ Av D c ( a ( i ) F ) = r ∈ D c ∑ a ( r ) F r = ∑ j = 1 h M j t a b j ( D c ) = \sum_{j=1}^{h} M_j t_{ab_j}(D^c) = j = 1 ∑ h M j t a b j ( D c ) ≡ − ∑ r ∈ D a ( r ) ∑ j = 1 h M j b j ( r ) ( m o d I A ) \equiv-\sum_{r\in D} a^{(r)} \sum_{j=1}^{h} M_j b_j^{(r)} \pmod{I_A} ≡ − r ∈ D ∑ a ( r ) j = 1 ∑ h M j b j ( r ) ( mod I A ) The congruence uses a b j ∈ m ab_j \in\mathfrak{m} a b j ∈ m . Each final inner sum is a determinant with a repeated row. Since ∣ D c ∣ |D^c| ∣ D c ∣ is invertible, the assertion follows. □
Odd invariants on the sign component.
Lemma 6.5. For a ∈ m a \in\mathfrak{m} a ∈ m , multiplication by q a q_a q a annihilates the sign submodule of R A ⊗ Λ N n R_A \otimes\Lambda N_n R A ⊗ Λ N n .
Proof. Sign tensors are spanned by the images of
Alt [ n ] ( θ J f ) , \operatorname{Alt}_{[n]}(\theta_J f), Alt [ n ] ( θ J f ) , f f f symmetric on J J J and alternating on J c J^c J c , where θ J \theta_J θ J is an exterior monomial with support J J J . Indeed, average first over the stabilizer S J × S J c S_J \times S_{J^c} S J × S J c : the sign of the exterior monomial cancels the sign projector on S J S_J S J .
Since q a q_a q a is invariant, its product with such a tensor is a sum over i ∈ J c i \in J^c i ∈ J c of terms
Alt [ n ] ( θ i θ J a ( i ) f ) . \operatorname{Alt}_{[n]}(\theta_i\theta_J a^{(i)}f). Alt [ n ] ( θ i θ J a ( i ) f ) . Fix one term and put D = J c ∖ { i } D = J^c \setminus\{i\} D = J c ∖ { i } . The exterior factor has support D c = J ∪ { i } D^c = J \cup\{i\} D c = J ∪ { i } . Inside the global sign projector, its coefficient can be replaced by its symmetric average over D c D^c D c : the exterior sign again cancels the sign projector. But f f f is alternating on D ∪ { i } D \cup\{i\} D ∪ { i } and symmetric on its complement. Lemma 6.4 therefore says that the averaged coefficient Av D c ( a ( i ) f ) \operatorname{Av}_{D^c}(a^{(i)}f) Av D c ( a ( i ) f ) is zero in R A R_A R A . Every term vanishes. □
Proof of Theorem 6.2. First quotient S S S by the even generators t a t_a t a , a ∈ m a \in\mathfrak{m} a ∈ m . The result is R A ⊗ Λ N n R_A \otimes\Lambda N_n R A ⊗ Λ N n . Next quotient by q 1 = θ Σ q_1 = \theta_\Sigma q 1 = θ Σ ; the result is R A ⊗ Λ W n R_A \otimes\Lambda W_n R A ⊗ Λ W n . Since A = K 1 ⊕ m A = K1 \oplus\mathfrak{m} A = K 1 ⊕ m , the only generators of J A J_A J A still to impose are q a q_a q a with a ∈ m a \in\mathfrak{m} a ∈ m .
Lemma 6.5 kills these operators on the sign part before imposing q 1 q_1 q 1 . Exactness of the sign projector shows that they also kill the sign part after imposing q 1 q_1 q 1 . Moreover multiplication by each q a q_a q a is S n S_n S n -equivariant. The sign part of its image is therefore its image on the sign part, and is zero. The remaining ideal has no sign component. Quotienting by that ideal preserves the sign submodule and proves (6.1). All the constructions used the canonical quotient maps and averaging projectors, which also proves the asserted naturality and grading compatibility.
For a homomorphism K → K ′ K \to K' K → K ′ , put A ′ = A ⊗ K K ′ A' = A \otimes_K K' A ′ = A ⊗ K K ′ . The averaging and sign projectors are split idempotents, so their images commute with arbitrary scalar extension. The augmentation on each invariant algebra is split by the unit; hence its kernel also commutes with scalar extension. Present I A I_A I A and J A J_A J A as images of the multiplication maps from these kernels. Right exactness of tensor product then gives canonical isomorphisms
R A ⊗ K K ′ ≅ R A ′ , Q A ⊗ K K ′ ≅ Q A ′ . R_A \otimes_K K' \cong R_{A'}, \qquad Q_A \otimes_K K' \cong Q_{A'}. R A ⊗ K K ′ ≅ R A ′ , Q A ⊗ K K ′ ≅ Q A ′ .
The splitting N n = K θ Σ ⊕ W n N_n = K\theta_\Sigma\oplus W_n N n = K θ Σ ⊕ W n and its finite-free exterior powers commute with scalar extension as well. Thus so does (6.1). □
Corollary 6.6 (Hook extraction on a pointed algebra). For 0 ≤ d ≤ n − 1 0 \le d \le n - 1 0 ≤ d ≤ n − 1 there are natural isomorphisms
Hom K [ S n ] ( sgn , Q A [ d ] ) ≅ Hom K [ S n ] ( S ( d + 1 , 1 n − d − 1 ) , R A ) . \operatorname{Hom}_{K[S_n]}(\operatorname{sgn}, Q_A[d]) \cong\operatorname{Hom}_{K[S_n]}(S^{(d+1,1^{n-d-1})}, R_A). Hom K [ S n ] ( sgn , Q A [ d ]) ≅ Hom K [ S n ] ( S ( d + 1 , 1 n − d − 1 ) , R A ) . where [ d ] [d] [ d ] denotes exterior degree and S μ S^\mu S μ is the base change to K K K of the integral Specht module. They commute with arbitrary base change and every augmented automorphism of A A A .
Proof. Theorem 6.2 identifies the left side with Hom K [ S n ] ( sgn , R A ⊗ Λ d W n ) \operatorname{Hom}_{K[S_n]}(\operatorname{sgn}, R_A \otimes\Lambda^d W_n) Hom K [ S n ] ( sgn , R A ⊗ Λ d W n ) . The standard pairing on N n N_n N n restricts to a perfect pairing on W n W_n W n : the complementary vector θ Σ \theta_\Sigma θ Σ has norm n n n , a unit. Since W n W_n W n is finite free and det W n = sgn \det W_n = \operatorname{sgn} det W n = sgn , exterior duality and the integral hook Specht model [15 Section 5.5(b)] give
sgn ⊗ Λ d W n ≅ S ( d + 1 , 1 n − d − 1 ) . \operatorname{sgn} \otimes\Lambda^d W_n \cong S^{(d+1,1^{n-d-1})}. sgn ⊗ Λ d W n ≅ S ( d + 1 , 1 n − d − 1 ) . Finite-free tensor-Hom adjunction gives the result. Equivariant Hom is the image of an averaging projector on ordinary Hom, so these identifications also commute with arbitrary base change. □
Example 6.7. Let A = K [ z 1 , … , z k ] / L A = K[z_1,\ldots,z_k]/L A = K [ z 1 , … , z k ] / L , where every element of L L L vanishes at the origin. Evaluation at the origin defines an augmentation. Theorem 6.2 applies whether or not the pointed affine scheme Spec A \operatorname{Spec} A Spec A is smooth or reduced. It applies, for example, to K [ u , v ] / ( u v ) K[u,v]/(uv) K [ u , v ] / ( uv ) and to K [ u 1 , … , u k ] / ( u 1 , … , u k ) m K[u_1,\ldots,u_k]/(u_1,\ldots,u_k)^m K [ u 1 , … , u k ] / ( u 1 , … , u k ) m . The theorem also applies over Z [ 1 / n ! ] \mathbb{Z}[1/n!] Z [ 1/ n !] and to every square-zero extension A = K ⊕ M A = K \oplus M A = K ⊕ M , with M 2 = 0 M^2 = 0 M 2 = 0 , for an arbitrary K K K -module M M M . In particular, M M M need not be flat. Over a field of characteristic p > n p > n p > n , the same theorem holds. The characteristic-zero Schur character identities in the following subsections use the polynomial case and additional representation-theoretic input.
Supersymmetric specialization
Return to a field K K K of characteristic zero and take A = K [ z 1 , … , z k ] A = K[z_1,\ldots,z_k] A = K [ z 1 , … , z k ] with augmentation at the origin. Then R A = R n , k R_A = R_{n,k} R A = R n , k . The algebra Q A Q_A Q A has one exterior permutation alphabet in addition to these bosonic alphabets. For a row alphabet q = ( q 1 , … , q k ) \mathbf{q} = (q_1,\ldots,q_k) q = ( q 1 , … , q k ) and one odd variable u u u , use the positive supersymmetric convention
s λ ( q ∣ u ) = ∑ λ / ρ a vertical strip s ρ ( q ) u ∣ λ ∣ − ∣ ρ ∣ . (28) s_\lambda(\mathbf{q}\mid u) = \sum_{\lambda/\rho\ \text{a vertical strip}} s_\rho(\mathbf{q})u^{|\lambda|-|\rho|}. \tag*{(28)} s λ ( q ∣ u ) = λ / ρ a vertical strip ∑ s ρ ( q ) u ∣ λ ∣ − ∣ ρ ∣ . (28)
In particular, [ u d ] s λ ( q ∣ u ) = ( e d ⊥ s λ ) ( q ) [u^d]s_\lambda(\mathbf{q}\mid u) = (e_d^\perp s_\lambda)(\mathbf{q}) [ u d ] s λ ( q ∣ u ) = ( e d ⊥ s λ ) ( q ) . Lentfer’s diagonal supersymmetry theorem [25, Theorem 1.2 and Corollary 1.3} states that the same universal coefficients c λ , μ ( n ) c_{\lambda,\mu}^{(n)} c λ , μ ( n ) describe the full coinvariant characters in arbitrary bosonic and fermionic alphabets. For the sign component with one odd alphabet, it says
ch Hom S n ( sgn , Q A ) = ∑ λ c λ , ( 1 n ) ( n ) s λ ( q ∣ u ) . (29) \operatorname{ch}\operatorname{Hom}_{S_n}(\operatorname{sgn},Q_A) = \sum_\lambda c_{\lambda,(1^n)}^{(n)}s_\lambda(\mathbf{q}\mid u). \tag*{(29)} ch Hom S n ( sgn , Q A ) = λ ∑ c λ , ( 1 n ) ( n ) s λ ( q ∣ u ) . (29)
Theorem 6.8 (All full hook identities). For 0 ≤ d ≤ n − 1 0 \le d \le n-1 0 ≤ d ≤ n − 1 ,
C ( d + 1 , 1 n − d − 1 ) ( n ) = e d ⊥ A n . (30) C_{(d+1,1^{n-d-1})}^{(n)} = e_d^\perp A_n. \tag*{(30)} C ( d + 1 , 1 n − d − 1 ) ( n ) = e d ⊥ A n . (30) Equivalently,
c ρ , ( d + 1 , 1 n − d − 1 ) ( n ) = ∑ λ / ρ a vertical d -strip c λ , ( 1 n ) ( n ) . c_{\rho,(d+1,1^{n-d-1})}^{(n)} = \sum_{\lambda/\rho\ \text{a vertical }d\text{-strip}} c_{\lambda,(1^n)}^{(n)}. c ρ , ( d + 1 , 1 n − d − 1 ) ( n ) = λ / ρ a vertical d -strip ∑ c λ , ( 1 n ) ( n ) . Proof. By Corollary 6.6, the coefficient of u d u^d u d on the left side of (29) is C ( d + 1 , 1 n − d − 1 ) ( n ) ( q ) C_{(d+1,1^{n-d-1})}^{(n)}(\mathbf{q}) C ( d + 1 , 1 n − d − 1 ) ( n ) ( q ) . By (28) , the coefficient on the right side is ( e d ⊥ A n ) ( q ) (e_d^\perp A_n)(\mathbf{q}) ( e d ⊥ A n ) ( q ) . The equality holds for every k k k , and hence holds as an identity of stable symmetric functions. □
The full hook identity follows from augmented sign transfer and diagonal supersymmetry. The two-bosonic-alphabet transfer in [18 ] gives its two-variable specialization; the augmented construction applies in every row dimension. We next construct the corresponding identity inside the operator module.
Remark 6.9 (Order of skewing and specialization). In k k k alphabets the formula is
C ( d + 1 , 1 n − d − 1 ) ( n ) ( q ) = ( e d ⊥ A n ) ( q ) . C_{(d+1,1^{n-d-1})}^{(n)}(\mathbf{q}) = (e_d^\perp A_n)(\mathbf{q}). C ( d + 1 , 1 n − d − 1 ) ( n ) ( q ) = ( e d ⊥ A n ) ( q ) .
It is generally impossible to replace A n A_n A n by the sum of its terms of length at most k k k before skewing. A vertical strip can remove the last box in each of d d d rows, so source partitions of length k + d k+d k + d can contribute to a target of length k k k .
Hook transfer in the polarization module
Invariant differential forms give the operator counterpart. An odd differential trace can be realized by a polarization into a fresh row. Repeating this operation and alternating the fresh rows lifts every hook vector to an alternating vector, which is the required inverse to exteriorization.
We work over a field of characteristic zero. For a finite row space V V V , write P ( V ) = Sym ( V ⊗ N n ) P(V) = \operatorname{Sym}(V \otimes N_n) P ( V ) = Sym ( V ⊗ N n ) , where N n = K n N_n = K^n N n = K n is the permutation representation. We use the polynomial row convention for the variables. All the subspaces below are polynomial row submodules compatible with enlarging the row space, with taking homogeneous row components, and with restriction obtained by setting auxiliary rows to zero. Let M ( V ) \mathcal{M}(V) M ( V ) be a subspace of the diagonal harmonics, stable under constant coefficient partial derivatives and generalized polarizations. Its stable alternating character is denoted by A M A^{\mathcal{M}} A M . For H \mathcal{H} H these compatibilities follow directly from the equations. For P \mathcal{P} P they follow from the commuting normal form and its specialization statement, Corollary 5.4.
For a nonzero multiindex α \alpha α on the old rows and a fresh row z z z , the mixed polarization
E z , α = ∑ i = 1 n z i ∂ i α (31) E_{z,\alpha}=\sum_{i=1}^{n}z_i\partial_i^\alpha \tag*{(31)} E z , α = i = 1 ∑ n z i ∂ i α (31)
is an allowed operation. Indeed, it is a linear combination of conjugates of ∑ i z i ∂ x i ∣ α ∣ \sum_i z_i\partial_{x_i}^{|\alpha|} ∑ i z i ∂ x i ∣ α ∣ under changes of the old rows that fix z z z : pure powers of linear forms span the homogeneous differential symbols of degree ∣ α ∣ |\alpha| ∣ α ∣ .
For 0 ≤ j ≤ n − 1 0\le j\le n-1 0 ≤ j ≤ n − 1 , add j j j auxiliary rows z 1 , … , z j z_1,\ldots,z_j z 1 , … , z j . Let E j M ( V ) \mathcal{E}^{\mathcal{M}}_j(V) E j M ( V ) be the space of elements of M ( V ⊕ K j ) s g n \mathcal{M}(V\oplus K^j)^{\mathrm{sgn}} M ( V ⊕ K j ) sgn that are multilinear in these rows and alternating under their permutations. Such a polynomial has a unique expression
F ( X , z ) = ∑ i 1 , … , i j f i 1 , … , i j ( X ) z 1 , i 1 ⋯ z j , i j , F(X,z)=\sum_{i_1,\ldots,i_j}f_{i_1,\ldots,i_j}(X)z_{1,i_1}\cdots z_{j,i_j}, F ( X , z ) = i 1 , … , i j ∑ f i 1 , … , i j ( X ) z 1 , i 1 ⋯ z j , i j ,
with alternating coefficient tensor. Set
Ψ j ( F ) = ∑ i 1 , … , i j f i 1 , … , i j ( X ) θ i 1 ⋯ θ i j . (32) \Psi_j(F)=\sum_{i_1,\ldots,i_j}f_{i_1,\ldots,i_j}(X)\theta_{i_1}\cdots\theta_{i_j}. \tag*{(32)} Ψ j ( F ) = i 1 , … , i j ∑ f i 1 , … , i j ( X ) θ i 1 ⋯ θ i j . (32)
Here the θ i \theta_i θ i generate Λ N n \Lambda N_n Λ N n . Translation invariance in each auxiliary row puts the coefficient tensor in W n ⊗ j W_n^{\otimes j} W n ⊗ j , where W n = ker ( N n → K ) W_n=\ker(N_n\to K) W n = ker ( N n → K ) ; hence the image is in Λ j W n \Lambda^jW_n Λ j W n . Taking coefficients in the auxiliary rows is a sequence of partial derivatives followed by setting those rows to zero, so every f i 1 , … , i j f_{i_1,\ldots,i_j} f i 1 , … , i j belongs to M ( V ) \mathcal{M}(V) M ( V ) . Consequently
\Psi_j:\mathcal{E}^{\mathcal{M}}_j(V)\lhook\joinrel\longrightarrow\bigl(\mathcal{M}(V)\otimes\Lambda^jW_n\bigr)^{\mathrm{sgn}} \tag*{(33)}
is a GL ( V ) \operatorname{GL}(V) GL ( V ) -equivariant injection. Injectivity follows directly: exteriorization is multiplication by j ! j! j ! on alternating tensors. The case j = 0 j=0 j = 0 is the identity.
Branching the stable alternating character from V ⊕ K j V\oplus K^j V ⊕ K j to GL ( V ) × GL j \operatorname{GL}(V)\times\operatorname{GL}_j GL ( V ) × GL j , and selecting the determinant representation of GL j \operatorname{GL}_j GL j , gives
ch E j M ( V ) = ( e j ⊥ A M ) ( V ) . (34) \operatorname{ch}\mathcal{E}^{\mathcal{M}}_j(V)=\bigl(e_j^{\perp}A^{\mathcal{M}}\bigr)(V). \tag*{(34)} ch E j M ( V ) = ( e j ⊥ A M ) ( V ) . (34)
Indeed, a multilinear alternating function of the auxiliary rows transforms by their determinant, and the relevant Littlewood–Richardson coefficients are c ρ , ( 1 j ) λ c_{\rho,(1^j)}^\lambda c ρ , ( 1 j ) λ .
Theorem 6.10 (Operator hook transfer). Let M \mathcal{M} M be a compatible system of harmonic submodules generated from alternating polynomials by constant coefficient derivatives and generalized polarizations. Then (6.9) is an isomorphism for every finite row space V V V and every 0 ≤ j ≤ n − 1 0\le j\le n-1 0 ≤ j ≤ n − 1 . In particular,
C ( j + 1 , 1 n − j − 1 ) M = e j ⊥ A M . (35) C^{\mathcal{M}}_{(j+1,1^{n-j-1})}=e_j^{\perp}A^{\mathcal{M}}. \tag*{(35)} C ( j + 1 , 1 n − j − 1 ) M = e j ⊥ A M . (35) For M = P \mathcal{M}=\mathcal{P} M = P , this is Bergeron’s Hook-Components Conjecture.
Proof. Put D = K [ ∂ x i ( a ) ] \mathcal{D}=K[\partial_{x_i^{(a)}}] D = K [ ∂ x i ( a ) ] . By Lemma 5.1, the evaluation map
D ⊗ B ( V ) ⟶ M ( V ) , D ⊗ g ⟼ D g , \mathcal{D}\otimes\mathcal{B}(V)\longrightarrow\mathcal{M}(V),\qquad\mathcal{D}\otimes g\longmapsto\mathcal{D}g, D ⊗ B ( V ) ⟶ M ( V ) , D ⊗ g ⟼ D g ,
is surjective, and S n S_n S n acts by its sign on B ( V ) \mathcal{B}(V) B ( V ) . Exactness of averaging therefore shows that every element of the right side of (6.9) is a sum of invariant differential forms in ( D ⊗ Λ j W n ) S n (\mathcal{D}\otimes\Lambda^jW_n)^{S_n} ( D ⊗ Λ j W n ) S n acting on elements of B ( V ) \mathcal{B}(V) B ( V ) .
The invariant algebra ( D ⊗ Λ N n ) S n (\mathcal{D}\otimes\Lambda N_n)^{S_n} ( D ⊗ Λ N n ) S n is generated by the even traces and odd traces
p α = ∑ i ∂ i α ( ∣ α ∣ > 0 ) , q α = ∑ i θ i ∂ i α ( ∣ α ∣ ≥ 0 ) . p_\alpha=\sum_i\partial_i^\alpha\quad(|\alpha|>0),\qquad q_\alpha=\sum_i\theta_i\partial_i^\alpha\quad(|\alpha|\ge0). p α = i ∑ ∂ i α ( ∣ α ∣ > 0 ) , q α = i ∑ θ i ∂ i α ( ∣ α ∣ ≥ 0 ) .
This is Lemma 6.3, applied to the polynomial algebra of differential symbols. Passing to Λ W n \Lambda W_n Λ W n sets q 0 = ∑ i θ i q_0=\sum_i\theta_i q 0 = ∑ i θ i to zero. Averaging is exact, so the images of these generators still generate all invariants in the quotient. The p α p_\alpha p α commute with all the q β q_\beta q β and annihilate the harmonic polynomial on which the form acts. A form of exterior degree j j j consequently reduces to a linear combination of
q α 1 ⋯ q α j g , ∣ α s ∣ > 0 , g ∈ B ( V ) . (36) q_{\alpha_1}\cdots q_{\alpha_j}g,\qquad|\alpha_s|>0,\quad g\in\mathcal{B}(V). \tag*{(36)} q α 1 ⋯ q α j g , ∣ α s ∣ > 0 , g ∈ B ( V ) . (36)
For this word, introduce the j j j fresh rows and form
F ~ = E z 1 , α 1 ⋯ E z j , α j g . \widetilde{F}=E_{z_1,\alpha_1}\cdots E_{z_j,\alpha_j}g. F = E z 1 , α 1 ⋯ E z j , α j g .
All derivatives in these operators use only the old rows. Thus the operators commute, their product is multilinear in the fresh rows, and
F ~ = ∑ i 1 , … , i j z 1 , i 1 ⋯ z j , i j ∂ i 1 α 1 ⋯ ∂ i j α j g . \widetilde{F}=\sum_{i_1,\ldots,i_j}z_{1,i_1}\cdots z_{j,i_j}\partial_{i_1}^{\alpha_1}\cdots\partial_{i_j}^{\alpha_j}g. F = i 1 , … , i j ∑ z 1 , i 1 ⋯ z j , i j ∂ i 1 α 1 ⋯ ∂ i j α j g .
It belongs to M ( V ⊕ K j ) s g n \mathcal{M}(V\oplus K^j)^{\mathrm{sgn}} M ( V ⊕ K j ) sgn by (31) . Let F F F be its normalized alternation under permutations of the fresh rows. Then F ∈ E j M ( V ) F\in\mathcal{E}_j^{\mathcal{M}}(V) F ∈ E j M ( V ) . Exteriorization already alternates these rows, so
Ψ j ( F ) = q α 1 ⋯ q α j g . \Psi_j(F)=q_{\alpha_1}\cdots q_{\alpha_j}g. Ψ j ( F ) = q α 1 ⋯ q α j g .
Every expression (6.12) is therefore in the image of Ψ j \Psi_j Ψ j , proving surjectivity. The character identity follows from (34) and sgn ⊗ Λ n W n ≅ S ( j + 1 , 1 n − j − 1 ) \operatorname{sgn}\otimes\Lambda^n W_n\cong S^{(j+1,1^{n-j-1})} sgn ⊗ Λ n W n ≅ S ( j + 1 , 1 n − j − 1 ) .
The construction also fixes the order of operations in (35) : add the j j j auxiliary rows, select their exterior component, and then specialize the remaining rows to V V V . This is stable skewing followed by finite-row specialization, as in Remark 6.9.
Simultaneous derivative embeddings
Taylor expansion gives simultaneous maps behind the character identities. Derivative closure keeps every Taylor coefficient inside the module. Schur–Weyl duality then separates all Littlewood–Richardson copies, so their contributions form an actual direct sum. This construction applies to any derivative-closed polynomial module.
Theorem 6.11 (The derivative coalgebra and Littlewood–Richardson embeddings). Let V V V , W W W be finite-dimensional vector spaces over a field of characteristic zero, let G ⊆ GL ( W ) G\subseteq\operatorname{GL}(W) G ⊆ GL ( W ) , and let M ⊆ Sym ( V ⊗ W ) \mathcal{M}\subseteq\operatorname{Sym}(V\otimes W) M ⊆ Sym ( V ⊗ W ) be a graded GL ( V ) × G \operatorname{GL}(V)\times G GL ( V ) × G -submodule closed under all constant-coefficient partial derivatives. Put
M λ = Hom GL ( V ) ( S λ V , M ) , M λ = 0 if ℓ ( λ ) > dim V . \mathcal{M}_{\lambda}=\operatorname{Hom}_{\operatorname{GL}(V)}(S_{\lambda}V,\mathcal{M}),\qquad\mathcal{M}_{\lambda}=0\quad\text{if }\ell(\lambda)>\dim V. M λ = Hom GL ( V ) ( S λ V , M ) , M λ = 0 if ℓ ( λ ) > dim V . Then M \mathcal{M} M is a subcoalgebra for the polynomial coproduct Δ F ( X , Y ) = F ( X + Y ) \Delta F(X,Y)=F(X+Y) Δ F ( X , Y ) = F ( X + Y ) . For every pair of partitions τ , ρ \tau,\rho τ , ρ , the coproduct induces a G G G -equivariant injection
\bigoplus_{\lambda}\mathcal{M}_{\lambda}^{\oplus c_{\tau,\rho}^{\lambda}}\lhook\joinrel\longrightarrow\mathcal{M}_{\tau}\otimes\mathcal{M}_{\rho}. \tag*{(37)}
All Littlewood–Richardson multiplicities occur independently. The maps are compatible with inclusions of derivative-closed submodules.
Proof. Taylor expansion and derivative closure give
Δ M ⊆ [ Sym ( V ⊗ W ) ⊗ M ] ∩ [ M ⊗ Sym ( V ⊗ W ) ] = M ⊗ M . \Delta\mathcal{M}\subseteq[\operatorname{Sym}(V\otimes W)\otimes\mathcal{M}]\cap[\mathcal{M}\otimes\operatorname{Sym}(V\otimes W)]=\mathcal{M}\otimes\mathcal{M}. Δ M ⊆ [ Sym ( V ⊗ W ) ⊗ M ] ∩ [ M ⊗ Sym ( V ⊗ W )] = M ⊗ M .
The counit is evaluation at zero. This proves the coalgebra assertion.
Put a = ∣ τ ∣ a=|\tau| a = ∣ τ ∣ and b = ∣ ρ ∣ b=|\rho| b = ∣ ρ ∣ . Project the bidegree-( a , b ) (a,b) ( a , b ) coproduct to the row types τ \tau τ and ρ \rho ρ in its two factors. The ambient Cauchy decomposition has column multiplicity S λ W S_{\lambda}W S λ W in row type λ \lambda λ . Schur–Weyl duality constructs this summand by the invariant pairing between two copies of the Specht module [ λ ] [\lambda] [ λ ] . Write L τ , ρ λ L_{\tau,\rho}^{\lambda} L τ , ρ λ for the multiplicity space of [ τ ] ⊠ [ ρ ] [\tau]\mathbin{\boxtimes}[\rho] [ τ ] ⊠ [ ρ ] in its restriction to S a × S b S_a\times S_b S a × S b ; its dimension is c τ , ρ λ c_{\tau,\rho}^{\lambda} c τ , ρ λ . The bidegree coproduct uses the restricted Specht pairing. Over the reals, choose this pairing positive and invariant, so its restriction to each isotypic component is nondegenerate. On the two copies of L τ , ρ λ L_{\tau,\rho}^{\lambda} L τ , ρ λ it is therefore a perfect pairing. In dual bases its tensor is the identity: each row coefficient selects a distinct column coefficient. We obtain exactly c τ , ρ λ c^{\lambda}_{\tau,\rho} c τ , ρ λ independent ambient maps
S λ W ⟶ S τ W ⊗ S ρ W . S_{\lambda}W \longrightarrow S_{\tau}W \otimes S_{\rho}W. S λ W ⟶ S τ W ⊗ S ρ W .
For different λ \lambda λ their images lie in distinct isotypic summands; for a fixed λ \lambda λ they fill its independent Littlewood–Richardson copies. Their direct sum is injective. All relevant pairings and projectors can be chosen over Q \mathbb{Q} Q , so the assertion holds over every characteristic-zero field.
The coalgebra inclusion places the projected coproduct inside the row spaces tensored with M τ ⊗ M ρ M_{\tau}\otimes M_{\rho} M τ ⊗ M ρ . Taking its independent row coefficients therefore restricts the ambient maps to (6.13). The column group is used only on the ambient polynomial ring; M \mathcal{M} M need only be G G G -stable. Fixed choices of ambient maps give compatibility with inclusions. □
Corollary 6.12 (Exterior derivative embeddings). Under the hypotheses of Theorem 6.11, for every partition ρ \rho ρ and j ≥ 0 j \ge0 j ≥ 0 there is a G G G -equivariant injection
\bigoplus_{\lambda/\rho\ \mathrm{a\ vertical}\ j\text{-strip}} M_{\lambda} \lhook\joinrel\longrightarrow\Lambda^{j}W \otimes M_{\rho}. \tag*{(38)}
The maps arise from the exterior row component of the j j j th derivatives and are compatible with inclusions.
Proof. Take τ = ( 1 j ) \tau=(1^{j}) τ = ( 1 j ) in (6.13) and use the vertical Pieri rule and M ( 1 j ) ⊆ Λ j W M_{(1^{j})}\subseteq\Lambda^{j}W M ( 1 j ) ⊆ Λ j W . □
Corollary 6.13 (Derivative realization and exact hook defects). For M = H \mathcal{M}=\mathcal{H} M = H or P \mathcal{P} P , let M ρ M M^{\mathcal{M}}_{\rho} M ρ M denote its stable row multiplicity space and put A λ M = Hom S n ( sgn , M λ M ) A^{\mathcal{M}}_{\lambda}=\operatorname{Hom}_{S_n}(\operatorname{sgn},M^{\mathcal{M}}_{\lambda}) A λ M = Hom S n ( sgn , M λ M ) . The exterior derivative maps give compatible isomorphisms
⨁ λ / ρ a v e r t i c a l d -strip A λ M → ∼ Hom S n ( S ( d + 1 , 1 n − d − 1 ) , M ρ M ) , 0 ≤ d < n , (39) \bigoplus_{\lambda/\rho\ \mathrm{a\ vertical}\ d\text{-strip}} A^{\mathcal{M}}_{\lambda}\xrightarrow{\sim}\operatorname{Hom}_{S_n}\left(S^{(d+1,1^{n-d-1})},M^{\mathcal{M}}_{\rho}\right),\qquad0\le d<n, \tag*{(39)} λ / ρ a vertical d -strip ⨁ A λ M ∼ Hom S n ( S ( d + 1 , 1 n − d − 1 ) , M ρ M ) , 0 ≤ d < n , (39) and induce isomorphisms on the quotients for P ⊆ H \mathcal{P}\subseteq\mathcal{H} P ⊆ H . Thus, writing D A = A n H − A n P D_A=A^{\mathcal{H}}_n-A^{\mathcal{P}}_n D A = A n H − A n P and D d = C ( d + 1 , 1 n − d − 1 ) H − C ( d + 1 , 1 n − d − 1 ) P D_d=C^{\mathcal{H}}_{(d+1,1^{n-d-1})}-C^{\mathcal{P}}_{(d+1,1^{n-d-1})} D d = C ( d + 1 , 1 n − d − 1 ) H − C ( d + 1 , 1 n − d − 1 ) P ,
C ( d + 1 , 1 n − d − 1 ) M = e d ⊥ A n M , (40) C^{\mathcal{M}}_{(d+1,1^{n-d-1})}=e_d^{\perp}A^{\mathcal{M}}_n, \tag*{(40)} C ( d + 1 , 1 n − d − 1 ) M = e d ⊥ A n M , (40) D d = e d ⊥ D A . (41) D_d=e_d^{\perp}D_A. \tag*{(41)} D d = e d ⊥ D A . (41) Restricting source and target to k k k rows gives
c ρ , ( d + 1 , 1 n − d − 1 ) M ≥ ∑ λ / ρ v e r t i c a l d -strip ℓ ( λ ) ≤ k c λ , ( 1 n ) M . c^{\mathcal{M}}_{\rho,(d+1,1^{n-d-1})}\ge\sum_{\substack{\lambda/\rho\ \mathrm{vertical}\ d\text{-strip}\\ \ell(\lambda)\le k}}c^{\mathcal{M}}_{\lambda,(1^n)}. c ρ , ( d + 1 , 1 n − d − 1 ) M ≥ λ / ρ vertical d -strip ℓ ( λ ) ≤ k ∑ c λ , ( 1 n ) M . Proof. Apply Corollary 6.12 with W = W n W=W_n W = W n and take Hom S n ( sgn , − ) . \operatorname{Hom}_{S_n}(\operatorname{sgn},-). Hom S n ( sgn , − ) . Since sgn ⊗ Λ d W n ≅ S ( d + 1 , 1 n − d − 1 ) \operatorname{sgn}\otimes\Lambda^dW_n\cong S^{(d+1,1^{n-d-1})} sgn ⊗ Λ d W n ≅ S ( d + 1 , 1 n − d − 1 ) , the target is the indicated hook multiplicity. Theorems 6.8 and 6.10 give equality of dimensions, so these compatible injections are isomorphisms. They induce the quotient isomorphisms and both character identities. Retaining only source partitions of length at most k k k gives the finite-row bound. □
Thus full alternating generation is equivalent to simultaneous generation of every hook label. The derivative maps retain the individual vertical-strip sources, and their quotients identify every remaining hook defect.
Two-column support and the alternating defect
After removing the linear moments, the labels form the standard representation W n W_n W n with its invariant inner product. The quadratic moments are its pairwise Gram products. Two alternating blocks whose total height exceeds dim W n \dim W_n dim W n vanish modulo those products. Applied to the first two columns of a row partition, this gives the support restriction used in the uniform generation proof.
Lemma 6.14 (Two alternating blocks in the Gram ideal). Let E E E be a vector space of dimension N N N over a field of characteristic zero, equipped with a nondegenerate symmetric bilinear form. A multilinear polynomial on E d + e E^{d+e} E d + e that is alternating in the first d d d arguments and in the last e e e arguments belongs to the ideal generated by the inner products between the two blocks if d + e > N d+e>N d + e > N .
Proof. Identify E E E with E ∗ E^* E ∗ by the form. On the tensor product of exterior algebras, using ordinary tensor signs, put
L = ∑ i ε i ⊗ ε i , c = ∑ i ι i ⊗ ι i . L=\sum_i\varepsilon_i\otimes\varepsilon_i,\qquad c=\sum_i\iota_i\otimes\iota_i. L = i ∑ ε i ⊗ ε i , c = i ∑ ι i ⊗ ι i . Invariantly the form and its inverse define these operators over K K K . They satisfy [ c , L ] = ( N − d − e ) id [c,L]=(N-d-e)\operatorname{id} [ c , L ] = ( N − d − e ) id in bidegree ( d , e ) (d,e) ( d , e ) . Finite-dimensional s l 2 \mathfrak{sl}_2 sl 2 theory makes
L : Λ d − 1 E ⊗ Λ e − 1 E ⟶ Λ d E ⊗ Λ e E L:\Lambda^{d-1}E\otimes\Lambda^{e-1}E\longrightarrow\Lambda^dE\otimes\Lambda^eE L : Λ d − 1 E ⊗ Λ e − 1 E ⟶ Λ d E ⊗ Λ e E surjective when d + e > N d+e>N d + e > N . Under the multilinear polynomial interpretation, its image consists of alternating sums of a cross-block inner product multiplied by a polynomial in the remaining arguments. This proves membership in the stated ideal. An explicit finite right inverse is recorded in (B.7). □
The next inequality is the admissibility condition for orthogonal traceless tensors; see, for example, [9 ] . The Gram-ideal argument proves it directly at the dimension n − 1 n-1 n − 1 relevant here.
Theorem 6.15 (Two-column support). For every coefficient of full diagonal harmonics,
c λ , μ ( n ) ≠ 0 ⟹ λ 1 ′ + λ 2 ′ ≤ n − 1. (42) c_{\lambda,\mu}^{(n)}\ne0\quad\Longrightarrow\quad\lambda'_1+\lambda'_2\le n-1. \tag*{(42)} c λ , μ ( n ) = 0 ⟹ λ 1 ′ + λ 2 ′ ≤ n − 1. (42) Here missing columns have length zero. The same restriction holds in P \mathcal{P} P .
Proof. Quotient by the linear invariants first. This identifies the polynomial ring with Sym ( V ⊗ W n ) \operatorname{Sym}(V\otimes W_n) Sym ( V ⊗ W n ) , where V = K k V=K^k V = K k and dim W n = n − 1 \dim W_n=n-1 dim W n = n − 1 . The quadratic invariants include all pairwise inner products of the row vectors for the nondegenerate form on W n W_n W n . Thus the full coinvariant ring is a quotient of the quotient by the Gram ideal.
The usual minor construction of the Cauchy decomposition spans a row-λ \lambda λ highest-weight space by products of minors with column heights λ 1 ′ , λ 2 ′ , … \lambda'_1,\lambda'_2,\ldots λ 1 ′ , λ 2 ′ , … . If the sum of the first two heights exceeds n − 1 n-1 n − 1 , regard the vectors in these two minors temporarily as independent. Their product is multilinear and alternating separately in the two blocks. Lemma 6.14 puts this product in the Gram ideal. Identifying the repeated row vectors and multiplying by the remaining minors keeps it in that ideal. If only one column is present, a height greater than n − 1 n-1 n − 1 is already excluded by the Cauchy decomposition. Every such highest-weight vector vanishes in the quotient, proving (6.18). The assertion for P \mathcal{P} P follows by inclusion. □
Corollary 6.16 (Support of the alternating defect). If the Schur coefficient of s λ s_\lambda s λ in D A D_A D A is nonzero, then
λ 2 ≥ 3 , λ 3 ≥ 2 , λ 1 ′ + λ 2 ′ ≤ n − 1. (43) \lambda_2\ge3,\qquad\lambda_3\ge2,\qquad\lambda'_1+\lambda'_2\le n-1. \tag*{(43)} λ 2 ≥ 3 , λ 3 ≥ 2 , λ 1 ′ + λ 2 ′ ≤ n − 1. (43) In particular, P \mathcal{P} P contains every stable alternating component and every stable hook-label component when n ≤ 6 n\le6 n ≤ 6 .
Proof. Haiman’s two-row operator theorem gives D j ( q 1 , q 2 ) = 0 D_j(q_1,q_2)=0 D j ( q 1 , q 2 ) = 0 for every j j j . If λ 3 ≤ 1 \lambda_3\le1 λ 3 ≤ 1 , delete one box from each nonempty row after the second; the remaining partition has at most two rows, and the removed boxes form a vertical strip. Schur positivity of D A D_A D A and the vertical Pieri rule would then give a nonzero coefficient in some e j ⊥ D A ( q 1 , q 2 ) e_j^\perp D_A(q_1,q_2) e j ⊥ D A ( q 1 , q 2 ) , a contradiction.
Every row-hook component of every label is generated, by Proposition 4.3. If λ 2 ≤ 2 \lambda_2\le2 λ 2 ≤ 2 , remove the last box from every row of length two below the first. The remaining partition is a row hook. Again positivity and (6.17) contradict generation of that row-hook component. The strip sizes used here are at most n − 1 n-1 n − 1 , since the row-length bound applies to the alternating label.
The final inequality is the two-column support theorem. The first two inequalities force the first two columns of λ \lambda λ to have at least three boxes each. Thus n ≥ 7 n \ge7 n ≥ 7 for a nonzero defect. (41) gives the final assertion. □
Finite-alphabet reconstruction
The hook identities also recover stable alternating information beyond the number of rows in which the hook characters are known.
Theorem 6.17 (Reconstruction from finitely many alphabets). Let M = H \mathcal{M}=\mathcal{H} M = H or P \mathcal{P} P , let n ≥ 3 n \ge3 n ≥ 3 , and put k = ⌊ ( n − 1 ) / 2 ⌋ k=\lfloor(n-1)/2\rfloor k = ⌊( n − 1 ) /2 ⌋ . The collection
{ C ( d + 1 , 1 n − d − 1 ) M ( q 1 , … , q k ) : 0 ≤ d ≤ n − 1 } (44) \left\{C^{\mathcal{M}}_{(d+1,1^{n-d-1})}(q_1,\ldots,q_k): 0 \le d \le n-1\right\} \tag*{(44)} { C ( d + 1 , 1 n − d − 1 ) M ( q 1 , … , q k ) : 0 ≤ d ≤ n − 1 } (44) determines every Schur coefficient of the stable alternating character A n M A_n^{\mathcal{M}} A n M .
Proof. Suppress the superscript M \mathcal{M} M , write A n = ∑ λ a λ s λ A_n=\sum_{\lambda}a_\lambda s_\lambda A n = ∑ λ a λ s λ , and form
F ( q , u ) = ∑ d = 0 n − 1 C ( d + 1 , 1 n − d − 1 ) ( q ) u d . F(\mathbf{q},u)=\sum_{d=0}^{n-1}C_{(d+1,1^{n-d-1})}(\mathbf{q})u^d. F ( q , u ) = d = 0 ∑ n − 1 C ( d + 1 , 1 n − d − 1 ) ( q ) u d . The hook identity gives
F ( q , u ) = ∑ λ a λ s λ ( q ∣ u ) . (45) F(\mathbf{q},u)=\sum_{\lambda}a_\lambda s_\lambda(\mathbf{q}\mid u). \tag*{(45)} F ( q , u ) = λ ∑ a λ s λ ( q ∣ u ) . (45) If λ k + 1 ≥ 2 \lambda_{k+1}\ge2 λ k + 1 ≥ 2 , then λ 1 ′ + λ 2 ′ ≥ 2 k + 2 > n − 1 \lambda'_1+\lambda'_2\ge2k+2>n-1 λ 1 ′ + λ 2 ′ ≥ 2 k + 2 > n − 1 . Theorem 6.15 excludes such a partition. Every supported partition therefore satisfies λ k + 1 ≤ 1 \lambda_{k+1}\le1 λ k + 1 ≤ 1 .
At u = 0 u=0 u = 0 , equation (6.21) determines all a λ a_\lambda a λ with ℓ ( λ ) ≤ k \ell(\lambda)\le k ℓ ( λ ) ≤ k , because their Schur functions in k k k variables are linearly independent. Subtract their entire supersymmetric contributions from F F F . Every remaining partition has the form
λ = ( λ 1 , … , λ k , 1 r ) , r > 0 , λ k ≥ 1. \lambda=(\lambda_1,\ldots,\lambda_k,1^r),\qquad r>0,\qquad\lambda_k\ge1. λ = ( λ 1 , … , λ k , 1 r ) , r > 0 , λ k ≥ 1. The standard factorization of a supersymmetric Schur function on the ( k , 1 ) (k,1) ( k , 1 ) hook gives
s λ ( q ∣ u ) = ∏ i = 1 k ( q i + u ) s ( λ 1 − 1 , … , λ k − 1 ) ( q ) u r . (46) s_\lambda(\mathbf{q}\mid u)=\prod_{i=1}^{k}(q_i+u)s_{(\lambda_1-1,\ldots,\lambda_k-1)}(\mathbf{q})u^r. \tag*{(46)} s λ ( q ∣ u ) = i = 1 ∏ k ( q i + u ) s ( λ 1 − 1 , … , λ k − 1 ) ( q ) u r . (46) It also follows directly from the vertical-strip expansion (6.4) and the Pieri rule. Divide the residual polynomial by ∏ i ( q i + u ) \prod_i(q_i+u) ∏ i ( q i + u ) . The polynomials s ν ( q ) u r s_\nu(\mathbf{q})u^r s ν ( q ) u r are linearly independent, and the pair ( ν , r ) (\nu,r) ( ν , r ) recovers λ = ( ν 1 + 1 , … , ν k + 1 , 1 r ) \lambda=(\nu_1+1,\ldots,\nu_k+1,1^r) λ = ( ν 1 + 1 , … , ν k + 1 , 1 r ) . This determines all remaining coefficients. □
Bergeron’s reconstruction from hook information [4 Section 3.1] is formulated under the corresponding structural conjectures. Theorem 6.17 establishes it for both the full and operator characters, with the number of rows fixed by the two-column bound. For seven labels, three-row hook characters determine the entire stable alternating character.
The sign component of the Theta conjecture at t = 0 t=0 t = 0 We prove the sign-isotypic component of the t = 0 t=0 t = 0 specialization of the Theta coinvariant conjecture of D’Adderio, Iraci and Vanden Wyngaerd [14 Conjecture 8.2] , in every rank. In particular, the alternating coinvariants with one bosonic and two fermionic alphabets have dimension 3 n − 1 3^{n-1} 3 n − 1 , proving the numerical prediction in Bergeron’s tbl:3 [5 ] . Iraci, Nadeau and Vanden Wyngaerd [20 Theorem 5.6] computed the symmetric-function prediction; the argument here identifies that prediction with the actual coinvariant sign component. Its ingredients are the full hook identity, diagonal supersymmetry, and the Fields character theorem.
Adjoining a fermionic alphabet
For k , j ≥ 0 k,j \ge0 k , j ≥ 0 , let
P n ( k , j ) = K [ x i ( a ) : 1 ≤ i ≤ n , 1 ≤ a ≤ k ] ⊗ Λ ( θ i ( b ) : 1 ≤ i ≤ n , 1 ≤ b ≤ j ) , P_n^{(k,j)} = K[x_i^{(a)} : 1 \le i \le n,\ 1 \le a \le k] \otimes\Lambda(\theta_i^{(b)} : 1 \le i \le n,\ 1 \le b \le j), P n ( k , j ) = K [ x i ( a ) : 1 ≤ i ≤ n , 1 ≤ a ≤ k ] ⊗ Λ ( θ i ( b ) : 1 ≤ i ≤ n , 1 ≤ b ≤ j ) ,
R n ( k , j ) = P n ( k , j ) / ⟨ ( P n ( k , j ) ) + S n ⟩ . R_n^{(k,j)} = P_n^{(k,j)}/\langle(P_n^{(k,j)})_+^{S_n}\rangle. R n ( k , j ) = P n ( k , j ) / ⟨( P n ( k , j ) ) + S n ⟩ .
Here K K K has characteristic zero, all odd generators anticommute, and S n S_n S n acts by simultaneously permuting their label indices. The subscript + + + denotes zero constant term. We record each bosonic and fermionic alphabet separately, by Q = ( q 1 , … , q k ) Q=(q_1,\ldots,q_k) Q = ( q 1 , … , q k ) and U = ( u 1 , … , u j ) U=(u_1,\ldots,u_j) U = ( u 1 , … , u j ) , respectively. Supersymmetric evaluation f [ Q ∣ U ] f[Q\mid U] f [ Q ∣ U ] uses the positive convention of (6.4); in particular, odd degrees have positive coefficients in the Hilbert series.
Write W n = ker ( K n ⟶ K ) W_n=\ker(K^n \longrightarrow K) W n = ker ( K n ⟶ K ) . If η 1 , … , η n \eta_1,\ldots,\eta_n η 1 , … , η n is a new fermionic alphabet, its invariant ∑ i η i \sum_i\eta_i ∑ i η i is zero in R n ( k , j + 1 ) R_n^{(k,j+1)} R n ( k , j + 1 ) . The quotient K n / K 1 K^n/K1 K n / K 1 identifies canonically with W n W_n W n by z ↦ z − n − 1 ( ∑ i z i ) 1 z\mapsto z-n^{-1}(\sum_i z_i)1 z ↦ z − n − 1 ( ∑ i z i ) 1 . Consequently the quotient map factors through R n ( k , j ) ⊗ Λ W n R_n^{(k,j)}\otimes\Lambda W_n R n ( k , j ) ⊗ Λ W n ; the tensor product uses the Koszul sign rule with the old odd variables.
Theorem 7.1 (Natural sign transfer in superspace). For all n ≥ 1 n \ge1 n ≥ 1 and k , j ≥ 0 k,j \ge0 k , j ≥ 0 , the natural quotient induces an isomorphism
( R n ( k , j ) ⊗ Λ W n ) s g n → ∼ ( R n ( k , j + 1 ) ) s g n . (47) \left(R_n^{(k,j)}\otimes\Lambda W_n\right)^{\mathrm{sgn}} \xrightarrow{\sim} \left(R_n^{(k,j+1)}\right)^{\mathrm{sgn}}. \tag*{(47)} ( R n ( k , j ) ⊗ Λ W n ) sgn ∼ ( R n ( k , j + 1 ) ) sgn . (47) It preserves every old grading and the degree in the new fermionic alphabet, and commutes with parity-preserving changes of the old row alphabets. In new fermionic degree 0 ≤ d ≤ n − 1 0 \le d \le n-1 0 ≤ d ≤ n − 1 , the induced isomorphism of multiplicity spaces is
Hom S n ( S ( d + 1 , 1 n − d − 1 ) , R n ( k , j ) ) ≅ Hom S n ( s g n , ( R n ( k , j + 1 ) ) deg η = d ) . (48) \operatorname{Hom}_{S_n}\left(S^{(d+1,1^{n-d-1})},R_n^{(k,j)}\right)\cong\operatorname{Hom}_{S_n}\left(\mathrm{sgn},\left(R_n^{(k,j+1)}\right)_{\deg_\eta=d}\right). \tag*{(48)} Hom S n ( S ( d + 1 , 1 n − d − 1 ) , R n ( k , j ) ) ≅ Hom S n ( sgn , ( R n ( k , j + 1 ) ) d e g η = d ) . (48) Both sides of (7.1) vanish in new fermionic degree greater than n − 1 n-1 n − 1 .
The case ( k , j ) = ( 2 , 0 ) (k,j)=(2,0) ( k , j ) = ( 2 , 0 ) is the sign-transfer theorem of González, Lentfer and Mularczyk [18 Theorem 4.21] . Here the full hook identity permits arbitrary old bosonic and fermionic alphabets.
Proof. The quotient is surjective on sign components, since the sign idempotent is exact in characteristic zero. By Lentfer’s diagonal supersymmetry theorem [25 Theorem 1.2 and Corollary 1.3] , the character of the S μ S^\mu S μ multiplicity space of R n ( k , j ) R_n^{(k,j)} R n ( k , j ) is C μ ( n ) [ Q ∣ U ] C_\mu^{(n)}[Q\mid U] C μ ( n ) [ Q ∣ U ] . Theorem 6.8 gives
C ( d + 1 , 1 n − d − 1 ) ( n ) = e d ⊥ A n ( 0 ≤ d ≤ n − 1 ) . C_{(d+1,1^{n-d-1})}^{(n)}=e_d^\perp A_n\qquad(0\le d\le n-1). C ( d + 1 , 1 n − d − 1 ) ( n ) = e d ⊥ A n ( 0 ≤ d ≤ n − 1 ) .
Super Schur branching in one additional odd variable v v v yields
A n [ Q ∣ U , v ] = ∑ d ≥ 0 v d ( e d ⊥ A n ) [ Q ∣ U ] . (49) A_n[Q\mid U,v]=\sum_{d\ge0}v^d(e_d^\perp A_n)[Q\mid U]. \tag*{(49)} A n [ Q ∣ U , v ] = d ≥ 0 ∑ v d ( e d ⊥ A n ) [ Q ∣ U ] . (49)
The skew operator acts on the stable symmetric function before the finite supersymmetric specialization is made.
The invariant dot product on W n W_n W n identifies its exterior powers with their duals, and s g n ⊗ Λ d W n ≅ S ( d + 1 , 1 n − d − 1 ) \mathrm{sgn}\otimes\Lambda^dW_n\cong S^{(d+1,1^{n-d-1})} sgn ⊗ Λ d W n ≅ S ( d + 1 , 1 n − d − 1 ) . Tensor–Hom adjunction therefore identifies the coefficient of v d v^d v d in the source sign multiplicity with C ( d + 1 , 1 n − d − 1 ) ( n ) [ Q ∣ U ] C_{(d+1,1^{n-d-1})}^{(n)}[Q\mid U] C ( d + 1 , 1 n − d − 1 ) ( n ) [ Q ∣ U ] . This equals the coefficient in (7.3), which is the target sign multiplicity. Every multigraded piece is finite-dimensional, so the surjection is an isomorphism in every multidegree, proving both displayed isomorphisms. The source has no exterior degree above n − 1 n-1 n − 1 ; the target also has none, because the relation ∑ i η i = 0 \sum_i\eta_i=0 ∑ i η i = 0 reduces its new exterior alphabet to W n W_n W n . Naturality follows from the quotient map itself. □
We use the usual grading convention when passing to harmonic models: a harmonic polynomial and its coinvariant class have the same multidegree. More explicitly, fix an order on the odd generators and use the apolar pairing
⟨ x α θ I , x β θ J ⟩ = α ! δ α , β δ I , J . \langle x^\alpha\theta_I,x^\beta\theta_J\rangle=\alpha!\,\delta_{\alpha,\beta}\delta_{I,J}. ⟨ x α θ I , x β θ J ⟩ = α ! δ α , β δ I , J .
The odd differential operators are taken in the reverse order to the corresponding monomial. The harmonic inverse system is the orthogonal complement of the invariant ideal for this pairing. The pairing is positive definite over the reals and is defined over Q \mathbb{Q} Q , so the quotient projection identifies the harmonic model and the coinvariant module in each multidegree over every characteristic-zero field. Thus no inversion of q q q , u u u , or v v v occurs when using the Fields formula (13) . On harmonic inverse systems, (47) also gives
( H n ( k , j + 1 ) ) s g n = ( H n ( k , j ) ⊗ Λ W n ) s g n . \left(\mathcal{H}^{(k,j+1)}_n\right)^{\mathrm{sgn}}=\left(\mathcal{H}^{(k,j)}_n\otimes\Lambda W_n\right)^{\mathrm{sgn}}. ( H n ( k , j + 1 ) ) sgn = ( H n ( k , j ) ⊗ Λ W n ) sgn .
the final harmonic space is contained in the intermediate one, and the quotient isomorphism makes their sign components equal.
The actual alternating character
Put [ m ] q ! = ∏ r = 1 m ( 1 + q + ⋯ + q r − 1 ) [m]_q! = \prod_{r=1}^{m}(1+q+\cdots+q^{r-1}) [ m ] q ! = ∏ r = 1 m ( 1 + q + ⋯ + q r − 1 ) , with [ 0 ] q ! = 1 [0]_q! = 1 [ 0 ] q ! = 1 , and use the Gaussian coefficients
[ N a , b , c ] q = [ N ] q ! [ a ] q ! [ b ] q ! [ c ] q ! ( a + b + c = N ) , [ N a ] q = [ N ] q ! [ a ] q ! [ N − a ] q ! . \begin{bmatrix} N \\ a,b,c \end{bmatrix}_q=\frac{[N]_q!}{[a]_q![b]_q![c]_q!}\quad(a+b+c=N),\qquad
\begin{bmatrix} N \\ a \end{bmatrix}_q=\frac{[N]_q!}{[a]_q![N-a]_q!}. [ N a , b , c ] q = [ a ] q ! [ b ] q ! [ c ] q ! [ N ] q ! ( a + b + c = N ) , [ N a ] q = [ a ] q ! [ N − a ] q ! [ N ] q ! .
Theorem 7.2 (The t = 0 t=0 t = 0 Theta sign component). For every n ≥ 1 n\ge1 n ≥ 1 ,
Hilb ( R n ( 1 , 2 ) s g n ; q ; u , v ) = ∑ a , b ≥ 0 a + b ≤ n − 1 u a v b q ( n − a − b 2 ) [ n − 1 a , b , n − 1 − a − b ] q . (50) \operatorname{Hilb}\left(R_n^{(1,2)\mathrm{sgn}};q;u,v\right)=\sum_{\substack{a,b\geq0\\a+b\leq n-1}}u^av^bq^{\binom{n-a-b}{2}}\begin{bmatrix}n-1\\a,b,n-1-a-b\end{bmatrix}_q. \tag*{(50)} Hilb ( R n ( 1 , 2 ) sgn ; q ; u , v ) = a , b ≥ 0 a + b ≤ n − 1 ∑ u a v b q ( 2 n − a − b ) [ n − 1 a , b , n − 1 − a − b ] q . (50)
Here q q q records the bosonic degree, and u , v u,v u , v record the two fermionic degrees. This is the sign coefficient of [14 Conjecture 8.2] at t = 0 t=0 t = 0 , in every fermionic bidegree. In particular,
Hilb ( R n ( 1 , 2 ) s g n ; 1 ; u , v ) = ( 1 + u + v ) n − 1 , dim ( R n ( 1 , 2 ) s g n ) = 3 n − 1 . (51) \operatorname{Hilb}\left(R_n^{(1,2)\mathrm{sgn}};1;u,v\right)=(1+u+v)^{n-1},\qquad\dim\left(R_n^{(1,2)\mathrm{sgn}}\right)=3^{n-1}. \tag*{(51)} Hilb ( R n ( 1 , 2 ) sgn ; 1 ; u , v ) = ( 1 + u + v ) n − 1 , dim ( R n ( 1 , 2 ) sgn ) = 3 n − 1 . (51)
Proof. Fix 0 ≤ b ≤ n − 1 0\leq b\leq n-1 0 ≤ b ≤ n − 1 and set N = n − 1 N=n-1 N = n − 1 , d = N − b d=N-b d = N − b . Theorem 7.1 identifies the coefficient of v b v^b v b with the multiplicity of the hook μ = ( b + 1 , 1 d ) \mu=(b+1,1^d) μ = ( b + 1 , 1 d ) in R n ( 1 , 1 ) R_n^{(1,1)} R n ( 1 , 1 ) . The Fields character theorem of Murai, Rhoades and Wilson [27 Theorem 5.2 and equation (2.5)] , in the form (13) , evaluates this multiplicity as
∑ T ∈ SYT ( μ ) q maj ( T ) ∏ r = 1 des ( T ) ( 1 + u q − r ) . \sum_{T\in\operatorname{SYT}(\mu)}q^{\operatorname{maj}(T)}\prod_{r=1}^{\operatorname{des}(T)}(1+uq^{-r}). T ∈ SYT ( μ ) ∑ q maj ( T ) r = 1 ∏ des ( T ) ( 1 + u q − r ) .
The hook-tableau enumeration used in (16) gives des ( T ) = d \operatorname{des}(T)=d des ( T ) = d and ∑ T q maj ( T ) = q ( d + 1 2 ) [ N d ] q \sum_Tq^{\operatorname{maj}(T)}=q^{\binom{d+1}{2}}\begin{bmatrix}N\\d\end{bmatrix}_q ∑ T q maj ( T ) = q ( 2 d + 1 ) [ N d ] q . It follows that
[ v b ] Hilb ( R n ( 1 , 2 ) s g n ; q ; u , v ) = q ( d + 1 2 ) [ N b ] q ∏ r = 1 d ( 1 + u q − r ) . (52) [v^b]\operatorname{Hilb}\left(R_n^{(1,2)\mathrm{sgn}};q;u,v\right)=q^{\binom{d+1}{2}}\begin{bmatrix}N\\b\end{bmatrix}_q\prod_{r=1}^{d}(1+uq^{-r}). \tag*{(52)} [ v b ] Hilb ( R n ( 1 , 2 ) sgn ; q ; u , v ) = q ( 2 d + 1 ) [ N b ] q r = 1 ∏ d ( 1 + u q − r ) . (52)
For 0 ≤ a ≤ d 0\leq a\leq d 0 ≤ a ≤ d , the Gaussian binomial theorem gives
[ u a ] ∏ r = 1 d ( 1 + u q − r ) = ∑ J ⊆ [ d ] ∣ J ∣ = a q − ∑ r ∈ J r = q − a d + ( a + 1 2 ) [ d a ] q . [u^a]\prod_{r=1}^{d}(1+uq^{-r})=\sum_{\substack{J\subseteq[d]\\|J|=a}}q^{-\sum_{r\in J}r}=q^{-ad+\binom{a+1}{2}}\begin{bmatrix}d\\a\end{bmatrix}_q. [ u a ] r = 1 ∏ d ( 1 + u q − r ) = J ⊆ [ d ] ∣ J ∣ = a ∑ q − ∑ r ∈ J r = q − a d + ( 2 a + 1 ) [ d a ] q .
For a > d a>d a > d the coefficient is zero. Since
( d + 1 2 ) − a d + ( a 2 ) = ( d − a + 1 2 ) = ( n − a − b 2 ) , [ N b ] q [ N − b a ] q = [ N a , b , N − a − b ] q , \binom{d+1}{2}-ad+\binom{a}{2}=\binom{d-a+1}{2}=\binom{n-a-b}{2},\qquad
\begin{bmatrix}N\\b\end{bmatrix}_q\begin{bmatrix}N-b\\a\end{bmatrix}_q=\begin{bmatrix}N\\a,b,N-a-b\end{bmatrix}_q, ( 2 d + 1 ) − a d + ( 2 a ) = ( 2 d − a + 1 ) = ( 2 n − a − b ) , [ N b ] q [ N − b a ] q = [ N a , b , N − a − b ] q ,
extracting u a u^a u a from (52) proves (50) .
For comparison with the stated conjecture, let Θ f \Theta_f Θ f and ∇ \nabla ∇ denote the symmetric-function operators of [14 ] . The sign coefficient predicted there in fermionic bidegree ( a , b ) (a,b) ( a , b ) , after setting t = 0 t=0 t = 0 , is
⟨ Θ e a Θ e b ∇ e n − a − b | t = 0 , e n ⟩ = q ( n − a − b 2 ) [ n − 1 a + b ] q [ a + b a ] q ( a + b < n ) . \left\langle\Theta_{e_a}\Theta_{e_b}\nabla e_{n-a-b}\middle|_{t=0},e_n\right\rangle=q^{\binom{n-a-b}{2}}\begin{bmatrix}n-1\\a+b\end{bmatrix}_{q}\begin{bmatrix}a+b\\a\end{bmatrix}_{q}\qquad(a+b<n). ⟨ Θ e a Θ e b ∇ e n − a − b t = 0 , e n ⟩ = q ( 2 n − a − b ) [ n − 1 a + b ] q [ a + b a ] q ( a + b < n ) .
by [20 Theorem 5.6] ; see also [24 Theorem 8.3] . The two Gaussian binomials multiply to the multinomial in (50) . Both series have zero sign coefficient for a + b ≥ n a+b\ge n a + b ≥ n . Finally, putting q = 1 q=1 q = 1 in that equation and applying the trinomial theorem proves (51) .
An explicit general linear action and Lefschetz maps
Let μ = ( c , τ ) ⊢ n \mu=(c,\tau)\vdash n μ = ( c , τ ) ⊢ n , put r = ∣ τ ∣ = n − c > 0 r=|\tau|=n-c>0 r = ∣ τ ∣ = n − c > 0 , and set
N = ( n 2 ) , S = r ( r + 1 ) 2 , L = N − S + 1. N=\binom{n}{2},\qquad S=\frac{r(r+1)}{2},\qquad L=N-S+1. N = ( 2 n ) , S = 2 r ( r + 1 ) , L = N − S + 1.
Write K a , j K_{a,j} K a , j for the primitive harmonic-form model of the row hook ( a , 1 j ) (a,1^j) ( a , 1 j ) from Lemma 4.1, and put
M μ = ⨁ a ≥ 1 Hom S n ( S μ , K a , r − 1 ) . \mathcal{M}_{\mu}=\bigoplus_{a\ge1}\operatorname{Hom}_{S_n}(S^{\mu},K_{a,r-1}). M μ = a ≥ 1 ⨁ Hom S n ( S μ , K a , r − 1 ) .
Its grading is the first row length a a a . We construct the representation behind Theorem 4.4, using the bounded-exponent model of the ordinary coinvariant algebra. The spaces K a , r − 1 K_{a,r-1} K a , r − 1 are the canonical harmonic-form models of stable row-hook multiplicity spaces. The auxiliary action is defined on M μ \mathcal{M}_{\mu} M μ . In first-row degree a a a , the ordinary G L k \mathrm{GL}_k GL k -isotypic component is S ( a , 1 r − 1 ) K k ⊗ M μ [ a ] S^{(a,1^{r-1})}K^k\otimes\mathcal{M}_{\mu}[a] S ( a , 1 r − 1 ) K k ⊗ M μ [ a ] , with the row group acting on the first factor.
Let C n = K [ x 1 , … , x n ] / ( K [ x 1 , … , x n ] + S n ) . C_n=K[x_1,\ldots,x_n]/(K[x_1,\ldots,x_n]^{S_n}_+). C n = K [ x 1 , … , x n ] / ( K [ x 1 , … , x n ] + S n ) . For 1 ≤ d < n 1\le d<n 1 ≤ d < n , let J n , d ⊂ C n J_{n,d}\subset C_n J n , d ⊂ C n be generated by the determinants det ( x b i a ) 1 ≤ a , b ≤ d \det(x_b^{i_a})_{1\le a,b\le d} det ( x b i a ) 1 ≤ a , b ≤ d over all d d d -element label sets. Write Ω n \Omega_n Ω n for the coinvariant algebra of differential forms, apolar dual to S H n SH_n S H n . Proposition B.1 proves the bounded-exponent complement and its Foulkes character, Ann C n ( J n , d ) = J n , n − d \operatorname{Ann}_{C_n}(J_{n,d})=J_{n,n-d} Ann C n ( J n , d ) = J n , n − d , and the prefix-contraction image used below.
Let U c = K [ t ] < c U_c=K[t]_{<c} U c = K [ t ] < c with ordered basis 1 , t , … , t c − 1 1,t,\ldots,t^{c-1} 1 , t , … , t c − 1 , and identify B c = U c ⊗ n B_c=U_c^{\otimes n} B c = U c ⊗ n with the bounded-exponent subspace
B c = span { x 1 a 1 ⋯ x n a n : 0 ≤ a i < c } ⊂ K [ x 1 , … , x n ] . B_c=\operatorname{span}\{x_1^{a_1}\cdots x_n^{a_n}:0\le a_i<c\}\subset K[x_1,\ldots,x_n]. B c = span { x 1 a 1 ⋯ x n a n : 0 ≤ a i < c } ⊂ K [ x 1 , … , x n ] .
The group G L ( U c ) \mathrm{GL}(U_c) GL ( U c ) acts on the tensor factors and commutes with S n S_n S n . Transporting this action will give the action on M μ \mathcal{M}_{\mu} M μ .
Lemma 8.1 (The bounded-exponent realization at the minimum descent number). The quotient projection induces a graded S n S_n S n -equivariant isomorphism
( B c ) μ ′ → ∼ ( C n / J n , c ) μ ′ , (B_c)_{\mu'}\xrightarrow{\sim}(C_n/J_{n,c})_{\mu'}, ( B c ) μ ′ ∼ ( C n / J n , c ) μ ′ , with the quotient and determinant ideal just defined.
Proof. The straightening argument in Appendix B gives the surjection B c → C n / J n , c B_c\to C_n/J_{n,c} B c → C n / J n , c . Applying the μ ′ \mu' μ ′ central idempotent preserves surjectivity. Schur–Weyl duality identifies the graded S μ S^\mu S μ multiplicity in the source with S μ ′ U c S_{\mu'}U_c S μ ′ U c , whose character is s μ ′ ( 1 , q , … , q c − 1 ) s_{\mu'}(1,q,\ldots,q^{c-1}) s μ ′ ( 1 , q , … , q c − 1 ) . Equations (B.1) and (B.5) identify the target multiplicity character as
∑ T ∈ SYT ( μ ′ ) des ( T ) < c q maj ( T ) . \sum_{\substack{T\in\operatorname{SYT}(\mu')\\ \operatorname{des}(T)<c}}q^{\operatorname{maj}(T)}. T ∈ SYT ( μ ′ ) des ( T ) < c ∑ q maj ( T ) .
The shape μ ′ \mu' μ ′ has exactly c c c rows, so each tableau has at least c − 1 c-1 c − 1 descents. Principal specialization on c c c variables identifies the displayed sum, consisting precisely of tableaux with c − 1 c-1 c − 1 descents, with s μ ′ ( 1 , q , … , q c − 1 ) s_{\mu'}(1,q,\ldots,q^{c-1}) s μ ′ ( 1 , q , … , q c − 1 ) . Thus the surjection has equal source and target dimensions in every degree, proving injectivity as well.
Use the harmonic-form operators
Θ j = ∑ i θ i ∂ x i j , ι = ∑ i x i ∂ θ i , d = d d R = Θ 1 . \Theta_j=\sum_i\theta_i\partial_{x_i}^j,\qquad\iota=\sum_i x_i\partial_{\theta_i},\qquad d=d_{\mathrm{dR}}=\Theta_1. Θ j = i ∑ θ i ∂ x i j , ι = i ∑ x i ∂ θ i , d = d dR = Θ 1 .
For a homogeneous b ∈ ( B c ) μ b\in(B_c)_\mu b ∈ ( B c ) μ of degree h h h , define
Φ μ ( b ) = 1 N − h − S + r ι Θ 1 Θ 2 ⋯ Θ r b ( ∂ ) Δ n . (53) \Phi_\mu(b)=\frac{1}{N-h-S+r}\iota\Theta_1\Theta_2\cdots\Theta_r b(\partial)\Delta_n. \tag*{(53)} Φ μ ( b ) = N − h − S + r 1 ι Θ 1 Θ 2 ⋯ Θ r b ( ∂ ) Δ n . (53)
The denominator is positive whenever the source component is nonzero. Indeed, S μ ′ U c = det ( U c ) ⊗ S τ ′ U c S_{\mu'}U_c=\det(U_c)\otimes S_{\tau'}U_c S μ ′ U c = det ( U c ) ⊗ S τ ′ U c , so its largest polynomial degree is at most ( c 2 ) + r ( c − 1 ) = N − S \binom{c}{2}+r(c-1)=N-S ( 2 c ) + r ( c − 1 ) = N − S . Consequently N − h − S ≥ 0 N-h-S\ge0 N − h − S ≥ 0 and N − h − S + r ≥ r > 0 N-h-S+r\ge r>0 N − h − S + r ≥ r > 0 .
Theorem 8.2 (An explicit Schur-module realization). The map (53) is a sign-intertwining isomorphism
( B c ) μ ′ → ∼ ⨁ a ≥ 1 ( K a , r − 1 ) μ , h ↦ a = L − h . (B_c)_{\mu'}\xrightarrow{\sim}\bigoplus_{a\ge1}(K_{a,r-1})_\mu,\qquad h\mapsto a=L-h. ( B c ) μ ′ ∼ a ≥ 1 ⨁ ( K a , r − 1 ) μ , h ↦ a = L − h . On multiplicity spaces, it constructs an auxiliary G L c \mathrm{GL}_c GL c action for which
M μ ≅ S τ ′ U c \mathcal{M}_\mu\cong S_{\tau'}U_c M μ ≅ S τ ′ U c as representations, with the displayed grading given by the principal cocharacter ρ ( z ) = diag ( 1 , z , … , z c − 1 ) \rho(z)=\operatorname{diag}(1,z,\ldots,z^{c-1}) ρ ( z ) = diag ( 1 , z , … , z c − 1 ) : weight a − 1 a-1 a − 1 is the harmonic multiplicity space M μ [ a ] \mathcal{M}_\mu[a] M μ [ a ] . Thus the graded character has the shift by q q q in Theorem 4.4 ; the group acts across first-row degrees. The action is obtained from the fixed bounded-exponent model and the displayed differential map; no choice of a graded basis in the harmonic multiplicity spaces is involved.
Proof. The perfect ordinary coinvariant pairing is
C n [ h ] ⊗ C n [ N − h ] ⟶ K Δ n . C_n[h]\otimes C_n[N-h]\longrightarrow K\Delta_n. C n [ h ] ⊗ C n [ N − h ] ⟶ K Δ n .
The complementary annihilator identity (B.3) states Ann C n ( J n , r ) = J n , c \operatorname{Ann}_{C_n}(J_{n,r})=J_{n,c} Ann C n ( J n , r ) = J n , c . Consequently the pairing restricts to a perfect sign-valued pairing between C n / J n , c C_n/J_{n,c} C n / J n , c and J n , r J_{n,r} J n , r , reversing degree around N N N .
On Ω n \Omega_n Ω n , put δ j = ∑ i x i j ∂ θ i \delta_j=\sum_i x_i^j\partial_{\theta_i} δ j = ∑ i x i j ∂ θ i . The contraction δ r ⋯ δ 1 : Ω n r → C n \delta_r\cdots\delta_1:\Omega_n^r\to C_n δ r ⋯ δ 1 : Ω n r → C n has image J n , r J_{n,r} J n , r , and its apolar adjoint is Θ 1 ⋯ Θ r \Theta_1\cdots\Theta_r Θ 1 ⋯ Θ r . On label μ \mu μ , the largest possible descent number is r r r . The Fields character [27 Theorem 5.2] in exterior degree r r r therefore has only the tableaux with exactly r r r descents and has polynomial degrees maj ( T ) − S \operatorname{maj}(T)-S maj ( T ) − S . The character of ( J n , r ) μ (J_{n,r})_\mu ( J n , r ) μ consists of those same tableaux, with degrees maj ( T ) \operatorname{maj}(T) maj ( T ) . Thus the prefix contraction is an isomorphism on that label, with degree shift S S S . Its adjoint identifies the perfect pairing above with the concrete map
b ↦ Θ 1 ⋯ Θ r b ( ∂ ) Δ n b\mapsto\Theta_1\cdots\Theta_r b(\partial)\Delta_n b ↦ Θ 1 ⋯ Θ r b ( ∂ ) Δ n
from ( C n / J n , c ) μ ′ (C_n/J_{n,c})_{\mu'} ( C n / J n , c ) μ ′ onto ( S H n r ) μ (SH_n^r)_\mu ( S H n r ) μ , with the sign twist. Its kernel on C n C_n C n is exactly the annihilator J n , c J_{n,c} J n , c , so the map factors through the stated quotient.
There is no label-μ \mu μ component in exterior degree r + 1 r+1 r + 1 . Hence every form in ( S H n r ) μ (SH_n^r)_\mu ( S H n r ) μ is killed by d d d . The Euler identity d ι + ι d = ( p + r ) id d\iota+\iota d=(p+r)\operatorname{id} d ι + ι d = ( p + r ) id in polynomial degree p p p identifies that space with primitive forms of exterior degree r − 1 r-1 r − 1 , by ι / ( p + r ) \iota/(p+r) ι / ( p + r ) , with inverse d d d . Taking p = N − h − S p=N-h-S p = N − h − S proves the explicit formula and the isomorphism, using the preceding bounded-exponent lemma.
Apolarity reverses polynomial degree. Reversing the monomial basis converts this to principal weights, and a determinant twist removes the full first column of μ ′ \mu' μ ′ . Explicitly, let J : U c → U c J:U_c\to U_c J : U c → U c send t j t^j t j to t c − 1 − j t^{c-1-j} t c − 1 − j . Transport to the target the following action on the source multiplicity space S μ ′ U c S_{\mu'}U_c S μ ′ U c :
g ↦ det ( g ) − 1 S μ ′ ( J g J − 1 ) . g\mapsto\det(g)^{-1}S_{\mu'}(JgJ^{-1}). g ↦ det ( g ) − 1 S μ ′ ( J g J − 1 ) .
Since μ ′ = ( 1 c ) + τ ′ \mu'=(1^c)+\tau' μ ′ = ( 1 c ) + τ ′ , this is the polynomial irreducible representation S τ ′ U c S_{\tau'}U_c S τ ′ U c . If g = diag ( 1 , z , … , z c − 1 ) g=\operatorname{diag}(1,z,\ldots,z^{c-1}) g = diag ( 1 , z , … , z c − 1 ) , its eigenvalue on source degree h h h is
z n ( c − 1 ) − h − ( c 2 ) = z L − h − 1 = z a − 1 . z^{n(c-1)-h-\binom{c}{2}}=z^{L-h-1}=z^{a-1}. z n ( c − 1 ) − h − ( 2 c ) = z L − h − 1 = z a − 1 .
Here N − S = ( c 2 ) + r ( c − 1 ) N-S=\binom{c}{2}+r(c-1) N − S = ( 2 c ) + r ( c − 1 ) . This proves the grading shift by one. Normalizing the Specht pairing or Vandermonde rescales the intertwiner and does not change the transported action. □ \square □
Corollary 8.3 (Lefschetz maps on harmonic multiplicities). There are explicitly defined operators E E E , F F F , H H H on M μ M_\mu M μ satisfying the s l 2 \mathfrak{sl}_2 sl 2 relations. They have grading changes + 1 +1 + 1 , − 1 -1 − 1 , 0 0 0 , and
H ∣ M μ [ a ] = ( 2 ( a − 1 ) − r ( c − 1 ) ) id . H|_{M_\mu[a]} = \left(2(a-1)-r(c-1)\right)\operatorname{id}. H ∣ M μ [ a ] = ( 2 ( a − 1 ) − r ( c − 1 ) ) id . In particular, whenever a ≤ 1 + r ( c − 1 ) / 2 a \le1+r(c-1)/2 a ≤ 1 + r ( c − 1 ) /2 ,
E 2 + r ( c − 1 ) − 2 a : M μ [ a ] → ∼ M μ [ 2 + r ( c − 1 ) − a ] . E^{2+r(c-1)-2a}:M_\mu[a]\xrightarrow{\sim}M_\mu[2+r(c-1)-a]. E 2 + r ( c − 1 ) − 2 a : M μ [ a ] ∼ M μ [ 2 + r ( c − 1 ) − a ] . Proof. On B c B_c B c , use the usual tensor action of the irreducible s l 2 \mathfrak{sl}_2 sl 2 module U c = Sym c − 1 K 2 U_c=\operatorname{Sym}^{c-1}K^2 U c = Sym c − 1 K 2 :
E B = ∑ i ( ( c − 1 ) x i − x i 2 ∂ x i ) , F B = ∑ i ∂ x i , H B = 2 ∑ i x i ∂ x i − n ( c − 1 ) . E_B=\sum_i\left((c-1)x_i-x_i^2\partial_{x_i}\right),\qquad F_B=\sum_i\partial_{x_i},\qquad H_B=2\sum_i x_i\partial_{x_i}-n(c-1). E B = i ∑ ( ( c − 1 ) x i − x i 2 ∂ x i ) , F B = i ∑ ∂ x i , H B = 2 i ∑ x i ∂ x i − n ( c − 1 ) . These preserve the bounded-exponent space and commute with S n S_n S n . Basis reversal exchanges E B E_B E B and F B F_B F B and negates H B H_B H B . Thus the desired operators are
E = Φ μ F B Φ μ − 1 , F = Φ μ E B Φ μ − 1 , H = − Φ μ H B Φ μ − 1 , E=\Phi_\mu F_B\Phi_\mu^{-1},\qquad F=\Phi_\mu E_B\Phi_\mu^{-1},\qquad H=-\Phi_\mu H_B\Phi_\mu^{-1}, E = Φ μ F B Φ μ − 1 , F = Φ μ E B Φ μ − 1 , H = − Φ μ H B Φ μ − 1 , on the sign-adjusted multiplicity spaces. The inverse means the unique bounded-exponent preimage under the explicit isomorphism above. The Cartan eigenvalue follows from a = L − h a=L-h a = L − h . Decomposition into finite dimensional s l 2 \mathfrak{sl}_2 sl 2 modules gives the stated Lefschetz isomorphisms. Since all row hooks are Vandermonde-generated, these operators act on the same actual subspaces in both modules. □
Tensor coinvariants, Schur resolutions and finite jets
The selected-label presentation retains the algebra of finite jets. A canonical Schur cover then describes every label and all its moment homology. Retain the augmented algebra B B B , its augmentation ideal m \mathfrak m m , and Q n ( B ) Q_n(B) Q n ( B ) from § 2 ; initially, only invertibility of n ! n! n ! in K K K is required.
Theorem 9.1 (First parabolics are jets). For every augmented commutative algebra B B B as above, inclusion of the first tensor factor gives a natural algebra isomorphism
Q n ( B ) S n − 1 ≅ B / m n (54) Q_n(B)^{S_{n-1}}\cong B/\mathfrak m^n \tag*{(54)} Q n ( B ) S n − 1 ≅ B / m n (54) For n ≥ 2 n\ge2 n ≥ 2 , let V n = S ( n − 1 , 1 ) V_n=S^{(n-1,1)} V n = S ( n − 1 , 1 ) be the standard representation. There is a natural isomorphism
Hom S n ( V n , Q n ( B ) ) ≅ m / m n (55) \operatorname{Hom}_{S_n}(V_n,Q_n(B))\cong\mathfrak m/\mathfrak m^n \tag*{(55)} Hom S n ( V n , Q n ( B )) ≅ m / m n (55) of modules for every augmentation-preserving symmetry group of B B B . If B B B is graded and ε \varepsilon ε preserves the grading, both isomorphisms preserve the grading.
Proof. Take r = 1 r=1 r = 1 in Theorem 2.1 . The relation is b n = 0 b^n=0 b n = 0 for every b ∈ m b\in\mathfrak m b ∈ m . The ideal generated by these powers is exactly m n \mathfrak m^n m n : the reverse inclusion follows by polarizing the n n n th power, or directly from
n ! b 1 ⋯ b n = ∑ J ⊆ { 1 , … , n } ( − 1 ) n − ∣ J ∣ ( ∑ j ∈ J b j ) n . n!b_1\cdots b_n=\sum_{J\subseteq\{1,\ldots,n\}}(-1)^{n-|J|}\left(\sum_{j\in J}b_j\right)^n. n ! b 1 ⋯ b n = J ⊆ { 1 , … , n } ∑ ( − 1 ) n − ∣ J ∣ j ∈ J ∑ b j n . Moreover Q n ( B ) S n = K Q_n(B)^{S_n}=K Q n ( B ) S n = K , by Reynolds averaging and the definition of the invariant-generated ideal. The permutation module Ind S n − 1 S n 1 \operatorname{Ind}_{S_{n-1}}^{S_n}\mathbf{1} Ind S n − 1 S n 1 splits as 1 ⊕ V n \mathbf{1}\oplus V_n 1 ⊕ V n . Frobenius reciprocity therefore identifies the standard multiplicity space with the augmentation kernel of Q n ( B ) S n − 1 Q_n(B)^{S_{n-1}} Q n ( B ) S n − 1 , which is m / m n \mathfrak m/\mathfrak m^n m / m n under (9.1). Concretely, the reciprocity map evaluates a homomorphism on e 1 − n − 1 ∑ i e i e_1-n^{-1}\sum_i e_i e 1 − n − 1 ∑ i e i , so the identification is natural. □
For B = Sym V B=\operatorname{Sym}V B = Sym V , the standard multiplicity space is ⨁ d = 1 n − 1 Sym d V \bigoplus_{d=1}^{n-1}\operatorname{Sym}^dV ⨁ d = 1 n − 1 Sym d V . This polynomial specialization recovers the bosonic standard-character formula of Jiang–Lenfer [21 ] , Proposition 6.5 and Theorem 6.6; their theorem also treats fermionic alphabets. The statement above identifies the algebra and its augmentation kernel for arbitrary augmented B B B .
Canonical Schur covers and their relations.
For the remainder of this section K K K has characteristic zero, and B = K ⊕ m B = K \oplus\mathfrak{m} B = K ⊕ m is any augmented commutative algebra. Put S B = Sym ( m ) S_B = \operatorname{Sym}(\mathfrak{m}) S B = Sym ( m ) ; its generator T b T_b T b , linear in b ∈ m b \in\mathfrak{m} b ∈ m , acts on P n ( B ) = B ⊗ n P_n(B) = B^{\otimes n} P n ( B ) = B ⊗ n by p n ( b ) = ∑ j b ( j ) p_n(b) = \sum_j b^{(j)} p n ( b ) = ∑ j b ( j ) . The trace presentation in Section 2 identifies Q n ( B ) Q_n(B) Q n ( B ) with P n ( B ) / ( S B ) + P n ( B ) P_n(B)/(S_B)_+P_n(B) P n ( B ) / ( S B ) + P n ( B ) .
The augmentation decomposes P n ( B ) P_n(B) P n ( B ) by occupied positions:
P n ( B ) = ⨁ j = 0 n P n ⟨ j ⟩ ( B ) , P n ⟨ j ⟩ ( B ) = ⨁ I ⊆ [ n ] ∣ I ∣ = j m ⊗ I , P_n(B) = \bigoplus_{j=0}^{n} P_n^{\langle j\rangle}(B), \qquad P_n^{\langle j\rangle}(B) = \bigoplus_{\substack{I\subseteq[n]\\ |I|=j}} \mathfrak{m}^{\otimes I}, P n ( B ) = j = 0 ⨁ n P n ⟨ j ⟩ ( B ) , P n ⟨ j ⟩ ( B ) = I ⊆ [ n ] ∣ I ∣ = j ⨁ m ⊗ I ,
with units elsewhere. If e μ e_\mu e μ is the central label idempotent, define the actual subspace
L n ( B ) = ⨁ μ ⊢ n e μ P n ( n − μ 1 ) ( B ) ⊆ P n ( B ) . L_n(B) = \bigoplus_{\mu\vdash n} e_\mu P_n^{(n-\mu_1)}(B) \subseteq P_n(B). L n ( B ) = μ ⊢ n ⨁ e μ P n ( n − μ 1 ) ( B ) ⊆ P n ( B ) .
Lemma 9.2 (Minimum occupancy). The quotient map L n ( B ) → Q n ( B ) L_n(B) \to Q_n(B) L n ( B ) → Q n ( B ) is surjective. For μ = ( t , τ ) ⊢ n \mu=(t,\tau)\vdash n μ = ( t , τ ) ⊢ n , its label-μ \mu μ source multiplicity is S τ m S_\tau\mathfrak{m} S τ m .
Proof. Put s = ∣ τ ∣ s=|\tau| s = ∣ τ ∣ . Selected elimination makes B ⊗ s → Q n ( B ) S t B^{\otimes s}\to Q_n(B)^{S_t} B ⊗ s → Q n ( B ) S t surjective. A selected tensor with a unit occupies at most s − 1 s-1 s − 1 positions, so its global orbit is induced from a subgroup containing S t + 1 S_{t+1} S t + 1 . Its label first row is therefore at least t + 1 t+1 t + 1 , and its μ \mu μ projection is zero. Fully occupied tensors cover the S t S_t S t -fixed part of label μ \mu μ , and its S n S_n S n -span covers the whole isotype. For the trivial label this says Q n ( B ) S n = K Q_n(B)^{S_n}=K Q n ( B ) S n = K .
Finally, P n ⟨ s ⟩ = Ind S s × S t S n ( m ⊗ s ⊗ 1 ) P_n^{\langle s\rangle}=\operatorname{Ind}_{S_s\times S_t}^{S_n}(\mathfrak{m}^{\otimes s}\otimes1) P n ⟨ s ⟩ = Ind S s × S t S n ( m ⊗ s ⊗ 1 ) . If ∣ ν ∣ = s |\nu|=s ∣ ν ∣ = s and μ / ν \mu/\nu μ / ν is horizontal, Pieri interlacing gives ν i ≥ τ i \nu_i\geq\tau_i ν i ≥ τ i ; equality of sizes forces ν = τ \nu=\tau ν = τ . □ \square □
For μ = ( t , τ ) \mu=(t,\tau) μ = ( t , τ ) write a = τ 1 a=\tau_1 a = τ 1 , with a = 0 a=0 a = 0 for the empty tail, and put
τ + = ( t + 1 , τ 2 , τ 3 , … ) , d 1 = ∣ τ + ∣ = n + 1 − a , M μ ( B ) = Hom S n ( S μ , Q n ( B ) ) . \tau^+=(t+1,\tau_2,\tau_3,\ldots), \qquad d_1=|\tau^+|=n+1-a, \qquad M_\mu(B)=\operatorname{Hom}_{S_n}(S^\mu,Q_n(B)). τ + = ( t + 1 , τ 2 , τ 3 , … ) , d 1 = ∣ τ + ∣ = n + 1 − a , M μ ( B ) = Hom S n ( S μ , Q n ( B )) .
Fix the usual one-dimensional Pieri identification lines.
Theorem 9.3 (The two-Schur presentation). There are natural exact presentations
S B ⊗ S τ + m → R μ , B S B ⊗ S τ m → Φ μ , B S μ ( B ) ⟶ 0 , S_B\otimes S_{\tau^+}\mathfrak{m}\xrightarrow{R_{\mu,B}} S_B\otimes S_\tau\mathfrak{m}\xrightarrow{\Phi_{\mu,B}} S_\mu(B)\longrightarrow0, S B ⊗ S τ + m R μ , B S B ⊗ S τ m Φ μ , B S μ ( B ) ⟶ 0 ,
S τ + m → β μ , B S τ m ⟶ M μ ( B ) ⟶ 0. S_{\tau^+}\mathfrak{m}\xrightarrow{\beta_{\mu,B}} S_\tau\mathfrak{m}\longrightarrow M_\mu(B)\longrightarrow0. S τ + m β μ , B S τ m ⟶ M μ ( B ) ⟶ 0.
Here Φ \Phi Φ is actual multiplication by traces on minimum occupancy. All maps commute with arbitrary augmented algebra maps, including ungraded and noninjective maps, and preserve compatible gradings and auxiliary symmetries. The left-hand maps need not be injective.
Proof. Filter F μ = S B ⊗ S τ m F_\mu=S_B\otimes S_\tau\mathfrak{m} F μ = S B ⊗ S τ m by input count: j j j formal traces and ∣ τ ∣ |\tau| ∣ τ ∣ inputs have count j + ∣ τ ∣ j+|\tau| j + ∣ τ ∣ . Filter the target by occupied positions. Collisions lower this count, so the associated-graded map is the same cover for the square-zero algebra B 0 = K ⊕ m B_0=K\oplus\mathfrak{m} B 0 = K ⊕ m . Giving m \mathfrak{m} m degree one, Lemma 9.2 makes this cover surjective by degree induction. Pieri and branching off the constant factor give
F μ = ⨁ ν / τ horizontal S ν m , F_\mu=\bigoplus_{\substack{\nu/\tau\ \text{horizontal}}} S_\nu\mathfrak{m}, F μ = ν / τ horizontal ⨁ S ν m ,
S μ ( K ⊕ m ) = ⨁ ν / τ horizontal ν 1 ≤ t S ν m . S_\mu(K\oplus\mathfrak{m})=\bigoplus_{\substack{\nu/\tau\ \text{horizontal}\\ \nu_1\leq t}} S_\nu\mathfrak{m}. S μ ( K ⊕ m ) = ν / τ horizontal ν 1 ≤ t ⨁ S ν m .
Let G μ G_\mu G μ be the sum of the retained source summands ν 1 ≤ t \nu_1\leq t ν 1 ≤ t , and Z μ Z_\mu Z μ the omitted sum. These are canonical Schur isotypic subfunctors. The square-zero map is zero on Z μ Z_\mu Z μ and an isomorphism on G μ G_\mu G μ . For B B B , the restriction Ψ B = Φ μ , B ∣ G μ \Psi_B=\Phi_{\mu,B}|_{G_\mu} Ψ B = Φ μ , B ∣ G μ has the same diagonal and only lower-count terms. Retained counts lie between ∣ τ ∣ |\tau| ∣ τ ∣ and n n n , so its triangular inverse is finite. The entire kernel is the graph
Z μ → ∼ ker Φ μ , B , z ⟼ z − Ψ B − 1 Φ μ , B ( z ) . (56) Z_\mu\xrightarrow{\sim} \ker\Phi_{\mu,B}, \qquad z\longmapsto z-\Psi_B^{-1}\Phi_{\mu,B}(z). \tag*{(56)} Z μ ∼ ker Φ μ , B , z ⟼ z − Ψ B − 1 Φ μ , B ( z ) . (56)
Every omitted ν \nu ν contains τ + \tau^+ τ + and ν / τ + \nu/\tau^+ ν / τ + is horizontal. The symmetric Pieri composition lemma [32 Lemma 2.1] and multiplicity one therefore give Z μ = S B ⋅ S τ + m Z_\mu=S_B\cdot S_{\tau^+}\mathbf m Z μ = S B ⋅ S τ + m for the square-zero map. Restrict (9.3) to this generating summand and extend S B S_B S B -linearly. Subtracting lifted generators with a prescribed leading term lowers count, so finite induction proves that they generate the actual kernel. Reduction modulo ( S B ) + (S_B)_+ ( S B ) + gives the second presentation, with the convention β μ , B = − ev T = 0 R μ , B \beta_{\mu,B}=-\operatorname{ev}_{T=0}R_{\mu,B} β μ , B = − ev T = 0 R μ , B .
Schur projections, multiplication, and the inverse of the natural isomorphism Ψ B \Psi_B Ψ B commute with augmented maps. For infinite-dimensional m \mathbf m m , the square-zero constructions pass through finite-dimensional vector subspaces; the formulas use finitely many products on each input. □
The graph is an identification of filtered vector spaces, not generally of S B S_B S B -modules. For example,
M ( 2 , 2 ) ( B ) = Sym 2 m / span { a b ⊙ c + a c ⊙ b + b c ⊙ a : a , b , c ∈ m } . M_{(2,2)}(B)=\operatorname{Sym}^2\mathbf m/\operatorname{span}\{ab\mathbin{\odot}c+ac\mathbin{\odot}b+bc\mathbin{\odot}a:a,b,c\in\mathbf m\}. M ( 2 , 2 ) ( B ) = Sym 2 m / span { ab ⊙ c + a c ⊙ b + b c ⊙ a : a , b , c ∈ m } .
The standard label gives m / m n \mathbf m/\mathbf m^n m / m n , as above. If m 2 = 0 \mathbf m^2=0 m 2 = 0 , every β \beta β is zero and M ( t , τ ) ( B ) = S τ m M_{(t,\tau)}(B)=S_\tau\mathbf m M ( t , τ ) ( B ) = S τ m in every rank.
Proposition 9.4 (The first obstruction). Suppose B B B is connected positively graded. The actual free cover on label μ \mu μ is an isomorphism in degrees less than d 1 = n + 1 − τ 1 d_1=n+1-\tau_1 d 1 = n + 1 − τ 1 , and every positive moment Tor \operatorname{Tor} Tor group on this label vanishes there. In degree d 1 d_1 d 1 there is a natural exact sequence
0 ⟶ Hom S n ( S μ , Tor 1 S B ( P n ( B ) , K ) d 1 ) ⟶ S τ + ( B 1 ) 0\longrightarrow\operatorname{Hom}_{S_n}\left(S^\mu,\operatorname{Tor}^{S_B}_1(P_n(B),K)_{d_1}\right)\longrightarrow S_{\tau^+}(B_1) 0 ⟶ Hom S n ( S μ , Tor 1 S B ( P n ( B ) , K ) d 1 ) ⟶ S τ + ( B 1 ) → β μ , B ( S τ m ) d 1 ⟶ M μ ( B ) d 1 ⟶ 0. \xrightarrow{\beta_{\mu,B}}(S_\tau\mathbf m)_{d_1}\longrightarrow M_\mu(B)_{d_1}\longrightarrow0. β μ , B ( S τ m ) d 1 ⟶ M μ ( B ) d 1 ⟶ 0. All higher Tor \operatorname{Tor} Tor groups on this label vanish in degree d 1 d_1 d 1 . More generally, i i i th Tor \operatorname{Tor} Tor vanishes in degrees less than d 1 + i − 1 d_1+i-1 d 1 + i − 1 .
Proof. Every omitted partition has size at least d 1 d_1 d 1 , with equality only for τ + \tau^+ τ + . The graph construction identifies the first possible kernel piece with S τ + ( B 1 ) S_{\tau^+}(B_1) S τ + ( B 1 ) . Tensor the free-cover sequence with K K K . Positive coefficients cannot reach that degree, giving the exact sequence. The Koszul chains of the kernel in homological degree i − 1 i-1 i − 1 have degree at least d 1 + i − 1 d_1+i-1 d 1 + i − 1 . □
A full resolution and its polynomial first degrees
The presentation extends in every homological degree. Extend the parts of μ \mu μ by zero and set
ρ i = ( μ 1 + 1 , … , μ i + 1 , μ i + 2 , μ i + 3 , … ) , d i = ∣ ρ i ∣ = n − μ i + 1 + i . (57) \rho_i=(\mu_1+1,\ldots,\mu_i+1,\mu_{i+2},\mu_{i+3},\ldots),\qquad d_i=\lvert\rho_i\rvert=n-\mu_{i+1}+i. \tag*{(57)} ρ i = ( μ 1 + 1 , … , μ i + 1 , μ i + 2 , μ i + 3 , … ) , d i = ∣ ρ i ∣ = n − μ i + 1 + i . (57)
Thus ρ 0 = τ \rho_0=\tau ρ 0 = τ and ρ 1 = τ + \rho_1=\tau^+ ρ 1 = τ + . The operation deletes row i + 1 i+1 i + 1 and adds one box to each preceding row.
Theorem 9.5 (A natural filtered Schur resolution). For every augmented commutative B B B in characteristic zero, S μ ( B ) S_\mu(B) S μ ( B ) has a natural S B S_B S B -free resolution with terms
F i = S B ⊗ S ρ i m . F_i=S_B\otimes S_{\rho_i}\mathbf m. F i = S B ⊗ S ρ i m . The augmentation and first differential are those of Theorem 9.3. Its associated graded for input count is the classical pure Pieri resolution; the lifted resolution need not be minimal. Modulo ( S B ) + (S_B)_+ ( S B ) + it gives a complex
⋯ ⟶ S ρ 2 m → ∂ 2 S ρ 1 m → ∂ 1 S τ m (58) \cdots\longrightarrow S_{\rho_2}\mathbf m\xrightarrow{\partial_2}S_{\rho_1}\mathbf m\xrightarrow{\partial_1}S_\tau\mathbf m \tag*{(58)} ⋯ ⟶ S ρ 2 m ∂ 2 S ρ 1 m ∂ 1 S τ m (58) computing Hom S n ( S μ , Tor i S B ( P n ( B ) , K ) ) \operatorname{Hom}_{S_n}(S^\mu,\operatorname{Tor}^{S_B}_i(P_n(B),K)) Hom S n ( S μ , Tor i S B ( P n ( B ) , K )) . The differential ∂ i \partial_i ∂ i uses exactly e i = μ i − μ i + 1 + 1 e_i=\mu_i-\mu_{i+1}+1 e i = μ i − μ i + 1 + 1 multiplications. It is natural for all augmented algebra maps and preserves compatible gradings. For dim m = q \dim\mathbf m=q dim m = q , the terms with i > q i>q i > q vanish.
Appendix C constructs the maps by a natural creation–annihilation contraction. Its Laplacian has a dimension-independent content eigenvalue, and multiplication is a finite perturbation. This constructs the full resolution without choices of higher splittings.
Nilpotent augmentation ideals. If m L = 0 m^L=0 m L = 0 with L ≥ 2 L\ge2 L ≥ 2 , then ∂ i = 0 \partial_i=0 ∂ i = 0 whenever d i > ( L − 1 ) d i − 1 d_i>(L-1)d_{i-1} d i > ( L − 1 ) d i − 1 : every term partitions its d i d_i d i inputs into d i − 1 d_{i-1} d i − 1 products, and one product must contain at least L L L inputs. In particular, the canonical presentation gives
M ( n − ∣ τ ∣ , τ ) ( B ) ≅ S τ m ( n ≥ ( L − 1 ) ∣ τ ∣ + τ 1 ) . M_{(n-|\tau|,\tau)}(B)\cong S_\tau\mathfrak{m}\qquad(n\ge(L-1)|\tau|+\tau_1). M ( n − ∣ τ ∣ , τ ) ( B ) ≅ S τ m ( n ≥ ( L − 1 ) ∣ τ ∣ + τ 1 ) .
This equality holds in every internal degree whenever B B B is graded. For square-zero algebras it holds at every rank supporting the label.
Corollary 9.6 (Sharp universal homological support). For connected positively graded B B B and i ≥ 1 i\ge1 i ≥ 1 , label μ \mu μ in degree-D D D moment homology requires
D ≥ n + i − μ i + 1 , D ≥ n + i − ⌊ n i + 1 ⌋ . D\ge n+i-\mu_{i+1},\qquad D\ge n+i-\left\lfloor\frac{n}{i+1}\right\rfloor. D ≥ n + i − μ i + 1 , D ≥ n + i − ⌊ i + 1 n ⌋ . Both bounds are attained by square-zero algebras with sufficiently large degree-one space.
Proof. The i i i th term of (58) begins in internal degree d i d_i d i , and μ i + 1 ≤ ⌊ n / ( i + 1 ) ⌋ \mu_{i+1}\le\lfloor n/(i+1)\rfloor μ i + 1 ≤ ⌊ n / ( i + 1 )⌋ . For square-zero multiplication all differentials are zero. A partition maximizing its ( i + 1 ) (i+1) ( i + 1 ) st part attains the second threshold. □ \square □
For a polynomial algebra, the first possible degree behaves better. The proof uses an ungraded specialization to orthogonal idempotents; the full naturality in Theorem 9.3 is essential.
Theorem 9.7 (Exactness in the first possible homological degree). Let A = Sym ( V ) A=\operatorname{Sym}(V) A = Sym ( V ) and S A = Sym ( A + ) S_A=\operatorname{Sym}(A_+) S A = Sym ( A + ) . If μ ≠ ( n ) \mu\ne(n) μ = ( n ) and i ≥ 1 i\ge1 i ≥ 1 , then
Hom S n ( S μ , Tor i S A ( A ⊗ n , K ) D ) = 0 ( D ≤ n + i − μ i + 1 ) . \operatorname{Hom}_{S_n}\left(S^\mu,\operatorname{Tor}^{S_A}_i(A^{\otimes n},K)_D\right)=0\qquad(D\le n+i-\mu_{i+1}). Hom S n ( S μ , Tor i S A ( A ⊗ n , K ) D ) = 0 ( D ≤ n + i − μ i + 1 ) . The same conclusion holds for μ = ( n ) \mu=(n) μ = ( n ) and i ≥ 2 i\ge2 i ≥ 2 . The sole exceptional first degree is
Hom S n ( 1 , Tor 1 S A ( A ⊗ n , K ) n + 1 ) = Sym n + 1 V . \operatorname{Hom}_{S_n}\left(\mathbf{1},\operatorname{Tor}^{S_A}_1(A^{\otimes n},K)_{n+1}\right)=\operatorname{Sym}^{n+1}V. Hom S n ( 1 , Tor 1 S A ( A ⊗ n , K ) n + 1 ) = Sym n + 1 V . Equivalently, every supported all-linear restriction ∂ i : S ρ i V → ( S ρ i − 1 A + ) d i \partial_i:S_{\rho_i}V\to(S_{\rho_{i-1}}A_+)_{d_i} ∂ i : S ρ i V → ( S ρ i − 1 A + ) d i is injective, except for μ = ( n ) , i = 1 \mu=(n),i=1 μ = ( n ) , i = 1 , where it is zero.
Proof. Take the split augmented algebra C = K q + 1 C=K^{q+1} C = K q + 1 , with augmentation its zeroth coordinate and augmentation ideal spanned by the orthogonal idempotents u 1 , … , u q u_1,\ldots,u_q u 1 , … , u q . On a primitive tensor idempotent of C ⊗ n C^{\otimes n} C ⊗ n , the moment p n ( u a ) p_n(u_a) p n ( u a ) acts by the number of positions of color a a a . Unless every position has color zero, some such count is a positive integer, and its Koszul complex contracts. Consequently
Q n ( C ) = K , Tor i Sym ( C + ) ( C ⊗ n , K ) = Λ i ( C + ) ⊗ 1. (59) Q_n(C)=K,\qquad\operatorname{Tor}^{\operatorname{Sym}(C_+)}_i(C^{\otimes n},K)=\Lambda^i(C_+)\otimes1. \tag*{(59)} Q n ( C ) = K , Tor i Sym ( C + ) ( C ⊗ n , K ) = Λ i ( C + ) ⊗ 1. (59)
Suppose μ ≠ ( n ) \mu\ne(n) μ = ( n ) and S ρ i K q ≠ 0 S_{\rho_i}K^q\ne0 S ρ i K q = 0 . The complex (58) for C C C is exact in every degree on this label. Its term dimensions are those of the square-zero pure resolution. Localizing that resolution at the fraction field of Sym ( K q ) \operatorname{Sym}(K^q) Sym ( K q ) also gives an exact complex: the square-zero coefficient module has finite length. Exactness determines the ranks of all differentials recursively from the term dimensions. Each supported pure Pieri differential is nonzero, so ∂ i \partial_i ∂ i for C C C is nonzero as well.
Set q = dim V q=\dim V q = dim V and extend a linear isomorphism V → C + V\to C_+ V → C + to an augmented algebra map A → C A\to C A → C . Naturality now makes the restriction of ∂ i \partial_i ∂ i to S ρ i V S_{\rho_i}V S ρ i V nonzero. It is GL ( V ) \operatorname{GL}(V) GL ( V ) -equivariant with irreducible source, hence injective. If that Schur module is zero, there is nothing to prove. For μ = ( n ) \mu=(n) μ = ( n ) and i ≥ 2 i\ge2 i ≥ 2 , take q = i q=i q = i . The top small-complex term is S ( n + 1 , 1 i − 1 ) K i S_{(n+1,1^{i-1})}K^i S ( n + 1 , 1 i − 1 ) K i , of dimension ( n + i − 1 n ) > 1 \binom{n+i-1}{n}>1 ( n n + i − 1 ) > 1 , whereas its kernel has dimension one by (9.6). Thus its differential is nonzero. The same specialization, followed by naturality in the row space, proves injectivity in every supporting dimension. For i = 1 i=1 i = 1 the target is K K K , so the positive-degree map is zero.
Finally, the degree-d i d_i d i part of the i i i th Schur term uses only V = A 1 V=A_1 V = A 1 , and the next term starts strictly above d i d_i d i . The assertions follow from the small complex and the preceding degree bounds. □ \square □
Theorem 9.8 (The first character correction in every degree). For every nonempty τ ⊢ s \tau\vdash s τ ⊢ s , every D > s D>s D > s , and n = D + τ 1 − 1 , μ = ( D − s + τ 1 − 1 , τ ) , τ ^ = ( D − s + τ 1 , τ 2 , τ 3 , … ) , n=D+\tau_{1}-1,\qquad\mu=(D-s+\tau_{1}-1,\tau),\qquad\widehat{\tau}=(D-s+\tau_{1},\tau_{2},\tau_{3},\ldots), n = D + τ 1 − 1 , μ = ( D − s + τ 1 − 1 , τ ) , τ = ( D − s + τ 1 , τ 2 , τ 3 , … ) , there is a natural exact sequence
0 ⟶ S τ V → β μ , A ( S τ A + ) D ⟶ Hom S n ( S μ , R n , k [ D ] ) ⟶ 0. (60) 0\longrightarrow S_{\tau}V \xrightarrow{\beta_{\mu,A}} (S_{\tau}A_{+})_{D}\longrightarrow\operatorname{Hom}_{S_{n}}(S^{\mu},R_{n,k}[D])\longrightarrow0. \tag*{(60)} 0 ⟶ S τ V β μ , A ( S τ A + ) D ⟶ Hom S n ( S μ , R n , k [ D ]) ⟶ 0. (60) All positive moment homology on this label vanishes through degree D D D . Thus, when dim V ≥ ℓ ( τ ) \dim V\geq\ell(\tau) dim V ≥ ℓ ( τ ) , the stable threshold n ≥ D + τ 1 n\geq D+\tau_{1} n ≥ D + τ 1 is sharp in every degree D > s D>s D > s .
Proof. Here D = d 1 D=d_{1} D = d 1 and ρ 1 = τ ^ \rho_{1}=\widehat{\tau} ρ 1 = τ . Apply Proposition 9.4 and Theorem 9.7. The correction has length ℓ ( τ ) \ell(\tau) ℓ ( τ ) and is nonzero under the stated dimension hypothesis. □
This identifies the canonical relation map at the first rank below the stable threshold.
Natural stable multiplicity spaces
Let B = K ⊕ m B=K\oplus\mathfrak{m} B = K ⊕ m be a connected positively graded commutative algebra over a field of characteristic zero. Retain Q n ( B ) Q_{n}(B) Q n ( B ) from Theorem 2.1. For τ ⊢ s \tau\vdash s τ ⊢ s write τ [ n ] = ( n − s , τ ) \tau[n]=(n-s,\tau) τ [ n ] = ( n − s , τ ) when this is a partition. For the empty partition set τ 1 = 0 \tau_{1}=0 τ 1 = 0 .
Theorem 9.9 (Natural stable multiplicity spaces). If D ≥ s D\geq s D ≥ s and n ≥ D + τ 1 n\geq D+\tau_{1} n ≥ D + τ 1 , then
Hom S n ( S τ [ n ] , Q n ( B ) D ) ≃ ( S τ m ) D . \operatorname{Hom}_{S_{n}}(S^{\tau[n]},Q_{n}(B)_{D})\simeq(S_{\tau}\mathfrak{m})_{D}. Hom S n ( S τ [ n ] , Q n ( B ) D ) ≃ ( S τ m ) D . Fixing the one-dimensional Pieri branching line makes the isomorphism natural for graded augmented algebra maps and auxiliary symmetries preserving the grading and augmentation. For D < s D<s D < s the left side is zero whenever τ [ n ] \tau[n] τ [ n ] is a partition.
Proof. The canonical cover in Theorem 9.3 has no relation in degrees below n + 1 − τ 1 n+1-\tau_{1} n + 1 − τ 1 , by Proposition 9.4. Its quotient modulo positive traces is therefore exactly S τ m S_{\tau}\mathfrak{m} S τ m in the stated range. This is the actual minimum-occupancy quotient map, so it has the asserted naturality. For D < s D<s D < s , occupancy gives zero. □
Corollary 9.10 (Universal separation and its sharp range). For D ≥ 1 D\geq1 D ≥ 1 and n ≥ 2 D − 1 n\geq2D-1 n ≥ 2 D − 1 , multiplication gives a canonical isomorphism
( P n ( B ) S n ⊗ L n ( B ) ) ≤ D → ∼ P n ( B ) ≤ D , \left(P_{n}(B)^{S_{n}}\otimes L_{n}(B)\right)_{\leq D}\xrightarrow{\ \sim\ }P_{n}(B)_{\leq D}, ( P n ( B ) S n ⊗ L n ( B ) ) ≤ D ∼ P n ( B ) ≤ D , and all positive moment homology vanishes through degree D D D . The bound is sharp for arbitrary connected graded augmented algebras.
Proof. A degree-d d d label has tail size at most d d d . At n = 2 d − 1 n=2d-1 n = 2 d − 1 , a tail with first part d d d would have to be ( d ) (d) ( d ) and would give the invalid label ( d − 1 , d ) (d-1,d) ( d − 1 , d ) . Thus every occurring label satisfies the individual bound n ≥ d + τ 1 n\geq d+\tau_{1} n ≥ d + τ 1 ; larger n n n are immediate. The trivial-label presentation shows that S B S_{B} S B and the actual invariant algebra agree through degree n n n , giving the multiplication statement.
For sharpness let D ≥ 2 D\geq2 D ≥ 2 , n = 2 D − 2 = 2 m n=2D-2=2m n = 2 D − 2 = 2 m , and B = K [ x ] / ( x 2 ) B=K[x]/(x^{2}) B = K [ x ] / ( x 2 ) . The sole moment
p = ∑ i = 1 2 m x i p=\sum_{i=1}^{2m}x_{i} p = i = 1 ∑ 2 m x i
kills the nonzero element
f = ∏ j = 1 m ( x 2 j − 1 − x 2 j ) . f=\prod_{j=1}^{m}(x_{2j-1}-x_{2j}). f = j = 1 ∏ m ( x 2 j − 1 − x 2 j ) .
Hence f e T x fe_{\mathrm{T}x} f e T x is nonzero first Koszul homology in degree D D D ; a one-generator complex has no boundaries there. The endpoint D = n = 1 D=n=1 D = n = 1 is included in the isomorphism. □
For a fixed tail, augmentation of the last factor gives compatible surjections M τ [ n ] ( B ) → M τ [ n − 1 ] ( B ) M_{\tau[n]}(B)\to M_{\tau[n-1]}(B) M τ [ n ] ( B ) → M τ [ n − 1 ] ( B ) whenever both labels are defined. With compatible Pieri lines they are quotients of the identity on S τ m S_{\tau}\mathfrak{m} S τ m , so their kernels decrease with n n n . Indeed, the occupancy-s s s projection is onto, and among its source labels of first row at least n − s n-s n − s , only τ [ n ] \tau[n] τ [ n ] branches to τ [ n − 1 ] \tau[n-1] τ [ n − 1 ] . Multiplicity monotonicity itself is known in the FI-module setting [8 Lemma 2.1.1] ; here the common source and maps are explicit for arbitrary augmented algebras.
Polynomial algebras and positive plethysm
Write
A = Sym V , A + = ⨁ d ≥ 1 Sym d V , H ( q ) = ∏ a = 1 k ( 1 − q a ) − 1 . A = \operatorname{Sym} V,\qquad A_{+} = \bigoplus_{d\geq1}\operatorname{Sym}^{d} V,\qquad H(\mathbf{q}) = \prod_{a=1}^{k}(1-q_{a})^{-1}. A = Sym V , A + = d ≥ 1 ⨁ Sym d V , H ( q ) = a = 1 ∏ k ( 1 − q a ) − 1 .
The grading on the Schur functor S τ ( A + ) S_{\tau}(A_{+}) S τ ( A + ) is induced by the usual polynomial grading of A + A_{+} A + . In particular, every homogeneous part is finite dimensional.
Corollary 9.11 (Explicit stable Specht multiplicities). For D ≥ 1 D \geq1 D ≥ 1 and n ≥ 2 D − 1 n \geq2D-1 n ≥ 2 D − 1 , there is an isomorphism of GL ( V ) × S n \operatorname{GL}(V)\times S_n GL ( V ) × S n -modules
( R n , k ) D ≅ ⨁ ∣ τ ∣ ≤ D τ [ n ] a partition ( S τ ( A + ) ) D ⊗ S ( n − ∣ τ ∣ , τ ) . (61) (R_{n,k})_{D} \cong\bigoplus_{\substack{|\tau|\leq D\\ \tau[n]\text{ a partition}}}(S_{\tau}(A_{+}))_{D}\otimes S^{(n-|\tau|,\tau)}. \tag*{(61)} ( R n , k ) D ≅ ∣ τ ∣ ≤ D τ [ n ] a partition ⨁ ( S τ ( A + ) ) D ⊗ S ( n − ∣ τ ∣ , τ ) . (61) Thus the multigraded stable multiplicity of the label τ [ n ] = ( n − ∣ τ ∣ , τ ) \tau[n]=(n-|\tau|,\tau) τ [ n ] = ( n − ∣ τ ∣ , τ ) is the positive plethysm
s τ [ H ( q ) − 1 ] . (62) s_{\tau}[H(\mathbf{q})-1]. \tag*{(62)} s τ [ H ( q ) − 1 ] . (62) For a fixed nonempty τ \tau τ and D ≥ ∣ τ ∣ D\geq|\tau| D ≥ ∣ τ ∣ , its degree-D D D multiplicity is already given by this formula when
n ≥ D + τ 1 . (63) n\geq D+\tau_{1}. \tag*{(63)} n ≥ D + τ 1 . (63) The term for τ = ∅ \tau=\varnothing τ = ∅ is the trivial label, which occurs only in degree zero.
Proof. Apply Theorem 9.9 to B = Sym V B=\operatorname{Sym} V B = Sym V . Corollary 9.10 covers every valid label when n ≥ 2 D − 1 n\geq2D-1 n ≥ 2 D − 1 . The same theorem gives the sharper individual range for every D ≥ 1 D\geq1 D ≥ 1 . Schur functor characters give (9.9), in each multidegree. □
For example, the tails τ = ( 1 ) , ( 2 ) , ( 1 , 1 ) \tau=(1),(2),(1,1) τ = ( 1 ) , ( 2 ) , ( 1 , 1 ) give A + A_{+} A + , Sym 2 ( A + ) \operatorname{Sym}^{2}(A_{+}) Sym 2 ( A + ) and ⋀ 2 ( A + ) \bigwedge^{2}(A_{+}) ⋀ 2 ( A + ) , respectively, before expansion in row Schur functions.
Finite jets and their maps
The stable multiplicities depend on the graded augmentation space. The compatible parabolic algebras retain more: they recover multiplication, and also every map of the finite jets. Here the algebras need not be graded. Put J n ( B ) = B / m B n J_n(B)=B/\mathfrak{m}_{B}^{n} J n ( B ) = B / m B n .
Theorem 9.12 (Full faithfulness of finite jets). Let n ≥ 2 n\geq2 n ≥ 2 , let K K K be a field in which n ! n! n ! is invertible, and let B , C B,C B , C be augmented commutative K K K -algebras. There is a natural bijection
Hom S n -alg ( Q n ( B ) , Q n ( C ) ) ≅ Hom aug-alg ( J n ( B ) , J n ( C ) ) . (64) \operatorname{Hom}_{S_n\text{-alg}}(Q_n(B),Q_n(C))\cong\operatorname{Hom}_{\text{aug-alg}}(J_n(B),J_n(C)). \tag*{(64)} Hom S n -alg ( Q n ( B ) , Q n ( C )) ≅ Hom aug-alg ( J n ( B ) , J n ( C )) . (64) Maps on the left are unital; they automatically preserve augmentation. Thus Q n Q_n Q n is fully faithful on augmented algebras whose augmentation ideals have n n n th power zero. If the two jets are finite dimensional, (9.11) is an isomorphism of affine schemes of homomorphisms, including the loci of isomorphisms.
Proof. By Theorem 9.1 , the first factor map B → Q n ( B ) B\to Q_n(B) B → Q n ( B ) kills m B n \mathfrak{m}_{B}^{n} m B n . The same is true in every factor, by permutation of labels. Consequently
Q n ( B ) ≅ Q n ( J n ( B ) ) . (65) Q_n(B)\cong Q_n(J_n(B)). \tag*{(65)} Q n ( B ) ≅ Q n ( J n ( B )) . (65)
Indeed, the factor maps give the inverse to the evident quotient map: the invariant ideal is generated by global traces of augmentation elements, and these traces vanish under the factor maps.
Reynolds averaging and the definition give Q n ( B ) S n = K Q_n(B)^{S_n}=K Q n ( B ) S n = K . Its Reynolds projection therefore equals its augmentation, since augmentation commutes with averaging and is the identity on K K K . An equivariant unital algebra map commutes with this projection and hence preserves augmentation. Restrict it to the invariants of S n − 1 S_{n-1} S n − 1 fixing the first factor. The first-parabolic theorem gives an augmented map J n ( B ) → J n ( C ) J_n(B)\to J_n(C) J n ( B ) → J n ( C ) . Conversely, apply Q n Q_n Q n to any such map and use (9.12). These operations are inverse: Q n ( B ) Q_n(B) Q n ( B ) is generated by the S n S_n S n -orbit of its first factor. They are natural in both algebras.
For finite-dimensional jets, repeat this argument after every commutative K K K -algebra extension R R R . Tensor powers, the invariant-generated ideal, and Reynolds idempotents commute with base change; the first parabolic is J n ( B ) ⊗ K R J_n(B) \otimes_K R J n ( B ) ⊗ K R . The same inverse constructions therefore work over nonreduced R R R as well. Yoneda’s lemma identifies the affine homomorphism schemes. Applying the construction to a map and its inverse identifies the isomorphism loci. □ \square □
Applying augmentation to the last tensor factor sends invariant traces on n n n factors to those on n − 1 n-1 n − 1 factors, defining Q n ( B ) → Q n − 1 ( B ) Q_n(B) \to Q_{n-1}(B) Q n ( B ) → Q n − 1 ( B ) . On first parabolics this is the quotient map B / m B n → B / m B n − 1 B/\mathfrak{m}_B^n \to B/\mathfrak{m}_B^{n-1} B / m B n → B / m B n − 1 .
Corollary 9.13 (Formal towers). In characteristic zero, compatible equivariant algebra maps between the towers { Q n ( B ) } n ≥ 2 \{Q_n(B)\}_{n\geq2} { Q n ( B ) } n ≥ 2 and { Q n ( C ) } n ≥ 2 \{Q_n(C)\}_{n\geq2} { Q n ( C ) } n ≥ 2 are exactly the continuous augmented algebra maps B ^ m B → C ^ m C \widehat{B}_{\mathfrak{m}_B} \to\widehat{C}_{\mathfrak{m}_C} B m B → C m C . The transition maps augment the last tensor factor.
Proof. The theorem identifies compatible tower maps with compatible maps of the jet quotients. Taking inverse limits gives a continuous augmented map of completions. Conversely, such a map takes the closure of m B n \mathfrak{m}_B^n m B n into the closure of m C n \mathfrak{m}_C^n m C n : approximate elements of the former by sums of products of n n n augmentation elements and use continuity. It therefore induces the required maps at every finite level. □ \square □
For example, take B 1 = K [ x ] / ( x 3 ) B_1 = K[x]/(x^3) B 1 = K [ x ] / ( x 3 ) , deg x = 1 \deg x = 1 deg x = 1 , \qquad B_2 = K[u,v]/(u,v)^2$, deg u = 1 \deg u = 1 deg u = 1 , deg v = 2 \deg v = 2 deg v = 2 . Their graded augmentation spaces are isomorphic: each has one basis vector in degrees one and two. After choosing this identification, Theorem 9.9 gives the same stable multiplicity spaces. Their first parabolics at n = 2 n=2 n = 2 differ, since dim J 2 ( B 1 ) = 2 \dim J_2(B_1)=2 dim J 2 ( B 1 ) = 2 and dim J 2 ( B 2 ) = 3 \dim J_2(B_2)=3 dim J 2 ( B 2 ) = 3 . Thus the stable representation and the parabolic algebra record different information.
Corollary 9.14 (Equivariant automorphisms of polynomial coinvariants). Let n ≥ 2 n\geq2 n ≥ 2 , let K K K have characteristic zero, and let dim V = k \dim V=k dim V = k . For R n , k = Q n ( Sym V ) R_{n,k}=Q_n(\operatorname{Sym} V) R n , k = Q n ( Sym V ) , the equivariant automorphism group scheme is the group of origin-fixing coordinate substitutions modulo degree n n n with invertible linear part. It is smooth of dimension
k ( ( k + n − 1 k ) − 1 ) , k\left(\binom{k+n-1}{k}-1\right), k ( ( k k + n − 1 ) − 1 ) ,
and its graded subgroup is GL ( V ) \operatorname{GL}(V) GL ( V ) .
Proof. An augmented endomorphism of Sym V / ( V ) n \operatorname{Sym} V/(V)^n Sym V / ( V ) n is specified freely by the image of V V V in ( V ) / ( V ) n (V)/(V)^n ( V ) / ( V ) n . It is invertible exactly when its linear part is invertible, as an inverse can then be constructed successively by degree. Its scheme is the product, as a variety, of GL ( V ) \operatorname{GL}(V) GL ( V ) and the vector space of higher coefficients. Count these coefficients and apply Theorem 9.12. □ \square □
Stable characters in every number of alphabets
The full character is the Euler character of the moment complex once positive homology vanishes. The Schur resolution controls its earliest possible degrees, and idempotent specialization excludes their boundary. Quadratic exactness gives the complementary label bound. These two inputs leave a short list of first-homology shapes; bivariate and hook exactness remove the routine cases, and one explicit cubic multiplication resolves the last one. The finer trace bounds are retained separately in Appendix D.
Quadratic characters and exactness
Let n ≥ 2 n\geq2 n ≥ 2 , V = K k V=K^k V = K k , and let W n W_n W n carry its invariant nondegenerate symmetric form. Removing the linear moments leaves P 0 = Sym ( V ⊗ W n ) P^0=\operatorname{Sym}(V\otimes W_n) P 0 = Sym ( V ⊗ W n ) . The quadratic moment map identifies Sym 2 V \operatorname{Sym}^2 V Sym 2 V with a subspace of P 2 0 P^0_2 P 2 0 ; put
Q n , k = P 0 / ( Sym 2 V ) . Q_{n,k}=P^0/(\operatorname{Sym}^2 V). Q n , k = P 0 / ( Sym 2 V ) .
The orthogonal Littlewood complex computes this quotient in every rank. Its modification rule also makes the quadratic Koszul complex exact in positive homological degrees through total degree n n n . These two facts give exact initial characters and provide the homological input for the character range proved below.
For a graded GL ( V ) × S n \operatorname{GL}(V) \times S_n GL ( V ) × S n character, let Π n \Pi_n Π n retain just the row Schur terms indexed by partitions ρ \rho ρ with ρ 1 ′ + ρ 2 ′ ≤ n − 1 \rho'_1+\rho'_2 \le n-1 ρ 1 ′ + ρ 2 ′ ≤ n − 1 . Here primes denote conjugate partitions.
Theorem 10.1 (Orthogonal quadratic character; Sam–Snowden–Weyman). In the degree-completed representation ring,
[ Q n , k ] = Π n ( [ Sym ( V ⊗ W n ) ] ∑ j ≥ 0 ( − 1 ) j [ Λ j ( Sym 2 V ) ] ) (66) [Q_{n,k}] = \Pi_n\left([ \operatorname{Sym}(V \otimes W_n)] \sum_{j\geq0}(-1)^j[\Lambda^j(\operatorname{Sym}^2 V)]\right) \tag*{(66)} [ Q n , k ] = Π n ( [ Sym ( V ⊗ W n )] j ≥ 0 ∑ ( − 1 ) j [ Λ j ( Sym 2 V )] ) (66) where Λ j ( Sym 2 V ) \Lambda^j(\operatorname{Sym}^2 V) Λ j ( Sym 2 V ) has polynomial degree 2 j 2j 2 j and trivial S n S_n S n -action. Equivalently,
Q n , k ≃ ⨁ ℓ ( ρ ) ≤ k ρ 1 ′ + ρ 2 ′ ≤ n − 1 S ρ V ⊗ S [ ρ ] W n ∣ S n , (67) Q_{n,k} \simeq\bigoplus_{\substack{\ell(\rho)\leq k\\ \rho'_1+\rho'_2\leq n-1}} S_\rho V \otimes S_{[\rho]}W_n\big|_{S_n}, \tag*{(67)} Q n , k ≃ ℓ ( ρ ) ≤ k ρ 1 ′ + ρ 2 ′ ≤ n − 1 ⨁ S ρ V ⊗ S [ ρ ] W n S n , (67) where S [ ρ ] S_{[\rho]} S [ ρ ] is the orthogonal Schur module.
Proof. The Koszul complex of the quadratic moment subspace is Sym ( V ⊗ W n ) ⊗ Λ ∙ ( Sym 2 V ) \operatorname{Sym}(V \otimes W_n)\otimes\Lambda^\bullet(\operatorname{Sym}^2 V) Sym ( V ⊗ W n ) ⊗ Λ ∙ ( Sym 2 V ) . Its row-ρ \rho ρ multiplicity complex is the orthogonal Littlewood complex. The orthogonal Littlewood theorem identifies its zeroth homology with S [ ρ ] W n S_{[\rho]}W_n S [ ρ ] W n for ρ 1 ′ + ρ 2 ′ ≤ dim W n \rho'_1+\rho'_2\leq\dim W_n ρ 1 ′ + ρ 2 ′ ≤ dim W n , and zero otherwise; moreover all positive homology vanishes for these admissible partitions [31 §4.2, Proposition 4.1 and Theorem 4.4] . Taking Euler characteristics only on those row components proves (66) ; taking zeroth homology gives (67) . Restriction from the full orthogonal group to S n ⊂ O ( W n ) S_n\subset O(W_n) S n ⊂ O ( W n ) retains all label multiplicities. The construction and the identities are defined over Q \mathbb{Q} Q , so characteristic-zero scalar extension gives the assertion over K K K . □
Corollary 10.2 (Three initial layers of every label). Let μ ⊢ n \mu\vdash n μ ⊢ n , and let δ μ ( k ) \delta_\mu(k) δ μ ( k ) be the least degree in which S μ S^\mu S μ occurs in the ambient polynomial ring P n , k P_{n,k} P n , k . For a row partition ρ ⊢ D \rho\vdash D ρ ⊢ D with ℓ ( ρ ) ≤ k \ell(\rho)\leq k ℓ ( ρ ) ≤ k , the joint ( ρ , μ ) (\rho,\mu) ( ρ , μ ) multiplicities in R n , k [ D ] R_{n,k}[D] R n , k [ D ] and Q n , k [ D ] Q_{n,k}[D] Q n , k [ D ] agree whenever
D − δ μ ( k ) ≤ 2 or ρ 1 ≤ 2. D-\delta_\mu(k)\leq2 \qquad\text{or}\qquad\rho_1\leq2. D − δ μ ( k ) ≤ 2 or ρ 1 ≤ 2. Here the empty partition has first part zero. Thus (66) computes the complete character in each of the three degrees
δ μ ( k ) , δ μ ( k ) + 1 , δ μ ( k ) + 2 , \delta_\mu(k),\qquad\delta_\mu(k)+1,\qquad\delta_\mu(k)+2, δ μ ( k ) , δ μ ( k ) + 1 , δ μ ( k ) + 2 , for every label and every number of alphabets, in every rank n ≥ 2 n\geq2 n ≥ 2 . It also computes the entire two-column row sector in all degrees. For n = 1 n=1 n = 1 , the quotient is K K K and all nonconstant components vanish.
Proof. A polarized power sum of degree d d d has row type Sym d V \operatorname{Sym}^d V Sym d V and trivial label type. To affect the joint ( ρ , μ ) (\rho,\mu) ( ρ , μ ) component, it must multiply a polynomial component of type ( σ , μ ) (\sigma,\mu) ( σ , μ ) in degree D − d D-d D − d , with ρ / σ \rho/\sigma ρ / σ a horizontal d d d -strip. Consequently
d ≤ ρ 1 , D − d ≥ δ μ ( k ) . d\leq\rho_1,\qquad D-d\geq\delta_\mu(k). d ≤ ρ 1 , D − d ≥ δ μ ( k ) .
Either hypothesis of the corollary excludes every d ≥ 3 d\geq3 d ≥ 3 . Thus higher moments have zero image in the indicated component, even before quotienting by the linear and quadratic moments. The quadratic quotient therefore has the full coinvariant character there, as given by Theorem 10.1. □
Theorem 11.1 computes δ μ ( k ) \delta_\mu(k) δ μ ( k ) and generates its entire first layer. The corollary computes the next two full harmonic layers as well; these degrees may lie outside the uniform stable range. Operator generation in the additional layers is governed by the criteria in Section 12.
Theorem 10.3 (Quadratic Koszul exactness through degree n n n ). For every n ≥ 2 n\geq2 n ≥ 2 , every finite-dimensional row space V V V , and every 0 ≤ D ≤ n 0\leq D\leq n 0 ≤ D ≤ n ,
H j ( Sym ( V ⊗ W n ) ⊗ Λ ∙ ( Sym 2 V ) ) D = 0 ( j > 0 ) , (68) H_j\left(\operatorname{Sym}(V\otimes W_n)\otimes\Lambda^\bullet(\operatorname{Sym}^2V)\right)_D=0 \qquad(j>0), \tag*{(68)} H j ( Sym ( V ⊗ W n ) ⊗ Λ ∙ ( Sym 2 V ) ) D = 0 ( j > 0 ) , (68) where the exterior generators have total degree two and the differential is multiplication by the quadratic moments.
Proof. Put m = n − 1 m=n-1 m = n − 1 and decompose the complex by its row partitions λ ⊢ D \lambda\vdash D λ ⊢ D . Its row-λ \lambda λ multiplicity complex is the orthogonal Littlewood complex. Write c i = λ i ′ c_i=\lambda_i' c i = λ i ′ for the column lengths. If c 1 + c 2 ≤ m c_1+c_2\le m c 1 + c 2 ≤ m , the admissible case of the Littlewood theorem gives zero positive homology.
For an inadmissible partition of size at most m + 1 m+1 m + 1 , the inequalities
m < c 1 + c 2 ≤ ∣ λ ∣ ≤ m + 1 m<c_1+c_2\le|\lambda|\le m+1 m < c 1 + c 2 ≤ ∣ λ ∣ ≤ m + 1
force c 1 + c 2 = m + 1 c_1+c_2=m+1 c 1 + c 2 = m + 1 . In the orthogonal modification rule, the dot action of the first reflection is
s 0 ⋅ ( c 1 , c 2 , c 3 , … ) = ( m + 1 − c 2 , m + 1 − c 1 , c 3 , … ) . s_0\cdot(c_1,c_2,c_3,\ldots)=(m+1-c_2,m+1-c_1,c_3,\ldots). s 0 ⋅ ( c 1 , c 2 , c 3 , … ) = ( m + 1 − c 2 , m + 1 − c 1 , c 3 , … ) .
It fixes this column sequence. The partition is therefore singular, and its Littlewood complex has zero homology in every degree [31 §4.4 and Theorem 4.4] . Every row partition of size D ≤ n D\le n D ≤ n is either admissible or singular, which proves (68) .
Consequently, in total degrees at most n n n , the projection Π n \Pi_n Π n in (66) can be omitted: the ordinary Koszul Euler character already equals [ Q n , k ] [Q_{n,k}] [ Q n , k ] . Every discarded row is singular and has Euler character zero in those degrees. This exactness holds uniformly in the row dimension because it is proved one Schur functor at a time; it requires no global regular-sequence assertion for the quadratic moments.
The full moment complex.
Let V = K k V=K^k V = K k and P = P n , k = Sym ( V ⊗ K n ) P=P_{n,k}=\operatorname{Sym}(V\otimes K^n) P = P n , k = Sym ( V ⊗ K n ) . Write p α = ∑ i = 1 n X α i p_\alpha=\sum_{i=1}^n X_\alpha^i p α = ∑ i = 1 n X α i for the polarized moment, and give Sym j V \operatorname{Sym}^j V Sym j V weight j j j in U D = ⨁ j = 1 D Sym j V U_D=\bigoplus_{j=1}^D\operatorname{Sym}^j V U D = ⨁ j = 1 D Sym j V . The moment Koszul complex is
K n , k , D = P ⊗ ⋀ U D , ∂ ( e α ) = p α . \mathcal{K}_{n,k,D}=P\otimes\bigwedge U_D,\qquad\partial(e_\alpha)=p_\alpha. K n , k , D = P ⊗ ⋀ U D , ∂ ( e α ) = p α .
Here e α e_\alpha e α is the exterior generator corresponding to the monomial of row weight α \alpha α in Sym ∣ α ∣ V \operatorname{Sym}^{|\alpha|}V Sym ∣ α ∣ V . Its total grading counts both polynomial degree and the assigned weights of exterior generators. Its zeroth homology agrees with R n , k R_{n,k} R n , k in total degrees at most D D D .
Our target is positive-degree exactness for D ≥ 3 D\ge3 D ≥ 3 , n ≥ max { D , 2 D − 6 } n\ge\max\{D,2D-6\} n ≥ max { D , 2 D − 6 } . Throughout this proof D ≤ n D\le n D ≤ n , λ \lambda λ is a row partition, and μ \mu μ a label partition. The Schur relation and quadratic reduction eliminate all homological degrees at least two. At n = 2 D − 5 n=2D-5 n = 2 D − 5 the only remaining source has at most two rows. At n = 2 D − 6 n=2D-6 n = 2 D − 6 , three labels remain; their only component not covered by bivariate or row-hook exactness is the multiplicity-one cubic source in Lemma A.1.
Parabolic detection of low-degree homology.
Lemma 10.4 (Parabolic detection of low-degree homology). Fix 3 ≤ D ≤ n 3\le D\le n 3 ≤ D ≤ n . In positive homological degree, every label partition μ \mu μ occurring in H j ( K n , k , D ) D H_j(\mathcal{K}_{n,k,D})_D H j ( K n , k , D ) D satisfies
n − D + 3 j ≤ μ 1 ≤ D − 2. (69) n-D+3j\le\mu_1\le D-2. \tag*{(69)} n − D + 3 j ≤ μ 1 ≤ D − 2. (69) In any total degree, if the moments through order a ≥ 1 a\ge1 a ≥ 1 form a regular sequence in P P P , one also has the lower bound μ 1 ≥ n − D + ( a + 1 ) j \mu_1\ge n-D+(a+1)j μ 1 ≥ n − D + ( a + 1 ) j .
Proof. Remove the linear moments and compute quadratic Koszul homology first. By Theorem 10.3, in total degrees at most n n n the remaining complex is
Q n , k ⊗ ⋀ ∙ ( ⨁ d ≥ 3 Sym d V ) , Q n , k = Sym ( V ⊗ W n ) / ( Sym 2 V ) . Q_{n,k}\otimes\bigwedge^\bullet\left(\bigoplus_{d\ge3}\operatorname{Sym}^d V\right),\qquad Q_{n,k}=\operatorname{Sym}(V\otimes W_n)/(\operatorname{Sym}^2V). Q n , k ⊗ ⋀ ∙ ( d ≥ 3 ⨁ Sym d V ) , Q n , k = Sym ( V ⊗ W n ) / ( Sym 2 V ) .
A degree-b b b coefficient has first label row at least n − b n-b n − b , because each monomial occupies at most b b b labels; the bound passes to quotients. The exterior factors consume at least 3 j 3j 3 j degrees, giving the lower bound. If moments through order a a a are regular, quotient by them first and use the minimum remaining exterior weight a + 1 a+1 a + 1 .
The trivial label has no positive moment homology through degree n n n , by Theorem 9.7. For any other label, that same theorem gives the stronger restriction
μ j + 1 ≥ n − D + j + 1. (70) \mu_{j+1}\ge n-D+j+1. \tag*{(70)} μ j + 1 ≥ n − D + j + 1. (70)
In particular μ 2 ≥ n − D + 2 \mu_2 \ge n-D+2 μ 2 ≥ n − D + 2 , so μ 1 ≤ D − 2 \mu_1 \le D-2 μ 1 ≤ D − 2 . More generally, summing the first j + 1 j+1 j + 1 label rows yields
μ 1 ≤ j D − ( j − 1 ) n − j ( j + 1 ) , n ≤ ( 1 + 1 / j ) D − j − 4 , D ≥ j ( j + 4 ) . \mu_1 \le jD-(j-1)n-j(j+1), \qquad n \le(1+1/j)D-j-4, \qquad D \ge j(j+4). μ 1 ≤ j D − ( j − 1 ) n − j ( j + 1 ) , n ≤ ( 1 + 1/ j ) D − j − 4 , D ≥ j ( j + 4 ) .
The latter two inequalities combine this estimate with the quadratic lower bound and D ≤ n D \le n D ≤ n . For example, second homology is absent in this range whenever D < 12 D < 12 D < 12 . □
Proposition 10.5 (Intersection support). If S λ V ⊗ S μ \mathbb{S}_{\lambda}V \otimes\mathbb{S}^{\mu} S λ V ⊗ S μ occurs in H i H_i H i in degree D D D with i ≥ 1 i \ge1 i ≥ 1 , then, with ρ i , d i \rho_i,d_i ρ i , d i from (9.4),
∣ λ ∩ ρ i ∣ ≥ 2 d i − D , D + λ 1 ≥ n + μ 1 − μ i + 1 + i + 1. |\lambda\cap\rho_i| \ge2d_i-D, \qquad D+\lambda_1 \ge n+\mu_1-\mu_{i+1}+i+1. ∣ λ ∩ ρ i ∣ ≥ 2 d i − D , D + λ 1 ≥ n + μ 1 − μ i + 1 + i + 1. Here the intersection is that of Young diagrams.
Proof. Inside S ρ i ( A + ) D \mathbb{S}_{\rho_i}(A_+)_D S ρ i ( A + ) D , at most D − d i D-d_i D − d i inputs have degree at least two. Splitting A + = V ⊕ Sym ≥ 2 V A_+ = V \oplus\operatorname{Sym}^{\ge2} V A + = V ⊕ Sym ≥ 2 V gives a linear-input shape ν ⊆ ρ i \nu\subseteq\rho_i ν ⊆ ρ i with ∣ ν ∣ ≥ 2 d i − D |\nu| \ge2d_i-D ∣ ν ∣ ≥ 2 d i − D . Every resulting row contains ν \nu ν , proving the first bound; also ν 1 ≥ ( ρ i ) 1 − ( D − d i ) \nu_1 \ge(\rho_i)_1-(D-d_i) ν 1 ≥ ( ρ i ) 1 − ( D − d i ) and ( ρ i ) 1 = μ 1 + 1 (\rho_i)_1=\mu_1+1 ( ρ i ) 1 = μ 1 + 1 . □
Appendix D gives further row-sensitive trace bounds and detects their closed upper boundary. They are not needed for the uniform character theorem below.
Separation through degree n n n in two alphabets
A Hilbert-series identity alone need not eliminate cancelling homology. Here multiplication already gives a free cover. Equality of dimensions then forces its kernel to begin in a higher degree.
Lemma 10.6 (Truncated freeness from a Hilbert identity). Let S S S be a connected positively graded commutative or supercommutative algebra over a field, and let M M M be a nonnegatively graded S S S -module. Assume their homogeneous pieces are finite dimensional, and put R = M / S + M R=M/S_+M R = M / S + M . If
H M ( t ) = H S ( t ) H R ( t ) ( m o d t N + 1 ) , H_M(t)=H_S(t)H_R(t) \pmod{t^{N+1}}, H M ( t ) = H S ( t ) H R ( t ) ( mod t N + 1 ) , then every homogeneous section R → M R \to M R → M induces an isomorphism
( S ⊗ R ) ≤ N ⟶ M ≤ N . (S \otimes R)_{\le N} \longrightarrow M_{\le N}. ( S ⊗ R ) ≤ N ⟶ M ≤ N . Moreover, Tor j S ( M , K ) d = 0 \operatorname{Tor}^{S}_j(M,K)_d=0 Tor j S ( M , K ) d = 0 for j > 0 j>0 j > 0 and d ≤ N d \le N d ≤ N . For superalgebras use ordinary dimensions and parity-preserving sections. Additional positive multigradings may be retained.
Proof. The multiplication map is surjective by degree induction: after subtracting a lift of the quotient class, an element belongs to S + M S_+M S + M and is a sum of positive-degree coefficients times elements of smaller degree. The Hilbert identity makes the source and target have equal dimensions through degree N N N . Its kernel L L L therefore has no homogeneous piece in those degrees. Choose a graded free resolution of L L L using generators only in degrees greater than N N N , and splice it with $0 \to L \to S \otimes R \to M \to 0$. Every positive term of the resulting free resolution of M M M is zero in degrees at most N N N . Tensoring with K K K proves the Tor assertion. All arguments respect the additional gradings. □
Theorem 10.7 (Bivariate separation through degree n n n ). Multiplication induces a canonical G L 2 × S n \mathrm{GL}_2 \times S_n GL 2 × S n -equivariant isomorphism
( P n , 2 S n ⊗ H n , 2 ) ≤ n ⟶ ( P n , 2 ) ≤ n . (P^{S_n}_{n,2} \otimes\mathcal{H}_{n,2})_{\le n} \longrightarrow(P_{n,2})_{\le n}. ( P n , 2 S n ⊗ H n , 2 ) ≤ n ⟶ ( P n , 2 ) ≤ n . Every positive moment Koszul homology group vanishes in internal degrees at most n n n . Consequently the full bivariate S n S_n S n character is given, in bidegree ( a , b ) (a,b) ( a , b ) whenever a + b ≤ n a+b \le n a + b ≤ n , by
tr ( σ ∣ ( R n , 2 ) a , b ) = [ q a t b ] ∏ i , j ≥ 0 i + j > 0 ( 1 − q i t j ) ∏ m ≥ 1 ( ( 1 − q m ) ( 1 − t m ) ) − X m ( σ ) . (71) \operatorname{tr}(\sigma\mid(R_{n,2})_{a,b})=[q^a t^b]\prod_{\substack{i,j\ge0\\i+j>0}}(1-q^i t^j)\prod_{m\ge1}\left((1-q^m)(1-t^m)\right)^{-X_m(\sigma)}. \tag*{(71)} tr ( σ ∣ ( R n , 2 ) a , b ) = [ q a t b ] i , j ≥ 0 i + j > 0 ∏ ( 1 − q i t j ) m ≥ 1 ∏ ( ( 1 − q m ) ( 1 − t m ) ) − X m ( σ ) . (71) Here X m ( σ ) X_m(\sigma) X m ( σ ) counts the m m m -cycles of σ \sigma σ . Equivalently, every row partition of length at most two has zero positive moment homology through degree n n n , in every row dimension.
Proof. For every fixed bidegree ( a , b ) (a,b) ( a , b ) , with D = a + b D=a+b D = a + b , Wang’s dimension theorem [36 Theorem 5.1] gives one polynomial in n n n equal to dim ( R n , 2 ) a , b \dim(R_{n,2})_{a,b} dim ( R n , 2 ) a , b throughout n ≥ D n\ge D n ≥ D . The Euler dimension of the moment complex is also polynomial in n n n . Corollary 9.10 gives eventual acyclicity, for example for n ≥ 2 D − 1 n\ge2D-1 n ≥ 2 D − 1 . The actual and Euler dimension polynomials therefore agree at all sufficiently large integers and are identical. Degrees zero, one, and two follow from the independent linear and quadratic moments. Hence the quotient has the Euler Hilbert series in every total degree at most n n n .
Let
F n = Sym ( ⨁ j = 1 n Sym j K 2 ) F_n=\operatorname{Sym}\left(\bigoplus_{j=1}^{n}\operatorname{Sym}^{j}K^{2}\right) F n = Sym ( j = 1 ⨁ n Sym j K 2 )
act on P n , 2 P_{n,2} P n , 2 by the formal moment evaluation. Polarized moments of degrees at most n n n generate the invariant algebra, so R n , 2 = P n , 2 / ( F n ) + P n , 2 R_{n,2}=P_{n,2}/(F_n)_+P_{n,2} R n , 2 = P n , 2 / ( F n ) + P n , 2 . The dimension identity just proved is H P n , 2 = H F n H R n , 2 H_{P_{n,2}}=H_{F_n}H_{R_{n,2}} H P n , 2 = H F n H R n , 2 through degree n n n . Lemma 10.6 gives freeness and positive Koszul acyclicity in this range.
The fundamental trace presentation has its first relation in degree n + 1 n+1 n + 1 : each relation uses n + 1 n+1 n + 1 positive inputs [16 ] . Thus F n F_n F n maps isomorphically onto the actual invariant algebra in degrees at most n n n . Choose the canonical harmonic section of R n , 2 R_{n,2} R n , 2 . The free cover becomes exactly the multiplication map in the statement.
Taking equivariant characters of the moment complex gives (71) . Each m m m -cycle contributes ( 1 − q m ) − 1 ( 1 − t m ) − 1 (1-q^{m})^{-1}(1-t^{m})^{-1} ( 1 − q m ) − 1 ( 1 − t m ) − 1 to the polynomial trace, and each invariant moment contributes its Euler factor. Finally, the complex is a polynomial functor of the row space. In characteristic zero its homology decomposes into Schur functors; every partition of length at most two is detected on K 2 K^{2} K 2 .
Row-hook exactness
The Fields character has a finite principal-specialization form that records both the exactness range and its first obstruction. Write ( q ; q ) m = ∏ i = 1 m ( 1 − q i ) (q;q)_m=\prod_{i=1}^{m}(1-q^i) ( q ; q ) m = ∏ i = 1 m ( 1 − q i ) and ( q ; q ) ∞ = ∏ i ≥ 1 ( 1 − q i ) (q;q)_\infty=\prod_{i\ge1}(1-q^i) ( q ; q ) ∞ = ∏ i ≥ 1 ( 1 − q i ) . Let Z Z Z be the Frobenius alphabet for S n S_n S n and set X = Z / ( 1 − q ) X=Z/(1-q) X = Z / ( 1 − q ) in plethystic notation. This means p d [ X ] = p d [ Z ] / ( 1 − q d ) p_d[X]=p_d[Z]/(1-q^d) p d [ X ] = p d [ Z ] / ( 1 − q d ) .
Let H n ( q , u ; Z ) \mathcal{H}_n(q,u;Z) H n ( q , u ; Z ) be the Frobenius series of the ( 1 , 1 ) (1,1) ( 1 , 1 ) bosonic–fermionic coinvariant ring: q q q records polynomial degree and u u u exterior degree. Let E n \mathcal{E}_n E n be the moment Euler character after evaluation on a row superspace of dimension ( 1 ∣ 1 ) (1|1) ( 1∣1 ) .
Lemma 10.8 (Finite Fields identity and its first discrepancy). For 0 ≤ b ≤ n 0\le b\le n 0 ≤ b ≤ n ,
[ u b ] H n = ∑ h = 0 b ( − 1 ) b − h ( q ; q ) n − h ( q ; q ) b − h e h [ X ] h n − h [ X ] . (72) [u^b]\mathcal{H}_n=\sum_{h=0}^{b}(-1)^{b-h}\frac{(q;q)_{n-h}}{(q;q)_{b-h}}e_h[X]h_{n-h}[X]. \tag*{(72)} [ u b ] H n = h = 0 ∑ b ( − 1 ) b − h ( q ; q ) b − h ( q ; q ) n − h e h [ X ] h n − h [ X ] . (72)
The corresponding coefficient of E n \mathcal{E}_n E n is obtained by replacing ( q ; q ) n − h (q;q)_{n-h} ( q ; q ) n − h with ( q ; q ) ∞ (q;q)_\infty ( q ; q ) ∞ . Hence the two characters agree in every total degree at most n n n , and
[ q n − b + 1 u b ] ( H n − E n ) = e b [ Z ] h n − b [ Z ] ( 0 ≤ b ≤ n ) . (73) [q^{n-b+1}u^b](\mathcal{H}_n-\mathcal{E}_n)=e_b[Z]h_{n-b}[Z]\qquad(0\le b\le n). \tag*{(73)} [ q n − b + 1 u b ] ( H n − E n ) = e b [ Z ] h n − b [ Z ] ( 0 ≤ b ≤ n ) . (73)
Proof. The Fields theorem [27 Theorem 5.2 and equation (2.5)] gives
H n = ∑ μ ⊢ n s μ [ Z ] ∑ T ∈ SYT ( μ ) q maj ( T ) ∏ r = 1 des ( T ) ( 1 + u q − r ) . \mathcal{H}_n=\sum_{\mu\vdash n}s_\mu[Z]\sum_{T\in\operatorname{SYT}(\mu)}q^{\operatorname{maj}(T)}\prod_{r=1}^{\operatorname{des}(T)}(1+uq^{-r}). H n = μ ⊢ n ∑ s μ [ Z ] T ∈ SYT ( μ ) ∑ q maj ( T ) r = 1 ∏ des ( T ) ( 1 + u q − r ) .
Put
A n ( t , q ; Z ) = ∑ μ ⊢ n s μ [ Z ] ∑ T ∈ SYT ( μ ) t des ( T ) q maj ( T ) . \mathcal{A}_n(t,q;Z)=\sum_{\mu\vdash n}s_\mu[Z]\sum_{T\in\operatorname{SYT}(\mu)}t^{\operatorname{des}(T)}q^{\operatorname{maj}(T)}. A n ( t , q ; Z ) = μ ⊢ n ∑ s μ [ Z ] T ∈ SYT ( μ ) ∑ t des ( T ) q maj ( T ) .
Principal specialization of Schur functions gives the character-valued Eulerian identity
A n ( t , q ; Z ) ∏ i = 0 n ( 1 − t q i ) = ∑ m ≥ 0 h n [ Z 1 − q m + 1 1 − q ] t m . \frac{\mathcal{A}_n(t,q;Z)}{\prod_{i=0}^{n}(1-tq^i)}=\sum_{m\ge0}h_n\left[Z\frac{1-q^{m+1}}{1-q}\right]t^m. ∏ i = 0 n ( 1 − t q i ) A n ( t , q ; Z ) = m ≥ 0 ∑ h n [ Z 1 − q 1 − q m + 1 ] t m .
One obtains it by standardizing semistandard tableaux and summing the weakly increasing entries with forced rises at descents. The standardization convention using comajor index gives the same identity, since tableau evacuation preserves the number of descents and interchanges major and comajor indices.
Expand
h n [ X − q m + 1 X ] = ∑ h = 0 n ( − 1 ) h q ( m + 1 ) h e h [ X ] h n − h [ X ] . h_n[X-q^{m+1}X]=\sum_{h=0}^{n}(-1)^h q^{(m+1)h}e_h[X]h_{n-h}[X]. h n [ X − q m + 1 X ] = h = 0 ∑ n ( − 1 ) h q ( m + 1 ) h e h [ X ] h n − h [ X ] .
Multiplication by the denominator above and evaluation at t = q − h t=q^{-h} t = q − h leave just one summand, yielding
A n ( q − h , q ; Z ) = q − ( h 2 ) ( q ; q ) h ( q ; q ) n − h e h [ X ] h n − h [ X ] . A_n(q^{-h},q;Z)=q^{-\binom{h}{2}}(q;q)_h(q;q)_{n-h}e_h[X]h_{n-h}[X]. A n ( q − h , q ; Z ) = q − ( 2 h ) ( q ; q ) h ( q ; q ) n − h e h [ X ] h n − h [ X ] .
The elementary-symmetric identity
e b ( q − 1 , … , q − d ) = ∑ h = 0 b ( − 1 ) b − h q ( h 2 ) ( q ; q ) h ( q ; q ) b − h q − h d e_b(q^{-1},\ldots,q^{-d})=\sum_{h=0}^{b}\frac{(-1)^{b-h}q^{\binom{h}{2}}}{(q;q)_h(q;q)_{b-h}}q^{-hd} e b ( q − 1 , … , q − d ) = h = 0 ∑ b ( q ; q ) h ( q ; q ) b − h ( − 1 ) b − h q ( 2 h ) q − h d
follows from the finite q q q -binomial theorem. Apply it with d = des ( T ) d=\operatorname{des}(T) d = des ( T ) and use the preceding evaluation to obtain (10.7).
The polynomial–exterior algebra has Frobenius series
∑ h = 0 n u h e h [ X ] h n − h [ X ] . \sum_{h=0}^{n}u^he_h[X]h_{n-h}[X]. h = 0 ∑ n u h e h [ X ] h n − h [ X ] .
The formal moment algebra contributes the reciprocal Euler factor
( q ; q ) ∞ ∏ a ≥ 0 ( 1 + u q a ) , ∏ a ≥ 0 ( 1 + u q a ) − 1 = ∑ j ≥ 0 ( − u ) j ( q ; q ) j . \frac{(q;q)_\infty}{\prod_{a\geq0}(1+uq^a)},\qquad
\prod_{a\geq0}(1+uq^a)^{-1}=\sum_{j\geq0}\frac{(-u)^j}{(q;q)_j}. ∏ a ≥ 0 ( 1 + u q a ) ( q ; q ) ∞ , a ≥ 0 ∏ ( 1 + u q a ) − 1 = j ≥ 0 ∑ ( q ; q ) j ( − u ) j .
This proves the asserted formula for E n \mathcal{E}_n E n .
The difference ( q ; q ) n − h − ( q ; q ) ∞ (q;q)_{n-h}-(q;q)_\infty ( q ; q ) n − h − ( q ; q ) ∞ begins in degree n − h + 1 n-h+1 n − h + 1 with coefficient one. In fixed exterior degree b b b , every term of the difference therefore has q q q -degree at least n − b + 1 n-b+1 n − b + 1 . At that first degree only h = b h=b h = b contributes, and the constant term of e b [ X ] h n − b [ X ] e_b[X]h_{n-b}[X] e b [ X ] h n − b [ X ] is e b [ Z ] h n − b [ Z ] e_b[Z]h_{n-b}[Z] e b [ Z ] h n − b [ Z ] . This proves both conclusions. □
Theorem 10.9 (Row-hook moment acyclicity). For every n n n , every row dimension k k k , every row hook λ ⊢ D \lambda\vdash D λ ⊢ D , and every D ≤ n D\leq n D ≤ n ,
Hom G L k ( S λ K k , H i ( K n , k , D ) D ) = 0 ( i > 0 ) . \operatorname{Hom}_{\mathrm{GL}_k}\left(S_\lambda K^k,H_i(K_{n,k,D})_D\right)=0\qquad(i>0). Hom GL k ( S λ K k , H i ( K n , k , D ) D ) = 0 ( i > 0 ) . Thus the full S n S_n S n character in every row-hook sector is its moment Euler character throughout D ≤ n D\leq n D ≤ n .
Proof. Evaluate the polynomial row functors on K 1 ∣ 1 K^{1|1} K 1∣1 . The coefficient algebra is P = K [ x 1 , … , x n ] ⊗ Λ ( θ 1 , … , θ n ) P=K[x_1,\ldots,x_n]\otimes\Lambda(\theta_1,\ldots,\theta_n) P = K [ x 1 , … , x n ] ⊗ Λ ( θ 1 , … , θ n ) . The free super moment algebra S S S has even generators of bidegree ( a , 0 ) (a,0) ( a , 0 ) for a ≥ 1 a\geq1 a ≥ 1 and odd generators of bidegree ( a , 1 ) (a,1) ( a , 1 ) for a ≥ 0 a\geq0 a ≥ 0 . They act by ∑ i x i a \sum_i x_i^a ∑ i x i a and ∑ i θ i x i a \sum_i\theta_i x_i^a ∑ i θ i x i a , respectively. Its ordinary Hilbert series is
H S ( q , u ) = ∏ a ≥ 0 ( 1 + u q a ) ( q ; q ) ∞ . H_S(q,u)=\frac{\prod_{a\geq0}(1+uq^a)}{(q;q)_\infty}. H S ( q , u ) = ( q ; q ) ∞ ∏ a ≥ 0 ( 1 + u q a ) .
Trace generation identifies P / S + P P/S_+P P / S + P with the superspace coinvariant ring. Lemma 10.8, after taking dimensions, says exactly that H P = H S H P / S + P H_P=H_SH_{P/S_+P} H P = H S H P / S + P through total degree n n n . Lemma 10.6 forces positive Tor to vanish there. The super Koszul resolution uses exterior powers on even generators and symmetric powers on odd generators, so it is precisely this evaluation of the moment complex.
To pass back to ordinary row spaces, express each homogeneous polynomial functor as ⨁ λ S λ ⊗ C λ \bigoplus_\lambda S_\lambda\otimes C_\lambda ⨁ λ S λ ⊗ C λ . In characteristic zero all maps act on the multiplicity spaces C λ C_\lambda C λ , and this decomposition can be evaluated on a super vector space with its signed tensor permutations. It commutes with homology. A nonempty Schur functor is nonzero on K 1 ∣ 1 K^{1|1} K 1∣1 exactly when its partition is a hook, and
s ( a , 1 b ) [ q ∣ u ] = ( q + u ) q a − 1 u b . s_{(a,1^b)}[q\mid u]=(q+u)q^{a-1}u^b. s ( a , 1 b ) [ q ∣ u ] = ( q + u ) q a − 1 u b .
Vanishing of the specialized homology therefore forces the vanishing of every row-hook multiplicity, including its entire S n S_n S n -module structure. □
Exact Euler characters and character polynomials.
Theorem 10.10 (Stable character polynomial formula). For D ≥ 3 D \ge3 D ≥ 3 and n ≥ max { D , 2 D − 6 } n \ge\max\{D,2D-6\} n ≥ max { D , 2 D − 6 } , the total-degree-D D D part of the moment Koszul complex is exact in positive degrees. The same conclusion holds for D = 0 , 1 D=0,1 D = 0 , 1 and n ≥ 1 n \ge1 n ≥ 1 , and for D = 2 D=2 D = 2 and n ≥ 2 n \ge2 n ≥ 2 . Consequently, if b = ( b 1 , … , b k ) \mathbf{b}=(b_1,\ldots,b_k) b = ( b 1 , … , b k ) has total size D D D , then for every σ ∈ S n \sigma\in S_n σ ∈ S n ,
tr ( σ ∣ ( R n , k ) b ) = [ q b ] ∏ 0 < ∣ α ∣ ≤ D ( 1 − q α ) ∏ j = 1 k ∏ a = 1 k ( 1 − q a j ) − X j ( σ ) . (74) \operatorname{tr}\left(\sigma\mid(R_{n,k})_{\mathbf{b}}\right)=[\mathbf{q}^{\mathbf{b}}]\prod_{0<|\alpha|\le D}(1-q^\alpha)\prod_{j=1}^{k}\prod_{a=1}^{k}(1-q_a^j)^{-X_j(\sigma)}. \tag*{(74)} tr ( σ ∣ ( R n , k ) b ) = [ q b ] 0 < ∣ α ∣ ≤ D ∏ ( 1 − q α ) j = 1 ∏ k a = 1 ∏ k ( 1 − q a j ) − X j ( σ ) . (74) where X j ( σ ) X_j(\sigma) X j ( σ ) is the number of j j j -cycles of σ \sigma σ . This gives the character polynomial explicitly in every multidegree and every number of alphabets.
For 3 ≤ D ≤ n 3 \le D \le n 3 ≤ D ≤ n , the same Koszul product computes every label coefficient S μ S^\mu S μ with μ 1 ≥ D − 1 \mu_1 \ge D-1 μ 1 ≥ D − 1 .
Proof. Suppose H j H_j H j is nonzero on label μ \mu μ in degree D ≤ n D \le n D ≤ n . Quadratic reduction gives μ 1 ≥ n − D + 3 j \mu_1 \ge n-D+3j μ 1 ≥ n − D + 3 j . Theorem 9.7 excludes the trivial label and gives μ 2 ≥ μ j + 1 ≥ n − D + j + 1 \mu_2 \ge\mu_{j+1} \ge n-D+j+1 μ 2 ≥ μ j + 1 ≥ n − D + j + 1 . Adding the first two rows therefore gives n ≤ 2 D − 4 j − 1 n \le2D-4j-1 n ≤ 2 D − 4 j − 1 . Thus j ≥ 2 j \ge2 j ≥ 2 is impossible when n ≥ 2 D − 6 n \ge2D-6 n ≥ 2 D − 6 . For j = 1 j=1 j = 1 the same inequality gives n ≤ 2 D − 5 n \le2D-5 n ≤ 2 D − 5 . This proves exactness for n ≥ max { D , 2 D − 4 } n \ge\max\{D,2D-4\} n ≥ max { D , 2 D − 4 } .
At n = 2 D − 5 ≥ D n=2D-5 \ge D n = 2 D − 5 ≥ D , the only possible label is ( D − 2 , D − 3 ) (D-2,D-3) ( D − 2 , D − 3 ) . Its first homology is a subquotient of S ( D − 1 ) ( A + ) D = Sym D − 2 V ⊗ Sym 2 V S_{(D-1)}(A_+)_D=\operatorname{Sym}^{D-2}V\otimes\operatorname{Sym}^2V S ( D − 1 ) ( A + ) D = Sym D − 2 V ⊗ Sym 2 V by the relation presentation. This has at most two rows and is removed by Theorem 10.7.
At n = 2 D − 6 ≥ D n=2D-6 \ge D n = 2 D − 6 ≥ D , put b = D − 3 ≥ 3 b=D-3 \ge3 b = D − 3 ≥ 3 . The bounds μ 1 ≥ b \mu_1 \ge b μ 1 ≥ b , μ 2 ≥ b − 1 \mu_2 \ge b-1 μ 2 ≥ b − 1 leave only
( b + 1 , b − 1 ) , ( b , b ) , ( b , b − 1 , 1 ) . (b+1,b-1),\qquad(b,b),\qquad(b,b-1,1). ( b + 1 , b − 1 ) , ( b , b ) , ( b , b − 1 , 1 ) . For ( b + 1 , b − 1 ) (b+1,b-1) ( b + 1 , b − 1 ) the first relation source has degree-D D D part Sym b + 1 V ⊗ Sym 2 V \operatorname{Sym}^{b+1}V\otimes\operatorname{Sym}^2V Sym b + 1 V ⊗ Sym 2 V , again with at most two rows. The other two labels first occur in coefficient degree b b b , with multiplicities Sym b V \operatorname{Sym}^bV Sym b V and S ( b − 1 , 1 ) V S_{(b-1,1)}V S ( b − 1 , 1 ) V , respectively. Positive invariant relations cannot change this first occurrence. Thus their degree-( b + 3 ) (b+3) ( b + 3 ) first chain groups after quadratic reduction are exactly Sym 3 V \operatorname{Sym}^3V Sym 3 V tensored with these spaces. The first has only two rows.
For the remaining label ( b , b − 1 , 1 ) (b,b-1,1) ( b , b − 1 , 1 ) , the relation source is S ( b + 1 , 1 ) ( A + ) b + 3 S_{(b+1,1)}(A_+)_{b+3} S ( b + 1 , 1 ) ( A + ) b + 3 , whose one-quadratic-input decomposition is
( S ( b , 1 ) V ⊕ Sym b + 1 V ) ⊗ Sym 2 V . \left(S_{(b,1)}V\oplus\operatorname{Sym}^{b+1}V\right)\otimes\operatorname{Sym}^2V. ( S ( b , 1 ) V ⊕ Sym b + 1 V ) ⊗ Sym 2 V . Pieri leaves only ( b + 1 , 1 , 1 ) (b+1,1,1) ( b + 1 , 1 , 1 ) and ( b , 2 , 1 ) (b,2,1) ( b , 2 , 1 ) after the two-row constituents are removed. Row-hook acyclicity (Theorem 10.9) removes the first. The second occurs once in the cubic source Sym 3 V ⊗ S ( b − 1 , 1 ) V \operatorname{Sym}^3V\otimes S_{(b-1,1)}V Sym 3 V ⊗ S ( b − 1 , 1 ) V , and its differential is nonzero by Lemma A.1. It is therefore injective there.
When dim V < 3 \dim V<3 dim V < 3 , both three-row components are absent.
For D = 3 , 4 , 5 D=3,4,5 D = 3 , 4 , 5 , n ≥ D n \ge D n ≥ D leaves only the earlier cases; the cubic witness is first needed at D = n = 6 D=n=6 D = n = 6 . Degrees zero and one use the linear sequence. In degree two the remaining map is the inclusion of the independent quadratic moments for n ≥ 2 n \ge2 n ≥ 2 . The alternating character of the exterior factor is ∏ 0 < ∣ α ∣ ≤ D ( 1 − q α ) \prod_{0<|\alpha|\le D}(1-q^\alpha) ∏ 0 < ∣ α ∣ ≤ D ( 1 − q α ) , and a j j j -cycle has polynomial trace ∏ a ( 1 − q a j ) − 1 \prod_a(1-q_a^j)^{-1} ∏ a ( 1 − q a j ) − 1 . Their product gives (10.9). The labelwise assertion follows from Lemma 10.4. □
Corollary 10.11 (Separation of invariants and harmonics). Multiplication induces an equivariant isomorphism
( P n , k S n ⊗ H n , k ) ≤ D ⟶ ∼ ( P n , k ) ≤ D \left(P_{n,k}^{S_n}\otimes\mathcal{H}_{n,k}\right)_{\le D}\overset{\sim}{\longrightarrow}(P_{n,k})_{\le D} ( P n , k S n ⊗ H n , k ) ≤ D ⟶ ∼ ( P n , k ) ≤ D when k ≤ 2 k\le2 k ≤ 2 and D < n D<n D < n , or for arbitrary k k k when D ≥ 3 D\ge3 D ≥ 3 and n ≥ max { D , 2 D − 6 } n\ge\max\{D,2D-6\} n ≥ max { D , 2 D − 6 } . For arbitrary k k k , its degree-D D D part is also an isomorphism on every row-hook isotype and every isotype whose row partition has length at most two, whenever D ≤ n D\le n D ≤ n .
Proof. The harmonic inclusion is an equivariant section of the coinvariant quotient. Apply Lemma 10.6 to the character identity, using the formal moment algebra, which agrees with
P n , k S n P_{n,k}^{S_n} P n , k S n through degree n n n . The bivariate case is Theorem 10.7 ; all smaller degrees satisfy the same bounds, with degrees zero, one and two immediate.
For the componentwise assertion, let L L L be the kernel of the free cover over the formal moment algebra S S S . Reduction modulo S + S_+ S + is an isomorphism, so L / S + L ≅ Tor 1 S ( P n , k , K ) L/S_+L \cong\operatorname{Tor}_1^S(P_{n,k},K) L / S + L ≅ Tor 1 S ( P n , k , K ) . A Littlewood–Richardson constituent contains the partition of each input factor. A kernel component of least degree among row hooks and partitions with at most two rows cannot lie in S + L S_+L S + L : its L L L factor would have lower degree and belong to one of those classes. It would therefore survive in Tor 1 \operatorname{Tor}_1 Tor 1 , contrary to Theorems 10.7 and 10.9 through degree n n n .
The coefficient extraction in (74) is finite. Expanding ( 1 − z ) − X (1-z)^{-X} ( 1 − z ) − X as ∑ v ≥ 0 ( X + v − 1 v ) z v \sum_{v\geq0}\binom{X+v-1}{v}z^v ∑ v ≥ 0 ( v X + v − 1 ) z v shows directly that it is a rational polynomial in the cycle counts, of weighted degree at most D D D when deg X j = j \deg X_j=j deg X j = j . At the identity it gives
dim ( R n , k ) b = [ q b ] ∏ 0 < ∣ α ∣ ≤ D ( 1 − q α ) ∏ a = 1 k ( 1 − q a ) − n . (75) \dim(R_{n,k})_{\mathbf b}=[\mathbf q^{\mathbf b}]\prod_{0<|\alpha|\leq D}(1-\mathbf q^\alpha)\prod_{a=1}^{k}(1-q_a)^{-n}. \tag*{(75)} dim ( R n , k ) b = [ q b ] 0 < ∣ α ∣ ≤ D ∏ ( 1 − q α ) a = 1 ∏ k ( 1 − q a ) − n . (75)
Its degree in n n n is exactly D D D , with leading coefficient 1 / ( b 1 ! ⋯ b k ! ) 1/(b_1!\cdots b_k!) 1/ ( b 1 ! ⋯ b k !) . For total degree D D D , the corresponding dimension is
[ t D ] ( 1 − t ) − k ( n − 1 ) ∏ j = 2 D ( 1 − t j ) ( k + j − 1 j ) , [t^D](1-t)^{-k(n-1)}\prod_{j=2}^{D}(1-t^j)^{\binom{k+j-1}{j}}, [ t D ] ( 1 − t ) − k ( n − 1 ) j = 2 ∏ D ( 1 − t j ) ( j k + j − 1 ) ,
with leading term k D n D / D ! k^D n^D/D! k D n D / D ! . For D ≥ 3 D\geq3 D ≥ 3 and n ≥ 2 D − 3 n\geq2D-3 n ≥ 2 D − 3 , Theorem 12.5 gives these same characters for the Vandermonde operator module.
This explicitly describes the eventual character and dimension polynomials asked for in [11 Problems 1.10 and 1.12] . The existence of eventual polynomiality and representation stability is due to Church–Ellenberg–Farb; see also the quantitative refinements of [2 , 8 ] . Outside this range, the same bounds locate every possible homological correction. The Euler product is the low-degree quotient proposed in [6 Section 4, equation (33)] . Bergeron’s conjecture asks for its validity throughout D ≤ n D\leq n D ≤ n .
Two-column homology, reflected cycles, and multiplication
Put W = W n W=W_n W = W n and m = dim W = n − 1 m=\dim W=n-1 m = dim W = n − 1 . For 0 ≤ q ≤ p 0\leq q\leq p 0 ≤ q ≤ p , let λ ( p , q ) = ( 2 q , 1 p − q ) \lambda(p,q)=(2^q,1^{p-q}) λ ( p , q ) = ( 2 q , 1 p − q ) , the partition with column heights p , q p,q p , q . Its degree is D = p + q D=p+q D = p + q . Write M p , q M_{p,q} M p , q for its stable row multiplicity space in full diagonal harmonics, with M 0 , 0 = K M_{0,0}=K M 0 , 0 = K . Only the invariant quadratic form survives in this sector; consequently its multiplicity spaces and moment homology carry an O ( W ) O(W) O ( W ) action extending their S n S_n S n action.
Theorem 10.12 (Two-term reduction in every degree). After eliminating the linear moments, the full moment Koszul complex in row λ ( p , q ) \lambda(p,q) λ ( p , q ) is
S λ ( p − 1 , q − 1 ) W → L S λ ( p , q ) W (76) S_{\lambda(p-1,q-1)}W \xrightarrow{L} S_{\lambda(p,q)}W \tag*{(76)} S λ ( p − 1 , q − 1 ) W L S λ ( p , q ) W (76) in homological degrees one and zero, with zero source when q = 0 q=0 q = 0 . The differential inserts the invariant symmetric form. There is no H j H_j H j for j ≥ 2 j\geq2 j ≥ 2 in any two-column row.
Proof. A moment of order at least three forces first row at least three in every Littlewood–Richardson constituent. The same is true of two or more quadratic Koszul generators, since ⋀ 2 Sym 2 V = S ( 3 , 1 ) V \bigwedge^2\operatorname{Sym}^2V=S_{(3,1)}V ⋀ 2 Sym 2 V = S ( 3 , 1 ) V and ⋀ j Sym 2 V \bigwedge^j\operatorname{Sym}^2V ⋀ j Sym 2 V is a summand of ⋀ 2 Sym 2 V ⊗ ⋀ j − 2 Sym 2 V \bigwedge^2\operatorname{Sym}^2V\otimes\bigwedge^{j-2}\operatorname{Sym}^2V ⋀ 2 Sym 2 V ⊗ ⋀ j − 2 Sym 2 V for j ≥ 2 j\geq2 j ≥ 2 . Thus only zero or one quadratic generator survives. Cauchy decomposition gives the target; horizontal Pieri removes one box from each column to give the source once. The differential is quadratic multiplication.
Let U = K 2 U=K^2 U = K 2 record two exterior copies of W W W , and put
E = ⋀ ( U ⊗ W ) , ω = ∑ i = 1 m θ i ∧ ϕ i , L = ω ∧ − , c = ∑ i = 1 m ∂ ϕ i ∂ θ i . \mathcal{E}=\bigwedge(U\otimes W),\qquad\omega=\sum_{i=1}^{m}\theta_i\wedge\phi_i,\qquad L=\omega\wedge-,\qquad c=\sum_{i=1}^{m}\partial_{\phi_i}\partial_{\theta_i}. E = ⋀ ( U ⊗ W ) , ω = i = 1 ∑ m θ i ∧ ϕ i , L = ω ∧ − , c = i = 1 ∑ m ∂ ϕ i ∂ θ i .
The displayed orthonormal notation denotes the operators defined by the form and its inverse, so no choice of an orthonormal basis over K K K is required. They satisfy [ c , L ] ∣ E e = ( m − e ) id [c,L]\vert_{\mathcal{E}_e}=(m-e)\operatorname{id} [ c , L ] ∣ E e = ( m − e ) id . For P e = ker ( c : E e → E e − 2 ) P_e=\ker(c:\mathcal{E}_e\to\mathcal{E}_{e-2}) P e = ker ( c : E e → E e − 2 ) , 0 ≤ e ≤ m 0\le e\le m 0 ≤ e ≤ m , the exterior Lefschetz decomposition gives
E = ⨁ e = 0 m ⨁ j = 0 m − e L j P e . (77) \mathcal{E}=\bigoplus_{e=0}^{m}\bigoplus_{j=0}^{m-e}L^jP_e. \tag*{(77)} E = e = 0 ⨁ m j = 0 ⨁ m − e L j P e . (77)
c L j f = j ( m − e − j + 1 ) L j − 1 f ( f ∈ P e ) . (78) cL^jf=j(m-e-j+1)L^{j-1}f \qquad(f\in P_e). \tag*{(78)} c L j f = j ( m − e − j + 1 ) L j − 1 f ( f ∈ P e ) . (78)
These follow directly from finite-dimensional s l 2 \mathfrak{sl}_2 sl 2 theory; see also the exterior Lefschetz treatment in [23 ] . The operator L L L commutes with the row s l 2 \mathfrak{sl}_2 sl 2 on U U U . On bidegree ( a , b ) (a,b) ( a , b ) , its formula in Λ a W ⊗ Λ b W \Lambda^aW\otimes\Lambda^bW Λ a W ⊗ Λ b W is ( − 1 ) a ∑ i ε i ⊗ ε i (-1)^a\sum_i\varepsilon_i\otimes\varepsilon_i ( − 1 ) a ∑ i ε i ⊗ ε i . Exterior multiplication by ω \omega ω fixes these signs uniformly.
Write η \eta η for the quadratic Koszul generator, of homological degree one, internal degree two, and G L ( U ) GL(U) G L ( U ) type det U \det U det U . The reduced complex is η E → L E \eta\mathcal{E}\xrightarrow{L}\mathcal{E} η E L E , with d ( η f ) = L f d(\eta f)=Lf d ( η f ) = L f . For its algebra formulas we use the totalized product, so η \eta η anticommutes with U ⊗ W U\otimes W U ⊗ W ; the complete sign convention is given in Appendix E.
Theorem 10.13 (All two-column homology and its reflection). For 0 ≤ q ≤ p ≤ n 0\le q\le p\le n 0 ≤ q ≤ p ≤ n , set ( p ∗ , q ∗ ) = ( n − q , n − p ) (p^*,q^*)=(n-q,n-p) ( p ∗ , q ∗ ) = ( n − q , n − p ) .
(1) If p + q ≤ n − 1 p+q\le n-1 p + q ≤ n − 1 , the sole homology is H 0 = M p , q = S [ λ ( p , q ) ] W H_0=M_{p,q}=S_{[\lambda(p,q)]}W H 0 = M p , q = S [ λ ( p , q )] W as O ( W ) O(W) O ( W ) modules.
(2) If p + q = n p+q=n p + q = n , the complex is acyclic.
(3) If p + q ≥ n + 1 p+q\ge n+1 p + q ≥ n + 1 , the sole homology is
H 1 ≅ M p ∗ , q ∗ ≅ M p ∗ , q ∗ ∗ ⊗ ( det W ) 2 H_1\cong M_{p^*,q^*}\cong M^*_{p^*,q^*}\otimes(\det W)^2 H 1 ≅ M p ∗ , q ∗ ≅ M p ∗ , q ∗ ∗ ⊗ ( det W ) 2 as O ( W ) O(W) O ( W ) modules, hence as S n S_n S n modules. The reflected row has degree 2 n − ( p + q ) 2n-(p+q) 2 n − ( p + q ) . If p > n p>n p > n , both terms vanish. These assertions hold in every row dimension in which the original row occurs; in case (3), p ∗ ≤ p − 1 p^*\le p-1 p ∗ ≤ p − 1 , so the reflected row occurs there as well.
More explicitly, if f ∈ P e f\in P_e f ∈ P e and k = m − e k=m-e k = m − e , reflection is the cycle map
ρ e ( f ) = η L k f k ! , ρ e − 1 ( η g ) = c k g k ! . (79) \rho_e(f)=\frac{\eta L^k f}{k!},\qquad\rho_e^{-1}(\eta g)=\frac{c^k g}{k!}. \tag*{(79)} ρ e ( f ) = k ! η L k f , ρ e − 1 ( η g ) = k ! c k g . (79)
Before extracting row multiplicity spaces, its source is P e ⊗ ( det U ) k + 1 P_e\otimes(\det U)^{k+1} P e ⊗ ( det U ) k + 1 and its target has internal degree 2 n − e 2n-e 2 n − e .
Proof. Each string in (77) contributes its initial vector to E / L E \mathcal{E}/L\mathcal{E} E / L E and its terminal vector to ker L \ker L ker L . Equation (78) gives L k + 1 f = 0 L^{k+1}f=0 L k + 1 f = 0 and c k L k f = ( k ! ) 2 f c^kL^kf=(k!)^2f c k L k f = ( k ! ) 2 f , proving (79) .
The row-λ ( p , q ) \lambda(p,q) λ ( p , q ) multiplicity space is the highest-weight kernel of the transfer Λ p W ⊗ Λ q W → Λ p + 1 W ⊗ Λ q − 1 W \Lambda^pW\otimes\Lambda^qW\to\Lambda^{p+1}W\otimes\Lambda^{q-1}W Λ p W ⊗ Λ q W → Λ p + 1 W ⊗ Λ q − 1 W . These kernels fix the Pieri normalization of (76) . Its low-degree primitive part is the orthogonal traceless Schur module: within-column traces vanish by antisymmetry, and the remaining trace is encoded by c c c . This is also the admissible case of the orthogonal Littlewood theorem [31 Theorem 4.4] . If f f f has G L ( U ) GL(U) G L ( U ) highest weight ( a , b ) (a,b) ( a , b ) , then e = a + b e=a+b e = a + b and (79) has weight
( a + k + 1 , b + k + 1 ) = ( n − b , n − a ) . (a+k+1,b+k+1)=(n-b,n-a). ( a + k + 1 , b + k + 1 ) = ( n − b , n − a ) .
Exterior Cauchy decomposition transposes this to the stated row reflection. The strings leave no homology in internal degree n n n . The dual formula follows from the exterior top-product pairing, whose orientation after adjoining η \eta η is ( det U ) n ⊗ ( det W ) 2 (\det U)^n\otimes(\det W)^2 ( det U ) n ⊗ ( det W ) 2 . Finally, if p > n p>n p > n , both Schur terms have more than dim W \dim W dim W rows. □
For the bivariate S n S_n S n -character R n ( u , v ) R_n(u,v) R n ( u , v ) of the exterior H 0 H_0 H 0 , reflection and the Euler character give
R n ( u , v ) − ( u v ) n R n ( u − 1 , v − 1 ) = ( 1 − u v ) λ u ( W n ) λ v ( W n ) , λ u ( W n ) = ∑ j [ Λ j W n ] u j . R_n(u,v)-(uv)^nR_n(u^{-1},v^{-1})=(1-uv)\lambda_u(W_n)\lambda_v(W_n),\qquad\lambda_u(W_n)=\sum_j[\Lambda^jW_n]u^j. R n ( u , v ) − ( uv ) n R n ( u − 1 , v − 1 ) = ( 1 − uv ) λ u ( W n ) λ v ( W n ) , λ u ( W n ) = j ∑ [ Λ j W n ] u j .
Products are tensor products in the representation ring. The dual operation disappears because S n S_n S n -modules are self-dual.
Corollary 10.14. (The full two-column first syzygies). In degree n + 1 n+1 n + 1 , for 1 ≤ b ≤ ⌊ ( n + 1 ) / 2 ⌋ 1 \le b \le\lfloor(n+1)/2 \rfloor 1 ≤ b ≤ ⌊( n + 1 ) /2 ⌋ , the row λ = ( 2 b , 1 n + 1 − 2 b ) \lambda= (2^b,1^{n+1-2b}) λ = ( 2 b , 1 n + 1 − 2 b ) has
H 1 , λ ≅ S [ ( 2 b − 1 , 1 n + 1 − 2 b ) ] W n ∣ S n ≅ sgn ⊗ S [ ( 2 b − 1 ) ] W n ∣ S n . H_{1,\lambda} \cong\mathbb{S}_{[(2^{b-1},1^{n+1-2b})]}W_n|_{S_n} \cong\operatorname{sgn}\otimes\mathbb{S}_{[(2^{b-1})]}W_n|_{S_n}. H 1 , λ ≅ S [( 2 b − 1 , 1 n + 1 − 2 b )] W n ∣ S n ≅ sgn ⊗ S [( 2 b − 1 )] W n ∣ S n . In particular, for n ≥ 4 n \ge4 n ≥ 4 ,
H 1 , ( 2 , 2 , 1 n − 3 ) ≅ S ( 2 , 1 n − 2 ) ⊕ S ( 2 , 2 , 1 n − 4 ) . H_{1,(2,2,1^{n-3})} \cong S^{(2,1^{n-2})} \oplus S^{(2,2,1^{n-4})}. H 1 , ( 2 , 2 , 1 n − 3 ) ≅ S ( 2 , 1 n − 2 ) ⊕ S ( 2 , 2 , 1 n − 4 ) . Proof. At this boundary, the reflected columns are obtained by subtracting one from each original column. Orthogonal determinant duality complements the first column in dimension n − 1 n-1 n − 1 , giving the second formula. For b = 2 b=2 b = 2 use S ( 2 ) W n = Sym 2 W n − K ≅ S ( n − 1 , 1 ) ⊕ S ( n − 2 , 2 ) \mathbb{S}_{(2)}W_n=\operatorname{Sym}^{2}W_n-K\cong S^{(n-1,1)}\oplus S^{(n-2,2)} S ( 2 ) W n = Sym 2 W n − K ≅ S ( n − 1 , 1 ) ⊕ S ( n − 2 , 2 ) for n ≥ 4 n\ge4 n ≥ 4 .
Thus at n = 3 n=3 n = 3 row ( 2 , 2 ) (2,2) ( 2 , 2 ) reflects to ( 2 ) (2) ( 2 ) and has H 1 ≅ S ( 2 , 1 ) H_1\cong S^{(2,1)} H 1 ≅ S ( 2 , 1 ) ; at n = 4 n=4 n = 4 row ( 2 , 2 , 1 ) (2,2,1) ( 2 , 2 , 1 ) reflects to ( 2 , 1 ) (2,1) ( 2 , 1 ) and has H 1 ≅ S ( 2 , 1 , 1 ) ⊕ S ( 2 , 2 ) H_1\cong S^{(2,1,1)}\oplus S^{(2,2)} H 1 ≅ S ( 2 , 1 , 1 ) ⊕ S ( 2 , 2 ) . The exterior H 0 H_0 H 0 character is also obtained in [23 , 25 ] , with explicit double-hook label formulas in [26 ] ; the positive homology is the width-two case of [31 Theorem 4.4] . The cycle map above will now identify its multiplication as well.
Proposition 10.15. (Multiplication on reflected homology). Put R = E / ( ω ) R=\mathcal{E}/(\omega) R = E / ( ω ) and I = η Ann E ( ω ) I=\eta\operatorname{Ann}_{\mathcal{E}}(\omega) I = η Ann E ( ω ) . For β ∈ U ⊗ W \beta\in U\otimes W β ∈ U ⊗ W , set ι β = [ c , ε β ] \iota_\beta=[c,\varepsilon_\beta] ι β = [ c , ε β ] . Then
β ρ e ( f ) = − ρ e − 1 ( ι β f ) , I = R ⋅ ( η P m ) . (80) \beta\,\rho_e(f)=-\rho_{e-1}(\iota_\beta f), \qquad I=R\cdot(\eta P_m). \tag*{(80)} β ρ e ( f ) = − ρ e − 1 ( ι β f ) , I = R ⋅ ( η P m ) . (80) In particular, the minimal R R R -generator space of I I I is η P m \eta P_m η P m , all in internal degree n + 1 n+1 n + 1 , and
dim ( I / R + I ) = ( 2 m m ) − ( 2 m m − 2 ) = 1 n + 1 ( 2 n n ) . \dim(I/R_+I)=\binom{2m}{m}-\binom{2m}{m-2}=\frac{1}{n+1}\binom{2n}{n}. dim ( I / R + I ) = ( m 2 m ) − ( m − 2 2 m ) = n + 1 1 ( n 2 n ) . The last formula uses the usual zero convention for binomial coefficients outside their range.
Proof. The identities [ L , ι β ] = ε β [L,\iota_\beta]=\varepsilon_\beta [ L , ι β ] = ε β and [ c , ι β ] = 0 [c,\iota_\beta]=0 [ c , ι β ] = 0 give ι β P e ⊆ P e − 1 \iota_\beta P_e\subseteq P_{e-1} ι β P e ⊆ P e − 1 . With k = m − e k=m-e k = m − e ,
L k + 1 ι β f = ( k + 1 ) β L k f L^{k+1}\iota_\beta f=(k+1)\beta L^k f L k + 1 ι β f = ( k + 1 ) β L k f because L k + 1 f = 0 L^{k+1}f=0 L k + 1 f = 0 . Division by ( k + 1 ) ! (k+1)! ( k + 1 )! , followed by moving β \beta β across η \eta η , gives the first identity. The contraction U ⊗ W ⊗ P e → P e − 1 U\otimes W\otimes P_e\to P_{e-1} U ⊗ W ⊗ P e → P e − 1 is a nonzero symplectic-equivariant map onto an irreducible primitive module. Contracting θ 1 ⋯ θ e \theta_1\cdots\theta_e θ 1 ⋯ θ e by ϕ e \phi_e ϕ e proves nonvanishing. It is therefore onto, and iteration from e = m e=m e = m proves generation. For m = 0 m=0 m = 0 the assertion is simply I = K η I=K\eta I = K η . The generator quotient is its first nonzero degree; the dimension formula is the primitive exterior-power formula.
The Catalan formula of [23 Corollary 4.4] counts the minimal syzygy-generating layer after evaluation on two odd rows. The identities are Sp 2 m \operatorname{Sp}_{2m} Sp 2 m -equivariant. For GL ( U ) × O ( W ) \operatorname{GL}(U)\times O(W) GL ( U ) × O ( W ) , contraction contributes the extra det U \det U det U needed between the source twists of ρ e \rho_e ρ e and ρ e − 1 \rho_{e-1} ρ e − 1 ; on row multiplicity spaces they are O ( W ) O(W) O ( W ) -equivariant.
Corollary 10.16. (All two-column syzygies are generated at the first boundary). Write H j , D ( V ) H_{j,D}(V) H j , D ( V ) for the degree-D D D moment homology as a polynomial functor of the ordinary row space V V V . For every D ≥ n + 1 D\ge n+1 D ≥ n + 1 , the cokernel of multiplication
H 0 , D − n − 1 ( V ) ⊗ H 1 , n + 1 ( V ) ⟶ H 1 , D ( V ) H_{0,D-n-1}(V)\otimes H_{1,n+1}(V)\longrightarrow H_{1,D}(V) H 0 , D − n − 1 ( V ) ⊗ H 1 , n + 1 ( V ) ⟶ H 1 , D ( V ) has no constituent S λ V S_\lambda V S λ V with λ 1 ≤ 2 \lambda_1\le2 λ 1 ≤ 2 . This holds in every row dimension in which that constituent occurs.
Proof. Purely odd two-dimensional evaluation is exact in characteristic zero and sends S λ ( Π U ) S_\lambda(\Pi U) S λ ( Π U ) to Π ∣ λ ∣ S λ U \Pi^{|\lambda|}S_\lambda U Π ∣ λ ∣ S λ U . It kills precisely the shapes with λ 1 > 2 \lambda_1>2 λ 1 > 2 and detects every remaining Schur multiplicity. It preserves the moment differential and multiplication. The evaluated map is onto by (10.15), so its original cokernel has no surviving two-column shape.
The quotient by shapes with more than two columns carries the induced algebra structure, since these shapes form a tensor ideal by Littlewood–Richardson containment. Appendix E identifies the complete odd differential algebra and shows that triple Massey products generate its positive homology.
Proposition 10.17 (A single-degree criterion for two-column generation). For fixed n ≥ 2 n \ge2 n ≥ 2 , Vandermonde generation in every two-column row and every degree is equivalent to generation of the two-column rows in degree n − 1 n-1 n − 1 . It suffices to test the rows
( 2 q , 1 n − 1 − 2 q ) , 2 ≤ q ≤ ⌊ ( n − 1 ) / 2 ⌋ . (2^q,1^{n-1-2q}), \qquad2 \le q \le\lfloor(n-1)/2 \rfloor. ( 2 q , 1 n − 1 − 2 q ) , 2 ≤ q ≤ ⌊( n − 1 ) /2 ⌋ . For n ≥ 5 n \ge5 n ≥ 5 , these tests require only n − 3 n-3 n − 3 ordinary alphabets. All two-column rows are generated for n ≤ 5 n \le5 n ≤ 5 .
Proof. Let G G G be the purely odd two-dimensional evaluation of the ordinary operator subfunctor P \mathcal P P . This means evaluation of the entire operator module, not of its Vandermonde seed alone. The corresponding full harmonic space is ⨁ e = 0 m P e \bigoplus_{e=0}^{m} P_e ⨁ e = 0 m P e . Derivative closure is expressed by the natural polynomial coaction
P D ( V ) ⟶ ( V ⊗ W ) ⊗ P D − 1 ( V ) . \mathcal P_D(V) \longrightarrow(V \otimes W) \otimes\mathcal P_{D-1}(V). P D ( V ) ⟶ ( V ⊗ W ) ⊗ P D − 1 ( V ) .
Exact odd evaluation therefore makes G G G closed under every constant-coefficient exterior contraction. The map ( U ⊗ W ) ∗ ⊗ P e → P e − 1 (U \otimes W)^* \otimes P_e \to P_{e-1} ( U ⊗ W ) ∗ ⊗ P e → P e − 1 is onto, by the same nonzero symplectic-equivariant contraction used in Proposition 10.15. Thus G m = P m G_m=P_m G m = P m propagates down to G e = P e G_e=P_e G e = P e for every e ≤ m e \le m e ≤ m . Odd evaluation detects every two-column shape, proving the equivalence. The cases q = 0 , 1 q=0,1 q = 0 , 1 are row hooks and follow from Proposition 4.3. Each remaining row has n − 1 − q ≤ n − 3 n-1-q \le n-3 n − 1 − q ≤ n − 3 parts, so Lemma 3.1 and Corollary 5.4 make n − 3 n-3 n − 3 alphabets detect every remaining multiplicity. The list is empty for n ≤ 4 n \le4 n ≤ 4 ; for n = 5 n=5 n = 5 its only row is ( 2 , 2 ) (2,2) ( 2 , 2 ) , covered by Haiman’s bivariate operator theorem as in Theorem 12.2. □
The remaining task is to generate the displayed degree-( n − 1 ) (n-1) ( n − 1 ) rows. The derivative coaction then propagates generation to every lower degree.
The first row-hook moment syzygies.
Theorem 10.18 (The first row-hook moment syzygies). In total degree n + 1 n+1 n + 1 ,
H 1 ( K n , k , n + 1 ) n + 1 ∣ row hooks ≅ ⨁ b = 0 n − 1 S ( n + 1 − b , 1 b ) K k ⊗ S ( n − b , 1 b ) . H_1(\mathcal K_{n,k,n+1})_{n+1}\big|_{\text{row hooks}} \cong\bigoplus_{b=0}^{n-1} S^{(n+1-b,1^b)}K^k \otimes S^{(n-b,1^b)}. H 1 ( K n , k , n + 1 ) n + 1 row hooks ≅ b = 0 ⨁ n − 1 S ( n + 1 − b , 1 b ) K k ⊗ S ( n − b , 1 b ) . Each summand occurs once. Row lengths exceeding k k k give zero Schur modules. Every H i H_i H i with i ≥ 2 i \ge2 i ≥ 2 has zero row-hook part in this total degree.
Proof. In the ( 1 ∣ 1 ) (1|1) ( 1∣1 ) free cover from the proof of Theorem 10.9, its kernel begins in degree n + 1 n+1 n + 1 . A positive term of the super Koszul complex on that kernel has internal degree greater than n + 1 n+1 n + 1 . The Tor long exact sequence therefore makes every H i H_i H i , i ≥ 2 i \ge2 i ≥ 2 , zero in degree n + 1 n+1 n + 1 . Consequently the character discrepancy (10.8) is exactly H 1 H_1 H 1 .
Let T b T_b T b be the label multiplicity space of the row ( n + 1 − b , 1 b ) (n+1-b,1^b) ( n + 1 − b , 1 b ) in this homology, and put T − 1 = 0 T_{-1}=0 T − 1 = 0 . The two monomials in its super Schur specialization give
ch T b + ch T b − 1 = e b [ Z ] h n − b [ Z ] ( 0 ≤ b ≤ n ) . \operatorname{ch} T_b+\operatorname{ch} T_{b-1}=e_b[Z]h_{n-b}[Z] \qquad(0 \le b \le n). ch T b + ch T b − 1 = e b [ Z ] h n − b [ Z ] ( 0 ≤ b ≤ n ) .
For 1 ≤ b ≤ n − 1 1 \le b \le n-1 1 ≤ b ≤ n − 1 , Pieri’s rule says
e b h n − b = s ( n − b , 1 b ) + s ( n − b + 1 , 1 b − 1 ) . e_bh_{n-b}=s_{(n-b,1^b)}+s_{(n-b+1,1^{b-1})}. e b h n − b = s ( n − b , 1 b ) + s ( n − b + 1 , 1 b − 1 ) .
At the endpoints the right side is h n h_n h n and e n e_n e n . Induction gives T b ≅ S ( n − b , 1 b ) T_b \cong S^{(n-b,1^b)} T b ≅ S ( n − b , 1 b ) for 0 ≤ b ≤ n − 1 0 \le b \le n-1 0 ≤ b ≤ n − 1 , and T n = 0 T_n=0 T n = 0 . Exact Schur evaluation yields the stated ordinary-row result and the higher-homology vanishing. □
For n = 1 n=1 n = 1 , the theorem gives only S ( 2 ) K k S_{(2)}K^k S ( 2 ) K k with trivial label in degree two. For n = 2 n=2 n = 2 , it gives S ( 3 ) K k ⊗ S ( 2 ) S_{(3)}K^k \otimes S^{(2)} S ( 3 ) K k ⊗ S ( 2 ) and S ( 2 , 1 ) K k ⊗ S ( 1 , 1 ) S_{(2,1)}K^k \otimes S^{(1,1)} S ( 2 , 1 ) K k ⊗ S ( 1 , 1 ) in degree three. These also follow by direct moment calculation.
10.9. Superspace characters and the first correction. Theorem 9.8 identifies the first unstable correction for every nonempty tail τ \tau τ and every D > ∣ τ ∣ D > |\tau| D > ∣ τ ∣ . Its idempotent specialization proves injectivity at that wall directly, without an Euler-character comparison.
Corollary 10.19 (Stable diagonal superspace characters). Let H + ( X ) = ∑ d ≥ 1 h d ( X ) H_{+}(X) = \sum_{d \ge1} h_d(X) H + ( X ) = ∑ d ≥ 1 h d ( X ) and define sp Q ∣ U ( p m ) = p m ( Q ) + ( − 1 ) m − 1 p m ( U ) \operatorname{sp}_{Q\mid U}(p_m) = p_m(Q) + (-1)^{m-1}p_m(U) sp Q ∣ U ( p m ) = p m ( Q ) + ( − 1 ) m − 1 p m ( U ) . For τ ⊢ s \tau\vdash s τ ⊢ s , D ≥ s D \ge s D ≥ s and n ≥ D + τ 1 n \ge D + \tau_1 n ≥ D + τ 1 , the degree-D D D multiplicity character in any bosonic–fermionic superspace is
ch Hom S n ( S τ [ n ] , R [ D ] ( k , j ) ) = sp Q ∣ U ( [ s τ [ H + ( X ) ] ] D ) . \operatorname{ch}\operatorname{Hom}_{S_n}\left(S^{\tau[n]},R^{(k,j)}_{[D]}\right)=\operatorname{sp}_{Q\mid U}\left([s_{\tau}[H_{+}(X)]]_D\right). ch Hom S n ( S τ [ n ] , R [ D ] ( k , j ) ) = sp Q ∣ U ( [ s τ [ H + ( X )] ] D ) . The full character-polynomial formula of Theorem 10.10 likewise transfers to arbitrary superspaces in its range D ≥ 3 D \ge3 D ≥ 3 , n ≥ max { D , 2 D − 6 } n \ge\max\{D,2D-6\} n ≥ max { D , 2 D − 6 } .
Proof. The bosonic identities hold in every row dimension and hence are identities of universal row symmetric functions. Apply Lentfer’s diagonal supersymmetry theorem [25 Theorem 1.2 and Corollary 1.3] . The displayed specialization records parity; at module level the corresponding functors are super Schur functors. □
Disjoint-label products and first occurrences
The first occurrence of a label has a direct monomial description: fill each row of its Young diagram with the least-weight one-label monomials. Each column then forms a lower monomial diagram. We generate the alternants of these diagrams on disjoint label sets and apply the Young symmetrizer. This generates every minimum-degree filling.
For V = K k V = K^k V = K k , define
b j = dim Sym j V = ( k + j − 1 k − 1 ) , N j = ( k + j k ) , N − 1 = 0. b_j = \dim\operatorname{Sym}^{j} V = \binom{k+j-1}{k-1}, \qquad N_j = \binom{k+j}{k}, \qquad N_{-1}=0. b j = dim Sym j V = ( k − 1 k + j − 1 ) , N j = ( k k + j ) , N − 1 = 0.
Let d i = j d_i = j d i = j for N j − 1 < i ≤ N j N_{j-1} < i \le N_j N j − 1 < i ≤ N j , and pad all partitions with zeros. Thus d 1 , d 2 , … d_1,d_2,\ldots d 1 , d 2 , … are the degrees of all monomials in k k k variables, in nondecreasing order.
Theorem 11.1 (Complete first occurrences). For every n , k ≥ 1 n,k \ge1 n , k ≥ 1 and every μ ⊢ n \mu\vdash n μ ⊢ n , put
δ μ ( k ) = ∑ i ≥ 1 μ i d i . \delta_\mu(k)=\sum_{i\ge1}\mu_i d_i. δ μ ( k ) = i ≥ 1 ∑ μ i d i . There is no S μ S^\mu S μ in P n , k P_{n,k} P n , k below this degree, and
P n , k [ δ μ ( k ) ] μ = H n , k [ δ μ ( k ) ] μ = P n , k [ δ μ ( k ) ] μ . (81) P_{n,k}[\delta_\mu(k)]_\mu=H_{n,k}[\delta_\mu(k)]_\mu=P_{n,k}[\delta_\mu(k)]_\mu. \tag*{(81)} P n , k [ δ μ ( k ) ] μ = H n , k [ δ μ ( k ) ] μ = P n , k [ δ μ ( k ) ] μ . (81) Its row multiplicity space is
Hom S n ( S μ , P n , k [ δ μ ( k ) ] ) ≅ ⨂ j ≥ 0 S ( μ N j − 1 + 1 , … , μ N j ) ( Sym j V ) . (82) \operatorname{Hom}_{S_n}\left(S^\mu,P_{n,k}[\delta_\mu(k)]\right)\cong\bigotimes_{j\ge0} \mathbb{S}_{(\mu_{N_{j-1}+1},\ldots,\mu_{N_j})}(\operatorname{Sym}^{j}V). \tag*{(82)} Hom S n ( S μ , P n , k [ δ μ ( k )] ) ≅ j ≥ 0 ⨂ S ( μ N j − 1 + 1 , … , μ N j ) ( Sym j V ) . (82)
The degree and tensor-product formula come from the semistandard construction of Schur functors; Romero records the ambient plethystic character in [30 Theorem 1.2 and §2] . For the alternating label, Wallach gives the determinant-twisted exterior-power description of the first layer [35 Theorem 1] . The assertion here generates that entire space. Its proof rests on a product theorem for arbitrary lower monomial diagrams.
Curvilinear quotients and colored determinants
A finite nonempty set Γ \Gamma Γ of monomials in u 1 , … , u k u_1,\ldots,u_k u 1 , … , u k is a lower diagram if it is closed under divisors. For labels C C C with ∣ C ∣ = ∣ Γ ∣ |C|=|\Gamma| ∣ C ∣ = ∣Γ∣ , put
Δ Γ ( X C ) = det ( X i α ) i ∈ C , α ∈ Γ , X i α = ∏ a = 1 k ( x i ( a ) ) α a . \Delta_{\Gamma}(X_C)=\det(X_i^\alpha)_{i\in C,\alpha\in\Gamma}, \qquad X_i^\alpha=\prod_{a=1}^{k}(x_i^{(a)})^{\alpha_a}. Δ Γ ( X C ) = det ( X i α ) i ∈ C , α ∈ Γ , X i α = a = 1 ∏ k ( x i ( a ) ) α a .
An ordering of C C C and Γ \Gamma Γ fixes the irrelevant overall sign.
Theorem 11.2 (Products of lower-diagram alternants). Partition { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } into nonempty sets C 1 , … , C r C_1,\ldots,C_r C 1 , … , C r . For each c c c , choose any lower diagram Γ c \Gamma_c Γ c of ∣ C c ∣ |C_c| ∣ C c ∣ monomials in k k k variables. Then
∏ c = 1 r Δ Γ c ( X C c ) ∈ P n , k . (83) \prod_{c=1}^{r}\Delta_{\Gamma_c}(X_{C_c})\in\mathcal{P}_{n,k}. \tag*{(83)} c = 1 ∏ r Δ Γ c ( X C c ) ∈ P n , k . (83) The lower diagrams may be chosen independently. The derivative and polarization module generated by this product is also contained in P n , k \mathcal{P}_{n,k} P n , k .
The product theorem follows by placing all label blocks in one quotient of a free module. Bérczi–Svendsen’s theorem moves that monomial quotient into the closure of the curvilinear locus, where a jet determinant gives its operator realization. We work over C \mathbb{C} C with the underlying reduced loci and then descend the resulting linear-span statement to K K K . The case n = 1 n=1 n = 1 is immediate.
For S = C [ u 1 , … , u k ] S=\mathbb{C}[u_1,\ldots,u_k] S = C [ u 1 , … , u k ] , the punctual Quot scheme parametrizes length-n n n quotients of S ⊕ r S^{\oplus r} S ⊕ r supported at the origin. Its curvilinear locus consists of quotients whose underlying S S S -module is the structure module of an embedded curvilinear length-n n n subscheme. In coordinates, such a quotient has the form
S ⊕ r ⟶ A = C [ t ] / ( t n ) , u a ⟼ f a ( t ) , e c ⟼ g c ( t ) , (84) S^{\oplus r}\longrightarrow A=\mathbb{C}[t]/(t^n),\qquad u_a\longmapsto f_a(t),\qquad e_c\longmapsto g_c(t), \tag*{(84)} S ⊕ r ⟶ A = C [ t ] / ( t n ) , u a ⟼ f a ( t ) , e c ⟼ g c ( t ) , (84)
where the f a f_a f a are pointed jets, at least one has nonzero linear coefficient, and the g c g_c g c generate A A A as an S S S -module. Equivalently, at least one g c g_c g c has nonzero constant term.
Theorem 11.3 (Bérczi–Svendsen). Every quotient
S / I 1 ⊕ ⋯ ⊕ S / I r S/I_1\oplus\cdots\oplus S/I_r S / I 1 ⊕ ⋯ ⊕ S / I r with each I c I_c I c a monomial ideal of finite colength is contained in the closure of the curvilinear locus in the length- n n n punctual Quot scheme, where n = ∑ c dim C S / I c n=\sum_c\dim_{\mathbb{C}}S/I_c n = ∑ c dim C S / I c .
This is [3 Theorem 1.3, equivalently Theorem 5.2] . The same paper’s Theorem 1.1 gives the r = 1 r=1 r = 1 Hilbert-scheme case. The rank-r r r Quot statement treats the direct sum of the independent monomial quotients and is the input for (11.3).
11.1.1. The graph determinant. For pointed jets f = ( f 1 , … , f k ) f=(f_1,\ldots,f_k) f = ( f 1 , … , f k ) modulo t n t^n t n , define
G f ( X ) = det ( ( [ t j ] exp ( ∑ a = 1 k x i ( a ) f a ( t ) ) ) 1 ≤ i ≤ n 0 ≤ j < n ) . (85) G_f(X)=\det\left(\left([t^j]\exp\left(\sum_{a=1}^{k}x_i^{(a)}f_a(t)\right)\right)_{\substack{1\leq i\leq n\\0\leq j<n}}\right). \tag*{(85)} G f ( X ) = det ( [ t j ] exp ( a = 1 ∑ k x i ( a ) f a ( t ) ) ) 1 ≤ i ≤ n 0 ≤ j < n . (85)
This is the jet determinant of Theorem 5.2, with the constant ∏ j = 0 n − 1 j ! \prod_{j=0}^{n-1}j! ∏ j = 0 n − 1 j ! omitted. On the graph chart f 1 ( t ) = t f_1(t)=t f 1 ( t ) = t , f a ( t ) = ∑ s a a , s t s f_a(t)=\sum_s a_{a,s}t^s f a ( t ) = ∑ s a a , s t s , that theorem gives
( ∏ j = 0 n − 1 j ! ) G f = exp ( ∑ a = 2 k ∑ s = 1 n − 1 a a , s Y a , s ) Δ n ∈ P n , k , Y a , s = ∑ i x i ( a ) ∂ x i s . (86) \left(\prod_{j=0}^{n-1}j!\right)G_f=\exp\left(\sum_{a=2}^{k}\sum_{s=1}^{n-1}a_{a,s}Y_{a,s}\right)\Delta_n\in\mathcal{P}_{n,k},\qquad Y_{a,s}=\sum_i x_i^{(a)}\partial_{x_i}^{s}. \tag*{(86)} ( j = 0 ∏ n − 1 j ! ) G f = exp ( a = 2 ∑ k s = 1 ∑ n − 1 a a , s Y a , s ) Δ n ∈ P n , k , Y a , s = i ∑ x i ( a ) ∂ x i s . (86)
Its inverse source reparametrization gives the same membership throughout the dense regular-jet locus. We now retain this membership under the colored specialization.
11.1.2. The colored Plücker map. To carry curvilinear closure into a linear span of polynomials, we use one linear map on a fixed exterior power. Its label colors remember which lower diagram belongs to each block. At a graph quotient it becomes a differential operator on the jet determinant; at a monomial quotient it becomes the desired product of alternants.
Every length-n n n module supported at the origin is annihilated by m n \mathfrak{m}^n m n , where m = ( u 1 , … , u k ) \mathfrak{m}=(u_1,\ldots,u_k) m = ( u 1 , … , u k ) . To see this, the strict nonzero steps in its m \mathfrak{m} m -adic filtration each decrease dimension; a repeated step would be zero by Nakayama’s lemma. All quotient points therefore belong to the fixed Grassmannian of n n n -dimensional quotients of
W = ( S / m n ) ⊕ r . W=(S/\mathfrak{m}^n)^{\oplus r}. W = ( S / m n ) ⊕ r .
Let c ( i ) = c c(i)=c c ( i ) = c for i ∈ C c i\in C_c i ∈ C c . Define a linear map
L C : ⋀ n W ∗ ⟶ P n , k , L_C:\bigwedge^n W^*\longrightarrow P_{n,k}, L C : ⋀ n W ∗ ⟶ P n , k ,
q 0 ∧ ⋯ ∧ q n − 1 ⟼ det ( ∑ ∣ α ∣ < n q j ( u α e c ( i ) ) X i α α ! ) i , j , α ! = ∏ a α a ! . (87) q_0\wedge\cdots\wedge q_{n-1}\longmapsto\det\left(\sum_{|\alpha|<n}q_j(u^\alpha e_{c(i)})\frac{X_i^\alpha}{\alpha!}\right)_{i,j},\qquad\alpha!=\prod_a\alpha_a!. \tag*{(87)} q 0 ∧ ⋯ ∧ q n − 1 ⟼ det ∣ α ∣ < n ∑ q j ( u α e c ( i ) ) α ! X i α i , j , α ! = a ∏ α a ! . (87)
Multilinearity and alternation in the q j q_j q j make this a well-defined linear map on the exterior power. Its normalization does not depend on the quotient point. Its values have total degree at most n ( n − 1 ) n(n-1) n ( n − 1 ) , so they lie in a fixed finite-dimensional polynomial space E = P n , k , ≤ n ( n − 1 ) E=P_{n,k,\leq n(n-1)} E = P n , k , ≤ n ( n − 1 ) .
For the graph quotient (84) , take q j q_j q j to be its coefficient-of-t j t^j t j functional. Then
L C ( q 0 ∧ ⋯ ∧ q n − 1 ) = det ( [ t j ] g c ( i ) ( t ) exp ( ∑ a X i ( a ) f a ( t ) ) ) i , j = ( ∏ i = 1 n g c ( i ) ( ∂ X i ) ) G f ∈ P n , k . (88) \begin{aligned}
L_C(q_0\wedge\cdots\wedge q_{n-1})=\det\left([t^j]g_{c(i)}(t)\exp\left(\sum_a X_i^{(a)}f_a(t)\right)\right)_{i,j} \\
&=\left(\prod_{i=1}^n g_{c(i)}(\partial_{X_i})\right)G_f\in P_{n,k}. \tag*{(88)}
\end{aligned} L C ( q 0 ∧ ⋯ ∧ q n − 1 ) = det ( [ t j ] g c ( i ) ( t ) exp ( a ∑ X i ( a ) f a ( t ) ) ) i , j = ( i = 1 ∏ n g c ( i ) ( ∂ X i ) ) G f ∈ P n , k . (88)
The second equality uses f 1 ( t ) = t f_1(t)=t f 1 ( t ) = t . Each derivative affects just one row of the determinant. Constant terms of the g c g_c g c are allowed and act as scalar summands of these differential operators.
The graph chart is dense in the curvilinear Quot locus: the regular jets form an irreducible open set, the condition that the g c g_c g c generate is also open, and f 1 ′ ( 0 ) ≠ 0 f_1'(0)\ne0 f 1 ′ ( 0 ) = 0 is a nonempty open condition. Inverse reparametrization preserves the quotient and produces graph form. For n = 1 n=1 n = 1 , the same assertion is immediate with all f a = 0 f_a=0 f a = 0 . Consequently the closed condition
L C ( ℓ ) ⊆ P n , k (89) L_C(\ell)\subseteq P_{n,k} \tag*{(89)} L C ( ℓ ) ⊆ P n , k (89)
holds on the entire curvilinear Quot component, where ℓ \ell ℓ is the Plücker line of a quotient. More explicitly, compose L C L_C L C with the map E → E / ( E ∩ P n , k ) E\to E/(E\cap P_{n,k}) E → E / ( E ∩ P n , k ) . Condition (89) says that this linear map vanishes on ℓ \ell ℓ , so it is a closed linear condition on the Grassmannian. It includes the points where L C ( ℓ ) = 0 L_C(\ell)=0 L C ( ℓ ) = 0 , and therefore passes through every specialization in the closure.
Proof of Theorem 11.2 . Let I c I_c I c be the monomial ideal whose standard monomials are Γ c \Gamma_c Γ c . The monomial quotient ⨁ c S / I c \bigoplus_c S/I_c ⨁ c S / I c has total length n n n , so Theorem 11.3 and (89) apply. Choose quotient functionals dual to the colored basis { u α e c : u α ∈ Γ c } \{u^\alpha e_c:u^\alpha\in\Gamma_c\} { u α e c : u α ∈ Γ c } . In (11.7), the corresponding column is zero outside the label block C c C_c C c . Thus, up to the ordering sign,
L C ( ℓ ) = K ( ∏ c = 1 n Δ Γ c ( X C c ) ∏ c = 1 n ∏ u α ∈ Γ c α ! ) . L_C(\ell)=K\left(\frac{\prod_{c=1}^n\Delta_{\Gamma_c}(X_{C_c})}{\prod_{c=1}^n\prod_{u^\alpha\in\Gamma_c}\alpha!}\right). L C ( ℓ ) = K ( ∏ c = 1 n ∏ u α ∈ Γ c α ! ∏ c = 1 n Δ Γ c ( X C c ) ) .
This is a nonzero line. It is contained in P n , k P_{n,k} P n , k by (89) , proving (11.3) over C \mathbb{C} C . The product, the defining operators, and the Vandermonde all have rational coefficients. The assertion is membership in a finite-dimensional rational linear span, so membership over C \mathbb{C} C implies membership over Q \mathbb{Q} Q and then over every characteristic-zero field K K K . Closure under derivatives and polarizations gives the last assertion. □ \square □
Each factor in (11.3) is harmonic for its own label set: a positive diagonal differential operator replaces a monomial column by a divisor column already present, or by zero. Since the label sets are disjoint, the product is harmonic for all n n n labels. Theorem 11.2 places these products in the operator module.
11.2. Multiplication on disjoint label sets. The product theorem extends from lower-diagram alternants to the whole family of operator modules. Private row spaces let the operators act independently on each label block; an invertible row shear and exact specialization then identify all row spaces.
Theorem 11.4 (Disjoint-label multiplication). Let C 1 , … , C r C_1,\ldots,C_r C 1 , … , C r partition { 1 , … , n } \{1,\ldots,n\} { 1 , … , n } , with ∣ C c ∣ = n c > 0 |C_c|=n_c>0 ∣ C c ∣ = n c > 0 . For arbitrary F c ∈ P n c , k F_c\in\mathcal{P}_{n_c,k} F c ∈ P n c , k , relabelled on C c C_c C c , one has
∏ c = 1 r F c ( X C c ) ∈ P n , k . \prod_{c=1}^{r} F_c(X_{C_c})\in\mathcal{P}_{n,k}. c = 1 ∏ r F c ( X C c ) ∈ P n , k . These products are associative, commutative, and equivariant under relabeling and the diagonal row action. They define canonical S n × G L k \mathfrak{S}_n\times\mathrm{GL}_k S n × GL k -equivariant maps
Ind S n 1 × ⋯ × S n r S n ( ⨂ c = 1 r P n c , k ) ⟶ P n , k . (90) \operatorname{Ind}_{\mathfrak{S}_{n_1}\times\cdots\times\mathfrak{S}_{n_r}}^{\mathfrak{S}_n}\left(\bigotimes_{c=1}^{r}\mathcal{P}_{n_c,k}\right)\longrightarrow\mathcal{P}_{n,k}. \tag*{(90)} Ind S n 1 × ⋯ × S n r S n ( c = 1 ⨂ r P n c , k ) ⟶ P n , k . (90) Proof. Introduce r r r private blocks of k k k variable rows, with variables x i ( c , a ) x_i^{(c,a)} x i ( c , a ) . For the c c c th factor take the lower diagram Γ c = { 1 , u c , 1 , … , u c , n c − 1 } \Gamma_c=\{1,u_{c,1},\ldots,u_{c,n_c-1}\} Γ c = { 1 , u c , 1 , … , u c , n c − 1 } in these k r kr k r one-label variables. Theorem 11.2 gives
∏ c Δ C c ( x ( c , 1 ) ) ∈ P n , k r . \prod_c\Delta_{C_c}(x^{(c,1)})\in\mathcal{P}_{n,kr}. c ∏ Δ C c ( x ( c , 1 ) ) ∈ P n , k r . Generate F c F_c F c using derivatives and polarizations whose source and target rows both lie in its private block. Such a global polarization acts only on the c c c th factor: every other factor is independent of its source row, and its derivative order is at least one. Label-specific partial derivatives have the same property. By linearity, this produces ∏ c F c ( X C c ) \prod_c F_c(X_{C_c}) ∏ c F c ( X C c ) inside P n , k r \mathcal{P}_{n,kr} P n , k r .
Apply the invertible row shear
x i ( c , a ) ⟼ x i ( c , a ) + x i ( 1 , a ) ( c > 1 ) , x i ( 1 , a ) ⟼ x i ( 1 , a ) . x_i^{(c,a)}\longmapsto x_i^{(c,a)}+x_i^{(1,a)}\quad(c>1),\qquad x_i^{(1,a)}\longmapsto x_i^{(1,a)}. x i ( c , a ) ⟼ x i ( c , a ) + x i ( 1 , a ) ( c > 1 ) , x i ( 1 , a ) ⟼ x i ( 1 , a ) . Then set all private rows with c > 1 c>1 c > 1 to zero. The shear preserves P n , k r \mathcal{P}_{n,kr} P n , k r , and Corollary 5.4 sends the result into P n , k \mathcal{P}_{n,k} P n , k . The resulting polynomial is the desired product in the common k k k rows. The associativity, commutativity, and equivariance assertions follow from ordinary multiplication in disjoint polynomial variables. Frobenius reciprocity gives (11.10). □
Corollary 11.5 (Tensor products under restriction). There is a natural injective map of G L k × S n 1 × ⋯ × S n r \mathrm{GL}_k\times\mathfrak{S}_{n_1}\times\cdots\times\mathfrak{S}_{n_r} GL k × S n 1 × ⋯ × S n r -modules
\bigotimes_c\mathcal{P}_{n_c,k}\lhook\joinrel\longrightarrow\operatorname{Res}^{\mathfrak{S}_n}_{\mathfrak{S}_{n_1}\times\cdots\times\mathfrak{S}_{n_r}}\mathcal{P}_{n,k}.
Consequently, for label partitions μ c ⊢ n c \mu_c\vdash n_c μ c ⊢ n c and a row partition λ \lambda λ of length at most k k k ,
∑ ν ⊢ n LR μ 1 , … , μ r ν P n , k λ , ν ≥ ∑ λ 1 , … , λ r LR λ 1 , … , λ r λ ∏ c = 1 r P n c , k λ c , μ c . (91) \sum_{\nu\vdash n}\operatorname{LR}^{\nu}_{\mu_1,\ldots,\mu_r}\mathcal{P}_{n,k}^{\lambda,\nu}\geq\sum_{\lambda_1,\ldots,\lambda_r}\operatorname{LR}^{\lambda}_{\lambda_1,\ldots,\lambda_r}\prod_{c=1}^{r}\mathcal{P}_{n_c,k}^{\lambda_c,\mu_c}. \tag*{(91)} ν ⊢ n ∑ LR μ 1 , … , μ r ν P n , k λ , ν ≥ λ 1 , … , λ r ∑ LR λ 1 , … , λ r λ c = 1 ∏ r P n c , k λ c , μ c . (91) Here LR denotes an iterated Littlewood–Richardson coefficient, and the sum on the right is over row partitions of length at most k k k .
Proof. Multiplication identifies the tensor product of polynomial rings on the disjoint label sets with the full polynomial ring. Its restriction to the tensor product of the indicated subspaces is therefore injective, and Theorem 11.4 places its image in P n , k \mathcal{P}_{n,k} P n , k . Decompose this inclusion under the row group and the label Young subgroup. The Littlewood–Richardson restriction and tensor product rules give (11.11). □
For fixed label blocks, multiplication is injective as stated in Corollary 11.5. The induced map (11.10) combines all relabelings of those blocks and can have a kernel.
Every positive-weight first layer
The first-occurrence construction works for every positive row weight. A minimum-weight column contains all cheaper monomials, so it remains a lower diagram even when many monomials have equal weight. Give the a a a th row a positive integer weight w a w_a w a . Write P n , k ( w ) [ d ] P_{n,k}^{(w)}[d] P n , k ( w ) [ d ] for weighted degree d d d and put
W d ( w ) = span K { u α : w ⋅ α = d } , b d ( w ) = dim W d ( w ) , W_d(w)=\operatorname{span}_K\{u^\alpha:w\cdot\alpha=d\},\qquad b_d(w)=\dim W_d(w), W d ( w ) = span K { u α : w ⋅ α = d } , b d ( w ) = dim W d ( w ) ,
N d ( w ) = ∑ e = 0 d b e ( w ) , N − 1 ( w ) = 0. N_d(w)=\sum_{e=0}^{d}b_e(w),\qquad N_{-1}(w)=0. N d ( w ) = e = 0 ∑ d b e ( w ) , N − 1 ( w ) = 0.
Order all monomials by nondecreasing weight, breaking ties in any fixed way. Let d i ( w ) d_i(w) d i ( w ) be the weight of the i i i th monomial and set
δ μ ( w ) = ∑ i μ i d i ( w ) , μ ( d , w ) = ( μ N d − 1 ( w ) + 1 , … , μ N d ( w ) ) . \delta_\mu(w)=\sum_i\mu_i d_i(w),\qquad\mu^{(d,w)}=(\mu_{N_{d-1}(w)+1},\ldots,\mu_{N_d(w)}). δ μ ( w ) = i ∑ μ i d i ( w ) , μ ( d , w ) = ( μ N d − 1 ( w ) + 1 , … , μ N d ( w ) ) .
Empty blocks contribute the trivial factor.
Theorem 11.6 (Complete generation on every lower weighted face). For every positive integral weight vector w w w and μ ⊢ n \mu\vdash n μ ⊢ n , δ μ ( w ) \delta_\mu(w) δ μ ( w ) is the first weighted degree containing S μ S^\mu S μ , and
P n , k ( w ) [ δ μ ( w ) ] μ = H n , k ( w ) [ δ μ ( w ) ] μ = P n , k ( w ) [ δ μ ( w ) ] μ . (92) P_{n,k}^{(w)}[\delta_\mu(w)]_\mu=H_{n,k}^{(w)}[\delta_\mu(w)]_\mu=P_{n,k}^{(w)}[\delta_\mu(w)]_\mu. \tag*{(92)} P n , k ( w ) [ δ μ ( w ) ] μ = H n , k ( w ) [ δ μ ( w ) ] μ = P n , k ( w ) [ δ μ ( w ) ] μ . (92) The multiplicity space of this layer is
⨂ d ≥ 0 S μ ( d , w ) ( W d ( w ) ) . (93) \bigotimes_{d\ge0}S_{\mu^{(d,w)}}(W_d(w)). \tag*{(93)} d ≥ 0 ⨂ S μ ( d , w ) ( W d ( w )) . (93) This is an isomorphism for the row torus and for the subgroup of G L k \mathrm{GL}_k GL k preserving the weighted grading.
Proof. As an S n S_n S n -module with row action, P n , k = ( Sym V ) ⊗ n P_{n,k}=(\operatorname{Sym} V)^{\otimes n} P n , k = ( Sym V ) ⊗ n . Schur–Weyl duality identifies the S μ S^\mu S μ multiplicity space with S μ ( Sym V ) S_\mu(\operatorname{Sym} V) S μ ( Sym V ) . In its semistandard tableau construction, an entry in row i i i is at least the i i i th element of the ordered monomial basis. Its weight is therefore at least d i ( w ) d_i(w) d i ( w ) . Filling row i i i with that i i i th monomial attains the bound δ μ ( w ) \delta_\mu(w) δ μ ( w ) .
Equality in the weight bound requires every box in row i i i to contain a monomial of weight exactly d i ( w ) d_i(w) d i ( w ) . Inside a weight block the fillings are the semistandard fillings for the partition μ ( d , w ) \mu^{(d,w)} μ ( d , w ) on W d ( w ) W_d(w) W d ( w ) . Between different blocks, column inequalities are automatic. This proves (93) and the claimed first degree. Every positive D α D_\alpha D α is S n S_n S n -equivariant and lowers weighted degree by w ⋅ α > 0 w\cdot\alpha>0 w ⋅ α > 0 . It annihilates this whole first isotype, proving equality of the P P P and H H H terms in (92) .
It remains to generate the entire space. Fix a standard label tableau T T T of shape μ \mu μ , with label sets C c C_c C c in its columns. Let F F F be any semistandard monomial filling attaining δ μ ( w ) \delta_\mu(w) δ μ ( w ) , and let Γ c \Gamma_c Γ c be the monomial set in its c c c th column. Column strictness makes these monomials distinct. For each weight strictly below the last weight in that column, the column crosses the complete corresponding block of rows. It must therefore contain every monomial of that weight; in the last block it contains a subset. Since all w a w_a w a are positive, a proper monomial divisor has strictly smaller weight. Thus every Γ c \Gamma_c Γ c is a lower diagram.
Let m F m_F m F be the monomial in P n , k P_{n,k} P n , k obtained by putting the monomial in each box of F F F at the label in the same box of T T T . If b T b_T b T is the column antisymmetrizer, then
b T m F = ∏ c Δ Γ c ( X C c ) b_Tm_F=\prod_c\Delta_{\Gamma_c}(X_{C_c}) b T m F = c ∏ Δ Γ c ( X C c )
up to the fixed column ordering signs. Theorem 11.2 places this polynomial in P n , k P_{n,k} P n , k . Applying the row symmetrizer a T a_T a T of the label tableau also preserves P n , k P_{n,k} P n , k . The standard semistandard construction says that the vectors a T b T m F a_Tb_Tm_F a T b T m F , as F F F ranges over these minimum-weight fillings, span the first-weight part of the image of the Young symmetrizer a T b T a_Tb_T a T b T . That image contains one fixed Specht line tensored with the whole multiplicity space (93) . Its S n S_n S n -span is the complete μ \mu μ -isotype. Since P n , k P_{n,k} P n , k is S n S_n S n -stable, this proves the remaining containment in (92) . □ \square □
Proof of Theorem 11.1. Take w = ( 1 , … , 1 ) w=(1,\ldots,1) w = ( 1 , … , 1 ) in Theorem 11.6. Then W j ( w ) = Sym j V W_j(w)=\operatorname{Sym}^j V W j ( w ) = Sym j V , b j ( w ) = b j b_j(w)=b_j b j ( w ) = b j , and N j ( w ) = N j N_j(w)=N_j N j ( w ) = N j . □ \square □
The proof generates monomial fillings one by one before assembling their row representations. It therefore carries every irreducible constituent and every plethystic multiplicity in (11.13) into the operator module.
Corollary 11.7 (The whole positive-weight lower boundary). For M = P , H \mathcal{M}=\mathcal{P},\mathcal{H} M = P , H , let ch μ M ( q ) \operatorname{ch}^{\mathcal{M}}_{\mu}(\mathbf{q}) ch μ M ( q ) be the row-torus character of the S μ S^{\mu} S μ multiplicity space in M n , k \mathcal{M}_{n,k} M n , k . For every w ∈ N > 0 k \mathbf{w}\in\mathbb{N}_{>0}^{k} w ∈ N > 0 k , the terms minimizing w ⋅ β \mathbf{w}\cdot\boldsymbol{\beta} w ⋅ β in the two characters are identical, including their coefficients. They are
∏ d ≥ 0 [ ∑ α ⋅ w = d q α ] s μ ( d , w ) (94) \prod_{d\geq0}\left[\sum_{\boldsymbol{\alpha}\cdot\mathbf{w}=d}q^{\boldsymbol{\alpha}}\right]^{s_{\mu}(d,\mathbf{w})} \tag*{(94)} d ≥ 0 ∏ [ α ⋅ w = d ∑ q α ] s μ ( d , w ) (94) and their common minimum is δ μ ( w ) \delta_{\mu}(\mathbf{w}) δ μ ( w ) . Consequently the generation defect has no component on any face of the μ \mu μ -character support exposed by a strictly positive weight vector.
Proof. Theorem 11.6 is equality of the corresponding row-torus modules. The character of a Schur functor of W d ( w ) W_d(\mathbf{w}) W d ( w ) is the plethystic factor in (11.14). Every face exposed by a strictly positive real vector is also exposed by a strictly positive rational vector: the relevant relatively open normal cone is defined by rational linear equalities and strict inequalities. Clearing denominators reduces that face to the integral-weight case. □
Strict positivity makes every proper monomial divisor cheaper and every positive invariant differential operator degree lowering. These are the two properties that drive the weighted proof.
Constructive generation in a linear range
The selected-label quotient now becomes an evaluation map for harmonic polynomials. Local generated lifts can be multiplied on disjoint label sets, and a two-label kernel saves one auxiliary label. These two operations prove the long-row and odd-row criteria. The hook identities then reduce the uniform range to one explicit rank-five lift.
Lemma 12.1 (Selected evaluation). Let V = K k V=K^k V = K k , B = Sym V B=\operatorname{Sym}V B = Sym V , m = B + \mathfrak{m}=B_{+} m = B + , μ = ( t , τ ) \mu=(t,\tau) μ = ( t , τ ) , ∣ τ ∣ = s |\tau|=s ∣ τ ∣ = s , and n = t + s n=t+s n = t + s . If D ≥ s D\geq s D ≥ s and n ≥ D + τ 1 n\geq D+\tau_1 n ≥ D + τ 1 , setting the auxiliary variables to zero identifies H n , k [ D ] μ S t \mathcal{H}_{n,k}[D]^{S_t}_{\mu} H n , k [ D ] μ S t with ( m ⊗ s ) D , τ (\mathfrak{m}^{\otimes s})_{D,\tau} ( m ⊗ s ) D , τ . More generally, in any range, suppose the span of selected restrictions of P n , k [ D ] S t \mathcal{P}_{n,k}[D]^{S_t} P n , k [ D ] S t contains the whole row-λ \lambda λ , selected-label-τ \tau τ component of ( m ⊗ s ) D (\mathfrak{m}^{\otimes s})_D ( m ⊗ s ) D . Then the entire row-λ \lambda λ , label-μ \mu μ harmonic component is generated.
Proof. Let A s = B ⊗ s = m ⊗ s ⊕ W A_s=B^{\otimes s}=\mathfrak{m}^{\otimes s}\oplus W A s = B ⊗ s = m ⊗ s ⊕ W , where W W W contains tensors with a unit position. The minimum-occupancy argument of Lemma 9.2 gives the selected surjection π : A s → R n , k S t \pi:A_s\to R^{S_t}_{n,k} π : A s → R n , k S t , eliminates W W W on label μ \mu μ , and identifies ( S μ ) S t (S^{\mu})^{S_t} ( S μ ) S t with S τ S^{\tau} S τ . For a selected polynomial f f f and an S t S_t S t -invariant harmonic g g g ,
⟨ π ( f ) , g ⟩ = ⟨ f , g ( X 1 , … , X s , 0 , … , 0 ) ⟩ . \langle\pi(f),g\rangle=\langle f,g(X_1,\ldots,X_s,0,\ldots,0)\rangle. ⟨ π ( f ) , g ⟩ = ⟨ f , g ( X 1 , … , X s , 0 , … , 0 )⟩ .
Restriction is therefore the Fischer adjoint of this cover and is injective. The occupancy summands are orthogonal, so on label μ \mu μ it takes values in the fully occupied summand. In the stable range Theorem 9.9 makes it an isomorphism.
For the general assertion, suppose a coinvariant class in the indicated global component annihilates all generated harmonics. Take a nonzero S t S_t S t -fixed vector in this annihilator and choose a selected representative f f f , projected to row type λ \lambda λ and selected-label type τ \tau τ . Pairing against the prescribed restrictions shows that the fully occupied part of f f f is zero. Hence f ∈ W f\in W f ∈ W . But π ( W ) \pi(W) π ( W ) has no global label μ \mu μ , a contradiction. Nondegeneracy of the coinvariant–harmonic pairing proves generation. □
We will build prescribed restrictions by multiplying harmonics on disjoint sets of labels. Each local lift uses auxiliary labels whose variables are later set to zero.
Theorem 12.2 (Long first rows). For every degree D ≥ 1 D \ge1 D ≥ 1 and every label μ ⊢ n \mu\vdash n μ ⊢ n ,
μ 1 ≥ D − 1 ⟹ P n , k [ D ] μ = H n , k [ D ] μ \mu_1 \ge D-1 \quad\Longrightarrow\quad\mathcal{P}_{n,k}[D]_\mu= \mathcal{H}_{n,k}[D]_\mu μ 1 ≥ D − 1 ⟹ P n , k [ D ] μ = H n , k [ D ] μ in every number k k k of alphabets.
Proof. For D ≤ 2 D \le2 D ≤ 2 , every row partition has length at most two. Lemma 3.1 reduces its highest-weight vectors to two alphabets, where Haiman’s operator theorem applies. The same reduction will generate the two-row local lifts below. Assume D ≥ 3 D \ge3 D ≥ 3 and put t = D − 1 t = D-1 t = D − 1 . If n < t n < t n < t there is no label in the stated range; if n = t n=t n = t , only the trivial label qualifies, and its positive-degree harmonic part is zero. We may therefore take s = n − t ≥ 1 s=n-t \ge1 s = n − t ≥ 1 selected labels.
The first selected relations have degree t + 1 = D t+1=D t + 1 = D . In that degree all their inputs are linear, so their row type is ( D ) (D) ( D ) . The selected quotient, and hence its dual harmonic restriction, is an isomorphism on every other row type. We will generate all of these row components with exactly D − 1 D-1 D − 1 auxiliary labels.
Fix an occupied selected multidegree d 1 + ⋯ + d r = D d_1+\cdots+d_r=D d 1 + ⋯ + d r = D , with each d i > 0 d_i>0 d i > 0 , and write
T = ⨂ i = 1 r Sym d i V , K i j = ker ( Sym d i V ⊗ Sym d j V ⟶ Sym d i + d j V ) . T=\bigotimes_{i=1}^{r}\operatorname{Sym}^{d_i}V,\qquad K_{ij}=\ker\left(\operatorname{Sym}^{d_i}V\otimes\operatorname{Sym}^{d_j}V\longrightarrow\operatorname{Sym}^{d_i+d_j}V\right). T = i = 1 ⨂ r Sym d i V , K ij = ker ( Sym d i V ⊗ Sym d j V ⟶ Sym d i + d j V ) .
The multiplication map gives
T / ∑ i < j K i j ⊗ ⨂ h ≠ i , j Sym d h V ≅ Sym D V . (95) T\mathbin{\Big/}\sum_{i<j}K_{ij}\otimes\bigotimes_{h\ne i,j}\operatorname{Sym}^{d_h}V\cong\operatorname{Sym}^{D}V. \tag*{(95)} T / i < j ∑ K ij ⊗ h = i , j ⨂ Sym d h V ≅ Sym D V . (95)
Indeed, realize T T T as the quotient of V ⊗ D V^{\otimes D} V ⊗ D by permutations within its blocks of sizes d i d_i d i . Quotienting by K i j K_{ij} K ij imposes symmetry on the union of the i i i th and j j j th blocks. These block-pair permutations generate all of S D S_D S D . Thus the pair kernels span every row component of T T T other than ( D ) (D) ( D ) .
Put d = d i + d j d=d_i+d_j d = d i + d j . By Pieri, K i j K_{ij} K ij has at most two rows and contains no copy of ( d ) (d) ( d ) . At local rank d + 1 d+1 d + 1 , take two selected labels and d − 1 d-1 d − 1 auxiliary labels. Its first selected relation has degree d d d and row type ( d ) (d) ( d ) , so restriction has a harmonic lift of every vector in K i j K_{ij} K ij . Haiman’s theorem generates these lifts. Each remaining selected factor Sym d h V \operatorname{Sym}^{d_h}V Sym d h V has a generated lift with d h d_h d h auxiliary labels: in one row, differentiate the Vandermonde on x , y 0 , … , y d h − 1 x,y_0,\ldots,y_{d_h-1} x , y 0 , … , y d h − 1 by ∏ q = 0 d h − 1 ∂ y q \prod_{q=0}^{d_h-1}\partial_{y_q} ∏ q = 0 d h − 1 ∂ y q and set the auxiliary labels to zero. The result is a nonzero multiple of x d h x^{d_h} x d h ; row polarization gives the whole symmetric power.
The total number of auxiliary labels is
( d − 1 ) + ∑ h ≠ i , j d h = D − 1. (d-1)+\sum_{h\ne i,j}d_h=D-1. ( d − 1 ) + h = i , j ∑ d h = D − 1.
Multiply these lifts on disjoint label sets using Theorem 11.4, and average over the auxiliary labels; unused selected labels contribute constants. Restriction sets every auxiliary label to zero, so averaging preserves the prescribed selected polynomial. Equation (12.1), summed over selected multidegrees, now generates every non-one-row harmonic component fixed by S t S_t S t . The ordinary operator theorem and row polarization generate the one-row component. Injectivity of harmonic restriction gives P n , k [ D ] S t = H n , k [ D ] S t \mathcal{P}_{n,k}[D]^{S_t}=\mathcal{H}_{n,k}[D]^{S_t} P n , k [ D ] S t = H n , k [ D ] S t . Every S μ S^\mu S μ with μ 1 ≥ t \mu_1\ge t μ 1 ≥ t has a nonzero S t S_t S t -fixed vector by branching. Exactness of invariants therefore gives the stated equality on every such label. □
Alternating pairs save one auxiliary label each. Placing such pairs in the columns of a label tail makes the savings compatible with its Young symmetrizer.
Theorem 12.3 (Odd-row generation criterion). For every μ ⊢ n \mu\vdash n μ ⊢ n ,
D ≤ μ 1 + μ 3 + μ 5 + ⋯ ⟹ P n , k [ D ] μ = H n , k [ D ] μ . D\le\mu_1+\mu_3+\mu_5+\cdots\quad\Longrightarrow\quad\mathcal{P}_{n,k}[D]_\mu=\mathcal{H}_{n,k}[D]_\mu. D ≤ μ 1 + μ 3 + μ 5 + ⋯ ⟹ P n , k [ D ] μ = H n , k [ D ] μ .
Proof. Write μ = ( t , τ ) \mu=(t,\tau) μ = ( t , τ ) and pair adjacent boxes vertically in every column of τ \tau τ . The number of pairs is p = τ 2 + τ 4 + ⋯ p=\tau_2+\tau_4+\cdots p = τ 2 + τ 4 + ⋯ . An alternating pair of positive-degree polynomials of total degree d d d has a generated lift on d + 1 d+1 d + 1 labels, with two selected labels. Indeed, its selected-sign component has no degree-d d d relation: the first selected relations have degree d d d and are S 2 S_2 S 2 -invariant. Its row types have length at most two, so Haiman’s operator theorem generates the harmonic lift. Such a pair costs d − 1 d-1 d − 1 auxiliary labels; a singleton of degree d d d costs d d d .
The disjoint products therefore use D − p D-p D − p auxiliary labels. Applying the Young symmetrizer for τ \tau τ spans its whole fully occupied selected component: each prealternated pair multiplies the column antisymmetrization by two. Averaging over the auxiliary labels and Lemma 12.1 prove generation whenever t ≥ D − p t \ge D-p t ≥ D − p , as asserted. □
Lemma 12.4 (Four universal tail degrees). The alternating component is generated in every degree d ≤ n + 2 d \le n+2 d ≤ n + 2 . Every hook label of tail s s s is generated in degrees d ≤ s + 3 d \le s+3 d ≤ s + 3 . More generally, if μ = τ [ n ] \mu=\tau[n] μ = τ [ n ] , s = ∣ τ ∣ s=|\tau| s = ∣ τ ∣ , then
0 ≤ D − s ≤ 3 , n ≥ D + τ 1 ⟹ P n , k [ D ] μ = H n , k [ D ] μ . 0 \le D-s \le3,\qquad n \ge D+\tau_1 \quad\Longrightarrow\quad\mathcal{P}_{n,k}[D]_\mu=\mathcal{H}_{n,k}[D]_\mu. 0 ≤ D − s ≤ 3 , n ≥ D + τ 1 ⟹ P n , k [ D ] μ = H n , k [ D ] μ .
Proof. The ambient sign multiplicity is ⋀ n ( Sym V ) \bigwedge^{n}(\operatorname{Sym} V) ⋀ n ( Sym V ) . Through degree n + 2 n+2 n + 2 , its nonlinear factors, after removing exterior powers of V V V and the possible constant polynomial, are among
Sym 2 V , Sym 3 V , ⋀ 2 Sym 2 V , Sym 4 V , Sym 3 V ⊗ Sym 2 V , ⋀ 3 Sym 2 V . \operatorname{Sym}^{2}V,\quad\operatorname{Sym}^{3}V,\quad\bigwedge^{2}\operatorname{Sym}^{2}V,\quad\operatorname{Sym}^{4}V,\quad\operatorname{Sym}^{3}V\otimes\operatorname{Sym}^{2}V,\quad\bigwedge^{3}\operatorname{Sym}^{2}V. Sym 2 V , Sym 3 V , ⋀ 2 Sym 2 V , Sym 4 V , Sym 3 V ⊗ Sym 2 V , ⋀ 3 Sym 2 V .
The last two decompose as S 5 V ⊕ S 41 V ⊕ S 32 V S_{5}V\oplus S_{41}V\oplus S_{32}V S 5 V ⊕ S 41 V ⊕ S 32 V and S 411 V ⊕ S 33 V S_{411}V\oplus S_{33}V S 411 V ⊕ S 33 V , respectively; ⋀ 2 Sym 2 V = S 31 V \bigwedge^{2}\operatorname{Sym}^{2}V=S_{31}V ⋀ 2 Sym 2 V = S 31 V . Vertical Pieri shows that none of these, after tensoring with an exterior power of V V V , has both λ 2 ≥ 3 \lambda_2 \ge3 λ 2 ≥ 3 and λ 3 ≥ 2 \lambda_3 \ge2 λ 3 ≥ 2 . Corollary 6.16 therefore rules out a sign defect. The exact hook-defect identities give the hook assertion.
Now consider a column of τ \tau τ of height c c c and degree d d d . Since the total excess is at most three, c ≤ d ≤ c + 3 c \le d \le c+3 c ≤ d ≤ c + 3 . The stable evaluation statement at local rank d + 1 d+1 d + 1 identifies its fully occupied alternating component with the hook ( d + 1 − c , 1 c ) (d+1-c,1^c) ( d + 1 − c , 1 c ) , which is generated by the assertion just proved. It costs d + 1 − c d+1-c d + 1 − c auxiliary labels. Summing over columns costs D − s + τ 1 D-s+\tau_1 D − s + τ 1 . Disjoint products, Young symmetrization and Lemma 12.1 prove the final claim. □
Theorem 12.5 (Uniform linear generation). For D ≥ 3 D \ge3 D ≥ 3 and n ≥ 2 D − 3 n \ge2D-3 n ≥ 2 D − 3 , in every number of alphabets,
P n , k [ d ] = H n , k [ d ] ( 0 ≤ d ≤ D ) . \mathcal{P}_{n,k}[d]=\mathcal{H}_{n,k}[d]\qquad(0\le d\le D). P n , k [ d ] = H n , k [ d ] ( 0 ≤ d ≤ D ) .
Proof. First prove the assertion in degree D D D . Write μ = ( t , τ ) \mu=(t,\tau) μ = ( t , τ ) , s = ∣ τ ∣ s=|\tau| s = ∣ τ ∣ . Occupancy gives s ≤ D s\le D s ≤ D . If t ≥ D − 1 t\ge D-1 t ≥ D − 1 , apply Theorem 12.2 . Otherwise t ≤ D − 2 t\le D-2 t ≤ D − 2 , and n ≥ 2 D − 3 n\ge2D-3 n ≥ 2 D − 3 forces s ≥ D − 1 s\ge D-1 s ≥ D − 1 . For s = D s=D s = D , Lemma 12.4 applies because n = D + t ≥ D + τ 1 n=D+t\ge D+\tau_1 n = D + t ≥ D + τ 1 . For s = D − 1 s=D-1 s = D − 1 , necessarily t = D − 2 t=D-2 t = D − 2 and n = 2 D − 3 n=2D-3 n = 2 D − 3 . The same lemma applies if τ 1 ≤ D − 3 \tau_1\le D-3 τ 1 ≤ D − 3 . The sole remaining label is therefore
μ = ( D − 2 , D − 2 , 1 ) . \mu=(D-2,D-2,1). μ = ( D − 2 , D − 2 , 1 ) .
Put s = D − 1 s=D-1 s = D − 1 . We reach its one exceptional row by combining a three-selected-label harmonic at rank five with linear lifts. The general part of Lemma 12.1 lets us work in the fully occupied selected space: spanning that space by restrictions suffices at this rank as well. The degree-s + 1 s+1 s + 1 part of S ( s − 1 , 1 ) ( m ) S_{(s-1,1)}(\mathfrak{m}) S ( s − 1 , 1 ) ( m ) has one quadratic slot. Its only row type of length greater than two which is not a row hook is ( s − 2 , 2 , 1 ) (s-2,2,1) ( s − 2 , 2 , 1 ) with multiplicity one; it occurs only for s ≥ 4 s\ge4 s ≥ 4 . This follows by removing one corner of ( s − 1 , 1 ) (s-1,1) ( s − 1 , 1 ) and applying the horizontal Pieri rule with S 2 V S_2V S 2 V . For s < 4 s<4 s < 4 , Haiman’s theorem of row-hook generation therefore finish the proof. Assume s ≥ 4 s\ge4 s ≥ 4 and dim V ≥ 3 \dim V\ge3 dim V ≥ 3 , so the exceptional row is present.
At rank five, a local selected-three vector of selected type ( 2 , 1 ) (2,1) ( 2 , 1 ) and row type ( 2 , 1 , 1 ) (2,1,1) ( 2 , 1 , 1 ) has a harmonic lift: the degree-four relations have at most two rows. Indeed, they come either from a degree-three relation multiplied by a linear form, with row source Sym 3 V ⊗ V \operatorname{Sym}^{3}V\otimes V Sym 3 V ⊗ V , or from a relation on two linear inputs and one quadratic input, with row source Sym 2 V ⊗ Sym 2 V \operatorname{Sym}^{2}V\otimes\operatorname{Sym}^{2}V Sym 2 V ⊗ Sym 2 V . Neither source contains ( 2 , 1 , 1 ) (2,1,1) ( 2 , 1 , 1 ) . Proposition 4.3 generates this lift. Multiply it with s − 3 s-3 s − 3 linear lifts. The number of auxiliary labels is 2 + ( s − 3 ) = s − 1 = D − 2 2+(s-3)=s-1=D-2 2 + ( s − 3 ) = s − 1 = D − 2 .
The local row projection and selected Young projection must both survive. In the realization S ( s − 1 , 1 ) m = ker ( Sym s − 1 m ⊗ m → Sym s m ) S_{(s-1,1)}\mathfrak{m}=\ker(\operatorname{Sym}^{s-1}\mathfrak{m}\otimes\mathfrak{m}\to\operatorname{Sym}^{s}\mathfrak{m}) S ( s − 1 , 1 ) m = ker ( Sym s − 1 m ⊗ m → Sym s m ) , the selected projection is
h ⊗ ( w ⊗ ( a ∧ b ) ) ⟼ ( h w a ) ⊗ b − ( h w b ) ⊗ a . (96) h\otimes(w\otimes(a\wedge b))\longmapsto(hwa)\otimes b-(hwb)\otimes a. \tag*{(96)} h ⊗ ( w ⊗ ( a ∧ b )) ⟼ ( h w a ) ⊗ b − ( h w b ) ⊗ a . (96)
It symmetrizes the first row and preserves the distinguished last box. The quadratic factor w ∈ Sym 2 V w \in\operatorname{Sym}^{2} V w ∈ Sym 2 V varies independently, so one can check its x 2 x^{2} x 2 component separately. Set W = Sym 2 V W = \operatorname{Sym}^{2} V W = Sym 2 V and choose independent x , y , z ∈ V x,y,z \in V x , y , z ∈ V . In W ⊗ ⋀ 2 V W \otimes\bigwedge^{2} V W ⊗ ⋀ 2 V put
v = x 2 ⊗ ( y ∧ z ) − x y ⊗ ( x ∧ z ) + x z ⊗ ( x ∧ y ) . v=x^{2}\otimes(y\wedge z)-xy\otimes(x\wedge z)+xz\otimes(x\wedge y). v = x 2 ⊗ ( y ∧ z ) − x y ⊗ ( x ∧ z ) + x z ⊗ ( x ∧ y ) .
This is a highest vector of type ( 2 , 1 , 1 ) (2,1,1) ( 2 , 1 , 1 ) , and
E 21 v = x y ⊗ ( y ∧ z ) − y 2 ⊗ ( x ∧ z ) + y z ⊗ ( x ∧ y ) . E_{21}v=xy\otimes(y\wedge z)-y^{2}\otimes(x\wedge z)+yz\otimes(x\wedge y). E 21 v = x y ⊗ ( y ∧ z ) − y 2 ⊗ ( x ∧ z ) + y z ⊗ ( x ∧ y ) .
Thus
x s − 4 y ⊗ v − x s − 3 ⊗ E 21 v x^{s-4}y\otimes v-x^{s-3}\otimes E_{21}v x s − 4 y ⊗ v − x s − 3 ⊗ E 21 v
is highest of type ( s − 2 , 2 , 1 ) (s-2,2,1) ( s − 2 , 2 , 1 ) . The image of the displayed highest vector has nonzero w = x 2 w=x^{2} w = x 2 component
x 2 ⊗ ( x s − 4 y 2 ⊗ z − x s − 4 y z ⊗ y ) . x^{2}\otimes\left(x^{s-4}y^{2}\otimes z-x^{s-4}yz\otimes y\right). x 2 ⊗ ( x s − 4 y 2 ⊗ z − x s − 4 y z ⊗ y ) .
The second summand has no such quadratic factor, so it cannot cancel. The target is Sym 2 V ⊗ Sym s − 2 V ⊗ V \operatorname{Sym}^{2}V\otimes\operatorname{Sym}^{s-2}V\otimes V Sym 2 V ⊗ Sym s − 2 V ⊗ V ; its first factor records the quadratic factor. Thus the selected Young projection is nonzero on the generated products. Row polarization reaches the whole multiplicity-one component, and averaging over auxiliary labels preserves its selected restriction. Lemma 12.1 gives generation of this last row-and-label component. All remaining rows are generated by Haiman or row-hook generation.
This proves degree D D D . For 3 ≤ d ≤ D 3\le d\le D 3 ≤ d ≤ D the same hypothesis gives n ≥ 2 d − 3 n\ge2d-3 n ≥ 2 d − 3 ; degrees zero, one and two follow from the ordinary and bivariate operator theorems. □
The labelwise criteria also locate every possible remaining defect. If ( H n , k [ D ] / P n , k [ D ] ) μ ≠ 0 (\mathcal{H}_{n,k}[D]/\mathcal{P}_{n,k}[D])_{\mu}\ne0 ( H n , k [ D ] / P n , k [ D ] ) μ = 0 , then
n − D + 1 ≤ μ 1 ≤ D − 2 , μ 1 + μ 3 + μ 5 + ⋯ ≤ D − 1. n-D+1\le\mu_{1}\le D-2,\qquad\mu_{1}+\mu_{3}+\mu_{5}+\cdots\le D-1. n − D + 1 ≤ μ 1 ≤ D − 2 , μ 1 + μ 3 + μ 5 + ⋯ ≤ D − 1.
Indeed, occupancy gives n − μ 1 ≤ D n-\mu_{1}\le D n − μ 1 ≤ D . Equality would make degree D D D a first occurrence, which is generated by Theorem 11.1 ; the other bounds are Theorems 12.2 and 12.3 .
Appendix A. The cubic multiplication at the character boundary
The proof of Theorem 10.10 leaves one possible multiplicity-one cubic source. We exhibit a highest vector whose image survives the quadratic quotient. The dual-number specialization below makes all unwanted moments vanish and leaves only one term of that image.
Lemma A.1 (A nonzero cubic moment multiplication). For a ≥ 3 a\ge3 a ≥ 3 , put n = 2 a n=2a n = 2 a , D = a + 3 D=a+3 D = a + 3 , and
τ = ( a − 1 , 1 ) , μ = ( a , a − 1 , 1 ) , λ = ( a , 2 , 1 ) . \tau=(a-1,1),\qquad\mu=(a,a-1,1),\qquad\lambda=(a,2,1). τ = ( a − 1 , 1 ) , μ = ( a , a − 1 , 1 ) , λ = ( a , 2 , 1 ) . For dim V ≥ 3 \dim V\ge3 dim V ≥ 3 , the cubic multiplication map
Sym 3 V ⊗ Hom S n ( S μ , Q n , k [ a ] ) ⟶ Hom S n ( S μ , Q n , k [ a + 3 ] ) \operatorname{Sym}^{3}V\otimes\operatorname{Hom}_{S_{n}}(S^{\mu},Q_{n,k}[a])\longrightarrow\operatorname{Hom}_{S_{n}}(S^{\mu},Q_{n,k}[a+3]) Sym 3 V ⊗ Hom S n ( S μ , Q n , k [ a ]) ⟶ Hom S n ( S μ , Q n , k [ a + 3 ]) is nonzero, and hence injective, on its multiplicity-one S λ V S_{\lambda}V S λ V summand. Here Q n , k Q_{n,k} Q n , k denotes the quotient by linear and quadratic moments.
Proof. It is enough to work in three rows x , y , z x,y,z x , y , z , because the map is a natural transformation of polynomial functors. Set m = a − 1 m=a-1 m = a − 1 and choose disjoint label blocks of sizes 3 , 2 , … , 2 , 1 3,2,\ldots,2,1 3 , 2 , … , 2 , 1 , with a − 2 a-2 a − 2 blocks of size two. Define
F = det ( 1 x 0 y 0 1 x 1 y 1 1 x 2 y 2 ) ∏ j = 1 a − 2 ( x 2 j + 1 − x 2 j + 2 ) . F=\det\begin{pmatrix}1&x_{0}&y_{0}\\1&x_{1}&y_{1}\\1&x_{2}&y_{2}\end{pmatrix}\prod_{j=1}^{a-2}(x_{2j+1}-x_{2j+2}). F = det 1 1 1 x 0 x 1 x 2 y 0 y 1 y 2 j = 1 ∏ a − 2 ( x 2 j + 1 − x 2 j + 2 ) .
The unused label is $2a-1$. This polynomial has row highest weight ( m , 1 ) (m,1) ( m , 1 ) , since ∑ i x i ∂ y i \sum_i x_i\partial_{y_i} ∑ i x i ∂ y i kills its determinant. It belongs to label μ \mu μ . To justify this last assertion directly, its column antisymmetry forces every label constituent to be dominated by μ \mu μ , hence to have first row at most a a a ; degree a a a forces a first row at least a a a . At first row exactly a a a , the row multiplicity in degree a a a is S ν V S_{\nu}V S ν V for label ( a , ν ) (a,\nu) ( a , ν ) , by the fully occupied, linear part of Schur–Weyl and Pieri branching (equivalently, Theorem 9.9 in its linear degree). The row highest weight $(a-1,1) therefore selects exactly μ \mu μ . All invariant differential operators annihilate this first occurrence, so F F F survives the quadratic quotient.
Write p i j k = ∑ s x s i y s j z s k p_{ijk}=\sum_s x_s^i y_s^j z_s^k p ij k = ∑ s x s i y s j z s k , and put
A = ∑ s y s ∂ x s , B = ∑ s z s ∂ x s , C = ∑ s z s ∂ y s . A=\sum_s y_s\partial_{x_s},\qquad B=\sum_s z_s\partial_{x_s},\qquad C=\sum_s z_s\partial_{y_s}. A = s ∑ y s ∂ x s , B = s ∑ z s ∂ x s , C = s ∑ z s ∂ y s .
The row weight ( a , 2 , 1 ) (a,2,1) ( a , 2 , 1 ) and the two simple raising equations determine the following combination of cubic moments and lowered copies of F F F :
G = p 300 ( A B − A 2 C ) F − ( m − 2 ) p 210 B F + ( 2 m − 1 ) p 210 A C F − ( m + 1 ) p 201 A F − ( m 2 − 1 ) p 120 C F + ( m 2 − 1 ) p 111 F . G=p_{300}(AB-A^2C)F-(m-2)p_{210}BF+(2m-1)p_{210}ACF
-(m+1)p_{201}AF-(m^2-1)p_{120}CF+(m^2-1)p_{111}F. G = p 300 ( A B − A 2 C ) F − ( m − 2 ) p 210 B F + ( 2 m − 1 ) p 210 A C F − ( m + 1 ) p 201 A F − ( m 2 − 1 ) p 120 C F + ( m 2 − 1 ) p 111 F .
It has weight ( a , 2 , 1 ) (a,2,1) ( a , 2 , 1 ) and is highest. For completeness, if X = ∑ s x s ∂ y s X=\sum_s x_s\partial_{y_s} X = ∑ s x s ∂ y s and Z = ∑ s y s ∂ z s Z=\sum_s y_s\partial_{z_s} Z = ∑ s y s ∂ z s , then
X A F = ( m − 1 ) F , X B F = − C F , X A C F = m C F , XAF=(m-1)F,\qquad XBF=-CF,\qquad XACF=mCF, X A F = ( m − 1 ) F , X B F = − C F , X A C F = m C F ,
X ( A B − A 2 C ) F = ( m − 2 ) B F − ( 2 m − 1 ) A C F , X(AB-A^2C)F=(m-2)BF-(2m-1)ACF, X ( A B − A 2 C ) F = ( m − 2 ) B F − ( 2 m − 1 ) A C F ,
while Z B F = Z A C F = A F ZBF=ZACF=AF Z B F = Z A C F = A F , Z C F = F ZCF=F Z C F = F , Z A F = 0 ZAF=0 Z A F = 0 , and Z ( A B − A 2 C ) F = 0 Z(AB-A^2C)F=0 Z ( A B − A 2 C ) F = 0 . The corresponding identities on the cubic moments show X G = Z G = 0 XG=ZG=0 X G = Z G = 0 . These are the simple raising operators of GL 3 \operatorname{GL}_3 GL 3 . Every operator used preserves the label μ \mu μ , and Pieri’s rule gives the indicated row summand once.
We show that G G G survives the quadratic quotient by an exact specialization over K ( ζ ) [ ϵ ] / ( ϵ 2 ) K(\zeta)[\epsilon]/(\epsilon^2) K ( ζ ) [ ϵ ] / ( ϵ 2 ) , where ζ \zeta ζ is a primitive 2 a 2a 2 a th root of unity. The frequencies 1 1 1 , a − 1 a-1 a − 1 , a a a sum to 2 a 2a 2 a for x y z xyz x y z . Every linear or quadratic frequency is nonzero modulo 2 a 2a 2 a , except that of z 2 z^2 z 2 , whose coefficient vanishes because ϵ 2 = 0 \epsilon^2=0 ϵ 2 = 0 . Set, for 0 ≤ i < 2 a 0\leq i<2a 0 ≤ i < 2 a ,
x i = ζ i , y i = ζ ( a − 1 ) i , z i = ϵ ζ a i . x_i=\zeta^i,\qquad y_i=\zeta^{(a-1)i},\qquad z_i=\epsilon\zeta^{ai}. x i = ζ i , y i = ζ ( a − 1 ) i , z i = ϵ ζ ai .
Every linear or quadratic moment is zero. For quadratic moments containing two z z z ’s this follows from ϵ 2 = 0 \epsilon^2=0 ϵ 2 = 0 ; every other vanishing is a geometric sum with exponent not divisible by 2 a 2a 2 a . The same observation gives
p 300 = p 210 = p 201 = p 120 = 0 , p 111 = 2 a ϵ . p_{300}=p_{210}=p_{201}=p_{120}=0,\qquad p_{111}=2a\epsilon. p 300 = p 210 = p 201 = p 120 = 0 , p 111 = 2 a ϵ .
All factors of F F F are nonzero. Its first determinant is ζ 2 − ζ − 2 \zeta^2-\zeta^{-2} ζ 2 − ζ − 2 , which is nonzero for a ≥ 3 a\geq3 a ≥ 3 , and each remaining factor is a difference of distinct roots of unity. Consequently
G = 2 a ( m 2 − 1 ) ϵ F ≠ 0. G=2a(m^2-1)\epsilon F\ne0. G = 2 a ( m 2 − 1 ) ϵ F = 0.
The specialization kills the quadratic ideal, so G G G is nonzero in Q n , 3 Q_{n,3} Q n , 3 . Characteristic-zero scalar descent and functoriality prove the assertion for all V V V . □
Appendix B. Eulerian ideals and bounded exponents
We prove the coinvariant algebra statements used in § 8 . The classical descent filtration and the Gram-ideal lemma give a short proof of the complementary annihilator identity inside the ordinary coinvariant algebra.
Put P n = K [ x 1 , … , x n ] P_n=K[x_1,\ldots,x_n] P n = K [ x 1 , … , x n ] , p s = ∑ i x i s p_s=\sum_i x_i^s p s = ∑ i x i s , I n = ( p 1 , … , p n ) I_n=(p_1,\ldots,p_n) I n = ( p 1 , … , p n ) and C n = P n / I n C_n=P_n/I_n C n = P n / I n . For a label set I I I , ordered increasingly, write
x I = ∏ i ∈ I x i , Δ I = det ( x i a − 1 ) 1 ≤ a ≤ ∣ I ∣ , i ∈ I . x_I=\prod_{i\in I}x_i,\qquad\Delta_I=\det(x_i^{a-1})_{1\leq a\leq|I|,\ i\in I}. x I = i ∈ I ∏ x i , Δ I = det ( x i a − 1 ) 1 ≤ a ≤ ∣ I ∣ , i ∈ I .
Thus the ideal J n , d J_{n,d} J n , d of § 8 is generated by g I = x I Δ I g_I=x_I\Delta_I g I = x I Δ I for ∣ I ∣ = d |I|=d ∣ I ∣ = d . Also put J n , 0 = C n J_{n,0}=C_n J n , 0 = C n and J n , n = 0 J_{n,n}=0 J n , n = 0 . For d ≥ 1 d\geq1 d ≥ 1 , let
B n , d = span { x α : 0 ≤ α i < d } , F n , d = image ( B n , d ⟶ C n ) . B_{n,d}=\operatorname{span}\{x^\alpha:0\leq\alpha_i<d\},\qquad F_{n,d}=\operatorname{image}(B_{n,d}\longrightarrow C_n). B n , d = span { x α : 0 ≤ α i < d } , F n , d = image ( B n , d ⟶ C n ) .
Proposition B.1 (Eulerian complements and annihilators). For 1 ≤ d ≤ n 1 \le d \le n 1 ≤ d ≤ n , the quotient projection induces a graded S n S_n S n -equivariant isomorphism
F n , d → ∼ C n / J n , d . (97) F_{n,d} \xrightarrow{\sim} C_n/J_{n,d}. \tag*{(97)} F n , d ∼ C n / J n , d . (97) Moreover, for 0 ≤ d ≤ n 0 \le d \le n 0 ≤ d ≤ n ,
Frob ( J n , d ; q ) = ∑ λ ⊢ n s λ ∑ T ∈ SYT ( λ ) des ( T ) ≥ d q maj ( T ) , (98) \operatorname{Frob}(J_{n,d};q)=\sum_{\lambda\vdash n}s_\lambda\sum_{\substack{T\in\operatorname{SYT}(\lambda)\\ \operatorname{des}(T)\ge d}}q^{\operatorname{maj}(T)}, \tag*{(98)} Frob ( J n , d ; q ) = λ ⊢ n ∑ s λ T ∈ SYT ( λ ) des ( T ) ≥ d ∑ q maj ( T ) , (98) Ann C n ( J n , d ) = J n , n − d . (99) \operatorname{Ann}_{C_n}(J_{n,d})=J_{n,n-d}. \tag*{(99)} Ann C n ( J n , d ) = J n , n − d . (99) On the coinvariant differential forms
Ω n = P n ⊗ Λ ( θ 1 , … , θ n ) ( p 1 , … , p n , d R p 1 , … , d R p n ) , δ j = ∑ i x i j ∂ θ i ( j ≥ 1 ) , \Omega_n=\frac{P_n\otimes\Lambda(\theta_1,\ldots,\theta_n)}{(p_1,\ldots,p_n,d_Rp_1,\ldots,d_Rp_n)},\qquad\delta_j=\sum_i x_i^j\partial_{\theta_i}\quad(j\ge1), Ω n = ( p 1 , … , p n , d R p 1 , … , d R p n ) P n ⊗ Λ ( θ 1 , … , θ n ) , δ j = i ∑ x i j ∂ θ i ( j ≥ 1 ) , the map D d = δ d ⋯ δ 1 : Ω n d → C n D_d=\delta_d\cdots\delta_1:\Omega_n^d\to C_n D d = δ d ⋯ δ 1 : Ω n d → C n has image J n , d J_{n,d} J n , d for 0 ≤ d < n 0\le d<n 0 ≤ d < n ; D 0 D_0 D 0 means the identity. Its apolar adjoint is Θ 1 ⋯ Θ d \Theta_1\cdots\Theta_d Θ 1 ⋯ Θ d .
Proof. Determinants and straightening. For 1 ≤ d ≤ n 1\le d\le n 1 ≤ d ≤ n , the two determinant presentations agree:
I n + ( Δ I : ∣ I ∣ = d + 1 ) = I n + ( x I Δ I : ∣ I ∣ = d ) . (100) I_n+(\Delta_I:|I|=d+1)=I_n+(x_I\Delta_I:|I|=d). \tag*{(100)} I n + ( Δ I : ∣ I ∣ = d + 1 ) = I n + ( x I Δ I : ∣ I ∣ = d ) . (100) Expansion along the constant row gives the inclusion from left to right. Conversely, append ( 1 , x j , … , x j d ) t (1,x_j,\ldots,x_j^d)^t ( 1 , x j , … , x j d ) t to the d d d columns indexed by a fixed I I I and sum the resulting determinants over j ∈ [ n ] j\in[n] j ∈ [ n ] . Each summand is zero or a signed ( d + 1 ) (d+1) ( d + 1 ) -label Vandermonde. The last column of the sum is ( n , p 1 , … , p d ) t (n,p_1,\ldots,p_d)^t ( n , p 1 , … , p d ) t , so modulo I n I_n I n the sum is ± n x I Δ I \pm nx_I\Delta_I ± n x I Δ I . This proves the reverse inclusion, and also proves g [ n ] = 0 g_{[n]}=0 g [ n ] = 0 in C n C_n C n when d = n d=n d = n .
We claim that every polynomial is congruent to an element of B n , d B_{n,d} B n , d modulo L n , d = I n + ( Δ I : ∣ I ∣ = d + 1 ) L_{n,d}=I_n+(\Delta_I:|I|=d+1) L n , d = I n + ( Δ I : ∣ I ∣ = d + 1 ) . Induct on d d d for all n n n , and then descend on monomial support for fixed ( n , d ) (n,d) ( n , d ) . If d = 1 d=1 d = 1 , the presentation above puts every x i x_i x i in L n , 1 L_{n,1} L n , 1 . If n ≤ d n\le d n ≤ d , the Artin basis, with exponents 0 ≤ α i < i 0\le\alpha_i<i 0 ≤ α i < i , suffices. Otherwise write a monomial of nonempty support I I I as x I f x_If x I f . Apply the induction hypothesis for ( ∣ I ∣ , d − 1 ) (|I|,d-1) ( ∣ I ∣ , d − 1 ) in K [ x i : i ∈ I ] K[x_i:i\in I] K [ x i : i ∈ I ] :
f = b + ∑ s = 1 ∣ I ∣ p s ( I ) h s + ∑ T ⊆ I ∣ T ∣ = d Δ T u T , p s ( I ) = ∑ i ∈ I x i s , f=b+\sum_{s=1}^{|I|}p_s(I)h_s+\sum_{\substack{T\subseteq I\\|T|=d}}\Delta_Tu_T,\qquad p_s(I)=\sum_{i\in I}x_i^s, f = b + s = 1 ∑ ∣ I ∣ p s ( I ) h s + T ⊆ I ∣ T ∣ = d ∑ Δ T u T , p s ( I ) = i ∈ I ∑ x i s , where every exponent in b b b is at most d − 2 d-2 d − 2 and all coefficients use only variables in I I I . After multiplication by x I x_I x I , the first term is bounded as required and every determinant term belongs to L n , d L_{n,d} L n , d by (B.4). Finally,
x I p s ( I ) h s ≡ − ∑ j ∉ I x j x I s h s ( m o d I n ) . x_Ip_s(I)h_s\equiv-\sum_{j\notin I}x_jx_I^sh_s\pmod{I_n}. x I p s ( I ) h s ≡ − j ∈ / I ∑ x j x I s h s ( mod I n ) . Each monomial on the right has larger support; for I = [ n ] I=[n] I = [ n ] the sum is zero. The empty-support monomial is 1 1 1 . This completes both inductions. Taking homogeneous components throughout proves that F n , d → C n / J n , d F_{n,d}\to C_n/J_{n,d} F n , d → C n / J n , d is a graded S n S_n S n -equivariant surjection.
The descent filtration. The classical descent straightening theorem identifies
F n , d = span { b w = ∏ i ∈ Des ( w ) ( x w ( 1 ) ⋯ x w ( i ) ) : des ( w ) < d } . F_{n,d}=\operatorname{span}\left\{b_w=\prod_{i\in\operatorname{Des}(w)}(x_{w(1)}\cdots x_{w(i)}):\operatorname{des}(w)<d\right\}. F n , d = span ⎩ ⎨ ⎧ b w = i ∈ Des ( w ) ∏ ( x w ( 1 ) ⋯ x w ( i ) ) : des ( w ) < d ⎭ ⎬ ⎫ . These descent monomials belong to the Garsia–Stanton basis of C n C_n C n , and the descent-representation refinement gives
Frob ( F n , d ; q ) = ∑ λ ⊢ n s λ ∑ T ∈ SYT ( λ ) des ( T ) < d q maj ( T ) . (101) \operatorname{Frob}(F_{n,d};q)=\sum_{\lambda\vdash n}s_\lambda\sum_{\substack{T\in\operatorname{SYT}(\lambda)\\ \operatorname{des}(T)<d}}q^{\operatorname{maj}(T)}. \tag*{(101)} Frob ( F n , d ; q ) = λ ⊢ n ∑ s λ T ∈ SYT ( λ ) des ( T ) < d ∑ q maj ( T ) . (101)
See [1 Theorems 3.12 and 4.1] for the descent representations and [17 Section 4, Theorem 4.3 and the paragraph following it] for the maximum-exponent realization. These statements are defined over Q \mathbb{Q} Q and hence hold over K K K . Put
r d = dim J n , d , R d = # { w ∈ S n : des ( w ) ≥ d } . r_d=\dim J_{n,d},\qquad R_d=\#\{w\in S_n:\operatorname{des}(w)\ge d\}. r d = dim J n , d , R d = # { w ∈ S n : des ( w ) ≥ d } .
The surjection just proved gives r d ≥ R d r_d\ge R_d r d ≥ R d .
The Gram ideal and complementary dimensions. Newton’s identities give p s = 0 p_s=0 p s = 0 in C n C_n C n for every s > 0 s>0 s > 0 . The standard inner product is nondegenerate on W n = { ( z i ) ∈ K n : ∑ i z i = 0 } W_n=\{(z_i)\in K^n:\sum_i z_i=0\} W n = {( z i ) ∈ K n : ∑ i z i = 0 } , since n ≠ 0 n\ne0 n = 0 in K K K . Thus
v s = ( x 1 s , … , x n s ) ∈ C n ⊗ W n , ⟨ v s , v t ⟩ = p s + t = 0 ( s , t ≥ 1 ) . v_s=(x_1^s,\ldots,x_n^s)\in C_n\otimes W_n,\qquad\langle v_s,v_t\rangle=p_{s+t}=0\quad(s,t\ge1). v s = ( x 1 s , … , x n s ) ∈ C n ⊗ W n , ⟨ v s , v t ⟩ = p s + t = 0 ( s , t ≥ 1 ) .
For 1 ≤ d < n 1\le d<n 1 ≤ d < n , a generator g I g_I g I of J n , d J_{n,d} J n , d is a coordinate minor of v 1 , … , v d v_1,\ldots,v_d v 1 , … , v d . The product of such a minor and a generator of J n , n − d J_{n,n-d} J n , n − d is multilinear and alternating in two blocks of vector arguments of total size n > dim W n n>\dim W_n n > dim W n . Apply Lemma 6.14 with the arguments independent, and then substitute these prefix vectors. Every Gram generator becomes zero. The polynomial ideal identity remains valid under this substitution in C n C_n C n , even when vectors occur in both blocks. Consequently
J n , d J n , n − d = 0. (102) J_{n,d}J_{n,n-d}=0. \tag*{(102)} J n , d J n , n − d = 0. (102)
The perfect Frobenius pairing of C n C_n C n now gives r d + r n − d ≤ n ! r_d+r_{n-d}\le n! r d + r n − d ≤ n ! . On the other hand, complementing every permutation value replaces its descent number by n − 1 − des ( w ) n-1-\operatorname{des}(w) n − 1 − des ( w ) , so R d + R n − d = n ! R_d+R_{n-d}=n! R d + R n − d = n ! . Hence
n ! = R d + R n − d ≤ r d + r n − d ≤ n ! , n!=R_d+R_{n-d}\le r_d+r_{n-d}\le n!, n ! = R d + R n − d ≤ r d + r n − d ≤ n ! ,
which forces r d = R d r_d=R_d r d = R d and r n − d = R n − d r_{n-d}=R_{n-d} r n − d = R n − d . The cases d = 0 , n d=0,n d = 0 , n are immediate. The graded surjection is therefore an isomorphism, proving (B.1); subtracting (B.5) from the ordinary coinvariant character proves (B.2). Equation (B.6) gives the actual inclusion J n , n − d ⊆ Ann ( J n , d ) J_{n,n-d}\subseteq\operatorname{Ann}(J_{n,d}) J n , n − d ⊆ Ann ( J n , d ) , and the same dimension equality makes it an equality.
The prefix map. The contractions descend to Ω n \Omega_n Ω n , since δ j ( d d R p s ) = s p s + j − 1 = 0 \delta_j(d_dR p_s)=s p_{s+j-1}=0 δ j ( d d R p s ) = s p s + j − 1 = 0 in C n C_n C n . Applying D d = δ d ⋯ δ 1 D_d=\delta_d\cdots\delta_1 D d = δ d ⋯ δ 1 to θ i 1 ⋯ θ i d \theta_{i_1}\cdots\theta_{i_d} θ i 1 ⋯ θ i d gives g I g_I g I . The map is C n C_n C n -linear, so its image is exactly J n , d J_{n,d} J n , d . Under the standard apolar pairing, δ j ∗ = Θ j \delta_j^*=\Theta_j δ j ∗ = Θ j ; reversing the composition order gives D d ∗ = Θ 1 ⋯ Θ d D_d^*=\Theta_1\cdots\Theta_d D d ∗ = Θ 1 ⋯ Θ d . □
A finite certificate for the Gram lemma
Let Z Z Z be a nondegenerate quadratic space of dimension N N N . On Λ d Z ⊗ Λ e Z \Lambda^d Z\otimes\Lambda^e Z Λ d Z ⊗ Λ e Z , use the raising and contraction operators L , c L,c L , c with the ordinary tensor signs of Lemma 6.14 . For t = d + e − N > 0 t=d+e-N>0 t = d + e − N > 0 , the operator
R d , e = ∑ j ≥ 0 ( − 1 ) j c j + 1 L j ( j + 1 ) ! t ( t + 1 ) ⋯ ( t + j ) (103) R_{d,e}=\sum_{j\ge0}\frac{(-1)^j c^{j+1}L^j}{(j+1)!\,t(t+1)\cdots(t+j)} \tag*{(103)} R d , e = j ≥ 0 ∑ ( j + 1 )! t ( t + 1 ) ⋯ ( t + j ) ( − 1 ) j c j + 1 L j (103)
satisfies L R d , e = id LR_{d,e}=\operatorname{id} L R d , e = id . Indeed, L c j + 1 L j = c j + 1 L j + 1 + ( j + 1 ) ( t + j ) c j L j Lc^{j+1}L^j=c^{j+1}L^{j+1}+(j+1)(t+j)c^jL^j L c j + 1 L j = c j + 1 L j + 1 + ( j + 1 ) ( t + j ) c j L j , so the sum telescopes; it is finite because exterior raising is nilpotent. Interpreting its terms as alternating multilinear polynomials gives an actual certificate in the ideal of cross-block inner products.
The canonical Schur resolution
We prove Theorem 9.5. Retain B = K ⊕ m B=K\oplus\mathfrak{m} B = K ⊕ m , S = S B S=S_B S = S B , and the shapes ρ i \rho_i ρ i , degrees d i d_i d i and differences ϵ i = μ i − μ i + 1 + 1 \epsilon_i=\mu_i-\mu_{i+1}+1 ϵ i = μ i − μ i + 1 + 1 from Section 9.2. The essential point is a natural contraction for square-zero multiplication. Its count filtration makes the perturbation for an arbitrary product finite.
A dimension-independent contraction
Put E = K 1 ⊕ m E = K1 \oplus\mathfrak{m} E = K 1 ⊕ m and give B 0 = K ⊕ m B_0 = K \oplus\mathfrak{m} B 0 = K ⊕ m the square-zero product. On
C i 0 = Λ i m ⊗ E ⊗ n C_i^0 = \Lambda^i \mathfrak{m} \otimes E^{\otimes n} C i 0 = Λ i m ⊗ E ⊗ n
let d 0 d_0 d 0 be its moment differential: remove a wedge input and place it in each empty coefficient position. Define c : C i 0 → C i + 1 0 c : C_i^0 \to C_{i+1}^0 c : C i 0 → C i + 1 0 by removing each occupied coefficient x x x in turn, replacing it by 1 1 1 , and sending the exterior factor ω \omega ω to x ∧ ω x \wedge\omega x ∧ ω . These formulas are natural for every linear map of m \mathfrak{m} m ; they require no bilinear form. Both operators preserve the count
k = number of occupied positions + exterior degree . k = \text{number of occupied positions} + \text{exterior degree}. k = number of occupied positions + exterior degree .
They satisfy d 0 2 = c 2 = 0 d_0^2 = c^2 = 0 d 0 2 = c 2 = 0 . With any temporary real orthonormal basis, c c c is the adjoint of d 0 d_0 d 0 for the standard tensor and exterior inner products.
Lemma C.1 (The content Laplacian). Let cont ( α ) = ∑ ( r , s ) ∈ α ( s − r ) \operatorname{cont}(\alpha) = \sum_{(r,s)\in\alpha}(s-r) cont ( α ) = ∑ ( r , s ) ∈ α ( s − r ) . On label μ ⊢ n \mu\vdash n μ ⊢ n and polynomial Schur type λ ⊢ k \lambda\vdash k λ ⊢ k in C i 0 C_i^0 C i 0 , the operator L = d 0 c + c d 0 L = d_0c + cd_0 L = d 0 c + c d 0 is the scalar
cont ( μ ) − cont ( λ ) + n − ( n − k + 1 2 ) (104) \operatorname{cont}(\mu) - \operatorname{cont}(\lambda) + n - \binom{n-k+1}{2} \tag*{(104)} cont ( μ ) − cont ( λ ) + n − ( 2 n − k + 1 ) (104) Here the binomial denotes its polynomial value, also for negative arguments. There is a natural contraction onto H = ker L \mathcal{H} = \ker L H = ker L , with inclusion j j j , projection p p p , and homotopy h h h satisfying
d 0 h + h d 0 = id − j p , h 2 = 0 , h j = p h = 0. d_0h + hd_0 = \operatorname{id} - jp,\qquad h^2 = 0,\qquad hj = ph = 0. d 0 h + h d 0 = id − j p , h 2 = 0 , hj = p h = 0. On label μ \mu μ , the harmonic space in homological degree i i i is H i ≅ S ρ i m \mathcal{H}_i \cong S_{\rho_i}\mathfrak{m} H i ≅ S ρ i m , in count d i d_i d i .
Proof. First let q = dim m < ∞ q = \dim\mathfrak{m} < \infty q = dim m < ∞ . On the coefficient module S μ ( E ) S_\mu(E) S μ ( E ) , write E a b E_{ab} E ab for the matrix-unit action, with index 0 0 0 denoting the unit. Exterior creation and contraction give
d 0 = ∑ a = 1 q E a 0 ι a , c = ∑ a = 1 q E 0 a ε a , L = ∑ a E a 0 E 0 a + i E 00 − ∑ a , b E a b ε b ι a . d_0 = \sum_{a=1}^{q} E_{a0}\iota_a,\qquad c = \sum_{a=1}^{q} E_{0a}\varepsilon_a,\qquad L = \sum_a E_{a0}E_{0a} + iE_{00} - \sum_{a,b} E_{ab}\varepsilon_b\iota_a. d 0 = a = 1 ∑ q E a 0 ι a , c = a = 1 ∑ q E 0 a ε a , L = a ∑ E a 0 E 0 a + i E 00 − a , b ∑ E ab ε b ι a .
Let Ω q = ∑ a , b E a b E b a \Omega_q = \sum_{a,b} E_{ab}E_{ba} Ω q = ∑ a , b E ab E ba be the quadratic Casimir. Its eigenvalue on S λ K q S_\lambda K^q S λ K q is 2 cont ( λ ) + q ∣ λ ∣ 2\operatorname{cont}(\lambda) + q|\lambda| 2 cont ( λ ) + q ∣ λ ∣ . The cross-Casimir identity rewrites the preceding formula as
2 L = Ω q + 1 μ − Ω q t o t a l + Ω q Λ i − E 00 2 + ( 2 i − q − 1 ) E 00 + n . 2L = \Omega_{q+1}^{\mu} - \Omega_q^{\mathrm{total}} + \Omega_q^{\Lambda^i} - E_{00}^2 + (2i-q-1)E_{00} + n. 2 L = Ω q + 1 μ − Ω q total + Ω q Λ i − E 00 2 + ( 2 i − q − 1 ) E 00 + n .
Now E 00 = n − k + i E_{00} = n-k+i E 00 = n − k + i and Ω q Λ i = i ( q + 1 − i ) \Omega_q^{\Lambda^i} = i(q+1-i) Ω q Λ i = i ( q + 1 − i ) . Substitution cancels q q q and i i i and gives (C.1).
Over the real numbers the adjoint identity implies ker L = ker d 0 ∩ ker c \ker L = \ker d_0 \cap\ker c ker L = ker d 0 ∩ ker c . Let L + L^+ L + invert every nonzero eigenvalue and be zero on the kernel. Then h = c L + h = cL^+ h = c L + and the spectral projection j p jp j p give the asserted contraction. Its formulas are rational: the eigenvalues in (C.1) are integers and the Schur projections are natural. They therefore hold over every characteristic-zero field and commute with arbitrary linear maps, including maps that lower dimension.
To identify the harmonic spaces, apply the pure-resolution theorem [32 ] , Theorem 2.2 to the square-zero presentation, with α = τ \alpha= \tau α = τ and β = τ + \beta= \tau^+ β = τ + . Its successive generator shapes are precisely ρ i \rho_i ρ i , and its degree shifts, including the initial shift ∣ τ ∣ |\tau| ∣ τ ∣ , are d i d_i d i . Hence the square-zero Koszul homology is S ρ i m S_{\rho_i}\mathfrak{m} S ρ i m in count d i d_i d i . This harmonic copy occurs once: its coefficient branching shape deletes row i + 1 i+1 i + 1 of μ \mu μ , and its exterior Pieri strip adds one box to each of the first i i i rows. The case i = 0 i = 0 i = 0 is minimum occupancy.
For infinite-dimensional m \mathfrak{m} m , each fixed-count construction passes through finite-dimensional vector subspaces. The scalar formula is independent of their dimension, so the contractions are compatible under inclusion and define the same natural operators on their union. □ \square □
The diagram convention in [32 ] is transposed, while its highest-weight tuples agree with those used here. This argument uses its full pure-resolution theorem; the earlier presentation used only the Pieri composition lemma.
Multiplication as a finite perturbation
Proof of Theorem 9.5 . We first transfer the collision differential to compute moment homology. We then adjoin formal trace variables and transfer once more to obtain the free resolution; setting those variables to zero recovers the first complex. Both transfers terminate because they lower coefficient count.
For the original algebra, write the moment differential as d B = d 0 + δ d_B=d_0+\delta d B = d 0 + δ . The collision operator δ \delta δ removes a wedge input and multiplies it into each occupied position. It lowers count by one, while h h h preserves count. The perturbation lemma, with our convention d 0 h + h d 0 = i d − j p d_0h+hd_0=\mathrm{id}-jp d 0 h + h d 0 = id − j p , gives the transferred differential
p δ ( 1 + h δ ) − 1 j = ∑ r ≥ 0 ( − 1 ) r p δ ( h δ ) r j . p\delta(1+h\delta)^{-1}j=\sum_{r\geq0}(-1)^r p\delta(h\delta)^rj. p δ ( 1 + h δ ) − 1 j = r ≥ 0 ∑ ( − 1 ) r p δ ( h δ ) r j .
Every series is finite on each vector. This is the standard formula of [13 §2, especially 2.4] , with its homotopy replaced by − h -h − h . A term from H i \mathcal{H}_i H i to H i − 1 \mathcal{H}_{i-1} H i − 1 must lower count from d i d_i d i to d i − 1 d_{i-1} d i − 1 , so only one term survives:
∂ i = ( − 1 ) e i − 1 p i − 1 δ ( h δ ) e i − 1 j i . (105) \partial_i=(-1)^{e_i-1}p_{i-1}\delta(h\delta)^{e_i-1}j_i. \tag*{(105)} ∂ i = ( − 1 ) e i − 1 p i − 1 δ ( h δ ) e i − 1 j i . (105)
Thus the small complex computes all moment homology, with no choices of higher splittings.
To obtain an actual free resolution, use C f r e e = S ⊗ Λ ∙ m ⊗ S μ ( E ) C_{\mathrm{free}}=S\otimes\Lambda^\bullet\mathfrak{m}\otimes S_\mu(E) C free = S ⊗ Λ ∙ m ⊗ S μ ( E ) . Let d T d_T d T remove a wedge input b b b and multiply the S S S factor by T b T_b T b . Subtracting d T d_T d T enforces the action T b = p n ( b ) T_b=p_n(b) T b = p n ( b ) . The differential d B − d T d_B-d_T d B − d T resolves S μ ( B ) S_\mu(B) S μ ( B ) under f ( T ) ⊗ v ↦ f ( p n ) v f(T)\otimes v\mapsto f(p_n)v f ( T ) ⊗ v ↦ f ( p n ) v . Indeed, these are polynomial-variable elimination operators p n ( b ) − T b p_n(b)-T_b p n ( b ) − T b ; eliminating any finite set is exact by successive substitution, and their directed union handles infinitely many variables.
Extend the preceding contraction S S S -linearly, starting with d 0 d_0 d 0 , and perturb it by A = δ − d T A=\delta-d_T A = δ − d T . Both terms lower coefficient count by one; the increase of formal polynomial degree under d T d_T d T is irrelevant to this count. Therefore the transferred free differential is the finite formula
D i = ( − 1 ) e i − 1 p i − 1 A ( h A ) e i − 1 j i . (106) D_i=(-1)^{e_i-1}p_{i-1}A(hA)^{e_i-1}j_i. \tag*{(106)} D i = ( − 1 ) e i − 1 p i − 1 A ( h A ) e i − 1 j i . (106)
It gives a chain homotopy equivalent free complex with the stated terms, and hence a resolution. Setting T = 0 T=0 T = 0 gives (C.2).
For the resolution filtration, give a formal trace count one and the generator of F i F_i F i count d i d_i d i . A term with no collisions has formal degree e i e_i e i and preserves this total count; every collision lowers it. At δ = 0 \delta=0 δ = 0 the complex is a minimal equivariant pure resolution, so its leading maps are the classical nonzero Pieri maps.
The corrected harmonic inclusion is ( 1 + h A ) − 1 j (1+hA)^{-1}j ( 1 + h A ) − 1 j . On degree zero, A j 0 = 0 Aj_0=0 A j 0 = 0 , so its augmentation is exactly the minimum-occupancy map. Choose the Schur identification line for H 1 \mathcal{H}_1 H 1 so that the leading part of D 1 D_1 D 1 equals the previously fixed Pieri inclusion. Every lower-count term lies below d 1 d_1 d 1 , where the omitted Schur summands are absent. The unique lift of that leading relation is therefore the graph lift (9.3), so D 1 = R μ , B D_1=R_{\mu,B} D 1 = R μ , B literally with this normalization. The earlier quotient map used β μ , B = − ϵ v T = 0 R μ , B \beta_{\mu,B}=-\epsilon v_T=0\,R_{\mu,B} β μ , B = − ϵ v T = 0 R μ , B ; accordingly ∂ 1 = − β μ , B \partial_1=-\beta_{\mu,B} ∂ 1 = − β μ , B . The kernels and cokernels are unchanged. For example, before this line normalize the raw adjoint convention at n = 1 n=1 n = 1 , m = K a m=K_a m = K a , a 2 = c a a^2=ca a 2 = c a gives D 1 = c T a − T a 2 D_1=cT_a-T_a^2 D 1 = c T a − T a 2 .
All operators are built from natural Schur projections, unit and augmentation, and multiplication. This proves the claimed functoriality and compatibility with gradings. It gives algebra-automorphism equivariance for general B B B ; full G L ( m ) \mathrm{GL}(\mathfrak{m}) GL ( m ) equivariance belongs to the square-zero specialization.
Finer row-sensitive character ranges
The uniform character theorem needs only the Schur relation, quadratic reduction, bivariate and hook exactness, and one cubic witness. This appendix retains the separate row-sensitive trace bounds and their closed boundary, which can give better ranges for individual rows.
Row-sensitive support of moment Koszul homology
The first row of the row partition gives a second measure of how early a tail relation can occur. A relation on t + 1 t+1 t + 1 positive inputs has many linear inputs until its degree becomes large. Their symmetric product forces a long first row. Comparing this trace relation with the free moment complex gives the following joint restriction on row and label.
Theorem D.1 (A row-and-label strip for positive moment homology). Suppose S λ V ⊗ S μ S_{\lambda}V \otimes S^{\mu} S λ V ⊗ S μ occurs in H j ( K n , k , D ) D H_j(K_{n,k,D})_D H j ( K n , k , D ) D , where j ≥ 1 j \ge1 j ≥ 1 and λ ⊢ D \lambda\vdash D λ ⊢ D . Write a = λ 1 a=\lambda_1 a = λ 1 . Then
n − D + 2 j ≤ μ 1 ≤ ⌊ D + a 2 ⌋ − j . (107) n-D+2j \le\mu_1 \le\left\lfloor\frac{D+a}{2} \right\rfloor-j. \tag*{(107)} n − D + 2 j ≤ μ 1 ≤ ⌊ 2 D + a ⌋ − j . (107) When 3 ≤ D ≤ n 3 \le D \le n 3 ≤ D ≤ n , the sharper strip is
n − D + 3 j ≤ μ 1 ≤ min { ⌊ D + a 2 ⌋ − j , D − 2 } . (108) n-D+3j \le\mu_1 \le\min\left\{\left\lfloor\frac{D+a}{2} \right\rfloor-j,\ D-2\right\}. \tag*{(108)} n − D + 3 j ≤ μ 1 ≤ min { ⌊ 2 D + a ⌋ − j , D − 2 } . (108) More generally, if all moments of orders at most m ≥ 1 m \ge1 m ≥ 1 form a regular sequence in P P P , then
n − D + ( m + 1 ) j ≤ μ 1 ≤ ⌊ D + a − ( m + 1 ) ( j − 1 ) − 2 2 ⌋ . (109) n-D+(m+1)j \le\mu_1 \le\left\lfloor\frac{D+a-(m+1)(j-1)-2}{2} \right\rfloor. \tag*{(109)} n − D + ( m + 1 ) j ≤ μ 1 ≤ ⌊ 2 D + a − ( m + 1 ) ( j − 1 ) − 2 ⌋ . (109)
Proof. For the general lower bound, remove the linear moments. Every remaining exterior generator has weight at least two, and a polynomial of degree at most D − 2 j D-2j D − 2 j has first label row at least n − D + 2 j n-D+2j n − D + 2 j . In the range D ≤ n D \le n D ≤ n , quadratic exactness raises the minimum remaining weight to three, as in Lemma 10.4.
We prove the new upper bound. For 1 ≤ t ≤ n 1 \le t \le n 1 ≤ t ≤ n , let S t S_t S t act on the last t t t labels, put r = n − t r=n-t r = n − t , and write
A r = Sym ( V ⊗ K r ) , A = Sym V , F = Sym ( A + ) , T = ( A ⊗ t ) S t . A_r=\operatorname{Sym}(V\otimes K^r),\qquad A=\operatorname{Sym}V,\qquad F=\operatorname{Sym}(A_+),\qquad T=(A^{\otimes t})^{S_t}. A r = Sym ( V ⊗ K r ) , A = Sym V , F = Sym ( A + ) , T = ( A ⊗ t ) S t .
The map F → T F\to T F → T sends the formal trace [ b ] [b] [ b ] to ∑ i = 1 t b ( X i ) \sum_{i=1}^{t}b(X_i) ∑ i = 1 t b ( X i ) . By the trace presentation used in Theorem 2.1, its kernel I t I_t I t is generated by the equivariant image of Sym t + 1 ( A + ) \operatorname{Sym}^{t+1}(A_+) Sym t + 1 ( A + ) . Explicitly, the relation on b 1 , … , b t + 1 b_1,\ldots,b_{t+1} b 1 , … , b t + 1 is
∑ π ∈ Π t + 1 ( − 1 ) t + 1 − ∣ π ∣ ∏ B ∈ π ( ∣ B ∣ − 1 ) ! ∏ B ∈ π [ ∏ i ∈ B b i ] . \sum_{\pi\in\Pi_{t+1}}(-1)^{t+1-|\pi|}\prod_{B\in\pi}(|B|-1)!\prod_{B\in\pi}\left[\prod_{i\in B}b_i\right]. π ∈ Π t + 1 ∑ ( − 1 ) t + 1 − ∣ π ∣ B ∈ π ∏ ( ∣ B ∣ − 1 )! B ∈ π ∏ [ i ∈ B ∏ b i ] .
The internal degree is ∑ i deg b i \sum_i\deg b_i ∑ i deg b i ; the same formula is extended multilinearly to arbitrary inputs in A + A_+ A + .
A degree-g g g generator has at least 2 t + 2 − g 2t+2-g 2 t + 2 − g linear inputs. Splitting Sym t + 1 ( A + ) \operatorname{Sym}^{t+1}(A_+) Sym t + 1 ( A + ) by the number ℓ \ell ℓ of linear inputs therefore expresses its relevant summands as Sym ℓ V ⊗ W \operatorname{Sym}^{\ell}V\otimes W Sym ℓ V ⊗ W , with ℓ ≥ 2 t + 2 − g \ell\ge2t+2-g ℓ ≥ 2 t + 2 − g and W W W polynomial. The Littlewood–Richardson rule shows that every constituent has first row at least 2 t + 2 − g 2t+2-g 2 t + 2 − g . Multiplication by any polynomial representation, or tensoring with one, cannot decrease that first row.
First eliminate the global linear moments in this comparison. They form a regular sequence both in A r ⊗ F A_r\otimes F A r ⊗ F and in A r ⊗ T = P S t A_r\otimes T=P^{S_t} A r ⊗ T = P S t . In the first ring their formal trace variables are independent. In the second, splitting the label representation into the invariant diagonal and the centered subspace gives a polynomial factor on the global linear moments; taking S t S_t S t -invariants preserves this factorization. Consequently the exact sequence with kernel A r ⊗ I t A_r\otimes I_t A r ⊗ I t remains exact after quotienting by those moments. Its new kernel I t ′ I'_t I t ′ is the image of that ideal, generated by the same equivariant trace-relation sources. Their first-row bounds are therefore unchanged.
The source after elimination is the polynomial algebra
A r ⊗ Sym ( A ≥ 2 ) . A_r\otimes\operatorname{Sym}(A_{\ge2}). A r ⊗ Sym ( A ≥ 2 ) .
In total degrees at most D D D , form its Koszul complex L ‾ \overline{\mathcal{L}} L on the remaining global moments [ X α ] + ∑ i ≤ r X i α [X^\alpha]+\sum_{i\le r}X_i^\alpha [ X α ] + ∑ i ≤ r X i α , 2 ≤ ∣ α ∣ ≤ D 2\le|\alpha|\le D 2 ≤ ∣ α ∣ ≤ D . A polynomial change of the independent formal trace variables makes this complex exact in positive degrees. It subjects onto the corresponding reduced moment complex of P S t P^{S_t} P S t , whose homology is that of K n , k , D S t K^{S_t}_{n,k,D} K n , k , D S t because the removed linear sequence is regular. Its kernel complex has terms
N i = I ˉ t ⊗ ⋀ i ( ⨁ d = 2 D Sym d V ) . \mathcal{N}_i=\bar{I}_t\otimes\bigwedge^i\left(\bigoplus_{d=2}^{D}\operatorname{Sym}^d V\right). N i = I ˉ t ⊗ ⋀ i ( d = 2 ⨁ D Sym d V ) .
All maps are equivariant. Using the full free trace algebra and then taking the stated degrees gives the same terms: variables and relations of weight greater than D D D cannot contribute.
Positive homology of the finite-label complex is detected one homological degree earlier in this trace-relation kernel. If a row λ ⊢ D \lambda\vdash D λ ⊢ D occurs in N j − 1 \mathcal{N}_{j-1} N j − 1 , it comes from a trace-relation generator of some degree g g g , polynomial factors, and j − 1 j-1 j − 1 exterior generators of weight at least two. Hence g ≤ D − 2 j + 2 g\le D-2j+2 g ≤ D − 2 j + 2 . By the first-row bound and Littlewood–Richardson containment,
a ≥ 2 t + 2 − g ≥ 2 t + 2 j − D . a\ge2t+2-g\ge2t+2j-D. a ≥ 2 t + 2 − g ≥ 2 t + 2 j − D .
The long exact sequence of 0 → N → L → K n , k , D S t → 0 0\to\mathcal{N}\to\mathcal{L}\to K^{S_t}_{n,k,D}\to0 0 → N → L → K n , k , D S t → 0 identifies positive homology in degree j ≥ 2 j\ge2 j ≥ 2 with H j − 1 ( N ) H_{j-1}(\mathcal{N}) H j − 1 ( N ) , and injects degree-one homology into H 0 ( N ) H_0(\mathcal{N}) H 0 ( N ) . Thus its row-λ \lambda λ part vanishes whenever D + a < 2 t + 2 j D+a<2t+2j D + a < 2 t + 2 j .
Now take t = μ 1 t=\mu_1 t = μ 1 . Branching gives a nonzero S t S_t S t -fixed vector in S μ S^\mu S μ , and exactness of invariants identifies H j ( K n , k , D ) S t H_j(K_{n,k,D})^{S_t} H j ( K n , k , D ) S t with the homology just considered. Therefore D + a ≥ 2 μ 1 + 2 j D+a\ge2\mu_1+2j D + a ≥ 2 μ 1 + 2 j , which is the desired upper bound. This also permits r = 0 r=0 r = 0 : the selected-label polynomial factor is then K K K .
Combining this estimate with Lemma 10.4 gives (D.2).
If the moments through order m m m are regular in P P P , their Koszul complex stays exact after taking S t S_t S t -invariants, so they are regular in P S t P^{S_t} P S t as well. The same comparison can therefore eliminate all their independent formal traces first. The trace-relation sources retain their row bound, while every remaining exterior generator has weight at least m + 1 m+1 m + 1 . Thus g ≤ D − ( m + 1 ) ( j − 1 ) g\le D-(m+1)(j-1) g ≤ D − ( m + 1 ) ( j − 1 ) , giving the upper bound in (D.3); removal of the same regular sequence gives its lower bound by monomial support. □
The row-λ \lambda λ Euler character is the S n S_n S n -representation
∑ j ≥ 0 ( − 1 ) j Hom GL ( V ) ( S λ V , ( P ⊗ ⋀ j U D ) D ) . \sum_{j\ge0}(-1)^j\operatorname{Hom}_{\operatorname{GL}(V)}\left(S^\lambda V,(P\otimes\bigwedge^j U_D)_D\right). j ≥ 0 ∑ ( − 1 ) j Hom GL ( V ) ( S λ V , ( P ⊗ ⋀ j U D ) D ) .
Its evaluation at σ ∈ S n \sigma\in S_n σ ∈ S n is obtained by projecting the total-degree-D D D Koszul product in (74) onto the Schur character s λ s_\lambda s λ . Thus exactness can be tested for one row partition while other rows in the same multidegree retain homology.
Detection of the first row-sensitive trace relations
At equality in the trace-relation bound, all positive inputs are linear or quadratic. This leaves a multiplicity-one source whose highest vector can be written with bordered minors. A factorial evaluation of that vector proves that the first possible relation survives negative-moment substitution. It therefore contributes no positive moment homology.
Lemma D.2 (The boundary trace is detected by selected labels.) Let λ ⊢ D \lambda\vdash D λ ⊢ D , write a = λ 1 a=\lambda_1 a = λ 1 , and suppose
D + a = 2 t + 2. (110) D+a=2t+2. \tag*{(110)} D + a = 2 t + 2. (110) Set q = t + 1 q=t+1 q = t + 1 , r = n − t r=n-t r = n − t , and use the free tail-trace algebra F = Sym ( A + ) F=\operatorname{Sym}(A_+) F = Sym ( A + ) , its kernel I t I_t I t , and the selected-label polynomial algebra A r A_r A r from the proof of Theorem D.1 . The degree-D D D , row-λ \lambda λ part of A r ⊗ I t A_r\otimes I_t A r ⊗ I t has multiplicity at most one. It is zero unless
λ = ( a , 2 κ ) , ∣ κ ∣ = D − a 2 , 2 κ 1 ≤ a . (111) \lambda=(a,2\kappa),\qquad|\kappa|=\frac{D-a}{2},\qquad2\kappa_1\le a. \tag*{(111)} λ = ( a , 2 κ ) , ∣ κ ∣ = 2 D − a , 2 κ 1 ≤ a . (111)
If (D.5) holds and r ≥ ℓ ( λ ) r \ge\ell(\lambda) r ≥ ℓ ( λ ) , the negative-moment specialization
ev − : A r ⊗ F ⟶ A r , [ b ] ⟼ − ∑ i = 1 r b ( X i ) , \operatorname{ev}_{-}: A_r \otimes F \longrightarrow A_r,\qquad[b] \longmapsto-\sum_{i=1}^{r} b(X_i), ev − : A r ⊗ F ⟶ A r , [ b ] ⟼ − i = 1 ∑ r b ( X i ) ,
is nonzero, hence injective, on that single irreducible summand. Consequently the row-λ \lambda λ part of H j ( K n , k , D ) D S t H_j(\mathcal{K}_{n,k,D})^{S_t}_D H j ( K n , k , D ) D S t vanishes for every j ≥ 1 j \ge1 j ≥ 1 if the even-tail condition fails, or if r ≥ ℓ ( λ ) r \ge\ell(\lambda) r ≥ ℓ ( λ ) . A row of length greater than dim V \dim V dim V is absent.
Proof. The trace-relation generators have source Sym q ( A + ) \operatorname{Sym}^q(A_+) Sym q ( A + ) . A degree-g g g generator has first row at least 2 q − g 2q-g 2 q − g . Under (D.4), a row-λ \lambda λ constituent of total degree D D D therefore requires g = D g=D g = D , with no positive-degree multiplier. Moreover its source must have exactly a a a linear inputs and m = ( D − a ) / 2 m=(D-a)/2 m = ( D − a ) /2 quadratic inputs. A larger number of linear inputs would force a larger first row, and any input of degree at least three would force more linear inputs. The entire possible source is consequently
Sym a V ⊗ Sym m ( Sym 2 V ) . (112) \operatorname{Sym}^a V \otimes\operatorname{Sym}^m(\operatorname{Sym}^2 V). \tag*{(112)} Sym a V ⊗ Sym m ( Sym 2 V ) . (112)
The symmetric-matrix Cauchy identity gives
Sym m ( Sym 2 V ) = ⨁ κ ⊢ m S 2 κ V . \operatorname{Sym}^m(\operatorname{Sym}^2 V)=\bigoplus_{\kappa\vdash m} \mathbb{S}_{2\kappa}V. Sym m ( Sym 2 V ) = κ ⊢ m ⨁ S 2 κ V .
Pieri’s rule now says that a constituent of (D.6) with first row exactly a a a must be ( a , 2 κ ) (a,2\kappa) ( a , 2 κ ) , since its horizontal strip gives 2 κ i ≥ λ i + 1 2\kappa_i \ge\lambda_{i+1} 2 κ i ≥ λ i + 1 , while both sides sum to D − a D-a D − a . This proves the asserted multiplicity bound and shape restriction.
It remains to show that this possible summand survives the map from trace-relation sources to selected polynomials. Multiplicity one reduces injectivity to a single nonzero image. We will choose a highest vector whose negative-moment image can be evaluated without cancellation. Realize (D.6) using vector generators u 1 , u 2 , … u_1,u_2,\ldots u 1 , u 2 , … of weight one and symmetric-matrix generators M i j = M j i M_{ij}=M_{ji} M ij = M j i of weight two. For j ≥ 1 j \ge1 j ≥ 1 let
p j ( u , M ) = det ( 0 u 1 ⋯ u j + 1 u 1 M 11 ⋯ M 1 , j + 1 ⋮ ⋮ ⋱ ⋮ u j + 1 M j + 1 , 1 ⋯ M j + 1 , j + 1 ) . (113) p_j(u,M)=\det\begin{pmatrix}
0 & u_1 & \cdots& u_{j+1}\\
u_1 & M_{11} & \cdots& M_{1,j+1}\\
\vdots& \vdots& \ddots& \vdots\\
u_{j+1} & M_{j+1,1} & \cdots& M_{j+1,j+1}
\end{pmatrix}. \tag*{(113)} p j ( u , M ) = det 0 u 1 ⋮ u j + 1 u 1 M 11 ⋮ M j + 1 , 1 ⋯ ⋯ ⋱ ⋯ u j + 1 M 1 , j + 1 ⋮ M j + 1 , j + 1 . (113)
It has u u u -degree two, M M M -degree j j j , and highest weight ( 2 j + 1 ) (2j+1) ( 2 j + 1 ) : each raising operator acts by an elementary row operation together with the corresponding column operation on a leading principal block. Set κ ℓ ( κ ) + 1 = 0 \kappa_{\ell(\kappa)+1}=0 κ ℓ ( κ ) + 1 = 0 . Then
P κ ( u , M ) = u 1 a − 2 κ 1 ∏ j ≥ 1 p j ( u , M ) κ j − κ j + 1 (114) P_\kappa(u,M)=u_1^{a-2\kappa_1}\prod_{j\ge1}p_j(u,M)^{\kappa_j-\kappa_{j+1}} \tag*{(114)} P κ ( u , M ) = u 1 a − 2 κ 1 j ≥ 1 ∏ p j ( u , M ) κ j − κ j + 1 (114)
is a highest-weight vector of weight ( a , 2 κ ) (a,2\kappa) ( a , 2 κ ) , with u u u -degree a a a and M M M -degree m m m . For κ = ∅ \kappa=\varnothing κ = ∅ the formula means P ∅ = u 1 a P_\varnothing=u_1^a P ∅ = u 1 a .
Introduce auxiliary variables ξ 1 , … , ξ r \xi_1,\ldots,\xi_r ξ 1 , … , ξ r and the linear functional
E ( ξ 1 α 1 ⋯ ξ r α r ) = ∏ i = 1 r α i ! . \mathcal{E}(\xi_1^{\alpha_1}\cdots\xi_r^{\alpha_r})=\prod_{i=1}^{r}\alpha_i!. E ( ξ 1 α 1 ⋯ ξ r α r ) = i = 1 ∏ r α i ! .
For the multilinear fundamental trace relation Ψ q \Psi_q Ψ q , its negative-moment specialization is exactly
ev − Ψ q ( b 1 , … , b q ) = ( − 1 ) q E [ ∏ s = 1 q ( ∑ i = 1 r ξ i b s ( X i ) ) ] . (115) \operatorname{ev}_{-}\Psi_q(b_1,\ldots,b_q)=(-1)^q\mathcal{E}\left[\prod_{s=1}^{q}\left(\sum_{i=1}^{r}\xi_i b_s(X_i)\right)\right]. \tag*{(115)} ev − Ψ q ( b 1 , … , b q ) = ( − 1 ) q E [ s = 1 ∏ q ( i = 1 ∑ r ξ i b s ( X i ) ) ] . (115)
Indeed, all set-partition terms in Ψ q \Psi_q Ψ q acquire the same sign ( − 1 ) q (-1)^q ( − 1 ) q . For a fixed assignment of the inputs to the selected labels, the sum over set partitions refining each fiber is the product of the fiber factorials: its blocks are the cycles of a permutation. These are precisely the coefficients given by E \mathcal{E} E . Thus the normalization in (D.9) is fixed, without any choice of polarization convention. Write Ψ ^ q : Sym q ( A + ) → I t \widehat{\Psi}_q:\operatorname{Sym}^q(A_+)\to I_t Ψ q : Sym q ( A + ) → I t for the linear map induced by the multilinear trace relation. The actual relation used here is Ψ ^ q ( P κ ) \widehat{\Psi}_q(P_\kappa) Ψ q ( P κ ) . Applying (D.9) to it amounts to substituting
u = ∑ i ξ i X i , M = ∑ i ξ i X i X i T , u=\sum_i \xi_i X_i,\qquad M=\sum_i \xi_i X_iX_i^{\mathsf T}, u = i ∑ ξ i X i , M = i ∑ ξ i X i X i T ,
then applying ( − 1 ) q E (-1)^q\mathcal{E} ( − 1 ) q E . Since r ≥ ℓ ( λ ) r\ge\ell(\lambda) r ≥ ℓ ( λ ) , specialize X i X_i X i to the i i i th coordinate vector for 1 ≤ i ≤ ℓ ( λ ) 1\le i\le\ell(\lambda) 1 ≤ i ≤ ℓ ( λ ) and set all remaining X i X_i X i equal to zero. Now u i = ξ i u_i=\xi_i u i = ξ i and M M M is diagonal, so
p j ( u , M ) = − ( ∏ i = 1 j + 1 ξ i ) ( ∑ i = 1 j + 1 ξ i ) . p_j(u,M)=-\left(\prod_{i=1}^{j+1}\xi_i\right)\left(\sum_{i=1}^{j+1}\xi_i\right). p j ( u , M ) = − ( i = 1 ∏ j + 1 ξ i ) ( i = 1 ∑ j + 1 ξ i ) .
The resulting P κ P_\kappa P κ is a nonzero polynomial all of whose nonzero coefficients have sign ( − 1 ) κ 1 (-1)^{\kappa_1} ( − 1 ) κ 1 . Its image under E \mathcal{E} E is therefore a nonzero integer. This proves nonvanishing over every field of characteristic zero. The case κ = ∅ \kappa=\varnothing κ = ∅ gives the same conclusion directly from E ( ξ 1 a ) = a ! \mathcal{E}(\xi_1^a)=a! E ( ξ 1 a ) = a ! .
Finally compare the free Koszul complex on [ X α ] + ∑ i ≤ r X i α [X^\alpha]+\sum_{i\le r}X_i^\alpha [ X α ] + ∑ i ≤ r X i α in A r ⊗ F A_r\otimes F A r ⊗ F with the moment complex of P S t P^{S_t} P S t . The free complex is exact in positive degrees; its zeroth homology in the degrees under consideration is A r A_r A r . Its kernel complex has terms
( A r ⊗ I t ) ⊗ ⋀ j U D . (A_r\otimes I_t)\otimes\bigwedge^j U_D. ( A r ⊗ I t ) ⊗ ⋀ j U D .
Under (D.4), none of its terms with j ≥ 1 j\ge1 j ≥ 1 can contain row λ \lambda λ in total degree D D D : an exterior factor would require a trace generator of degree strictly less than D D D , whose first row is greater than a a a . All homology in homological degrees at least two is therefore zero in this row. The remaining part of the long exact sequence is
0 ⟶ H 1 ( K n , k , D ) D , λ S t ⟶ ( A r ⊗ I t ) D , λ → ev − ( A r ) D , λ . 0\longrightarrow H_1(\mathcal{K}_{n,k,D})^{S_t}_{D,\lambda}\longrightarrow(A_r\otimes I_t)_{D,\lambda}\xrightarrow{\operatorname{ev}_{-}}(A_r)_{D,\lambda}. 0 ⟶ H 1 ( K n , k , D ) D , λ S t ⟶ ( A r ⊗ I t ) D , λ ev − ( A r ) D , λ .
The source is zero in the excluded shapes, and in the remaining case it is a single irreducible on which the displayed map is nonzero. This proves the homology assertion. □
Corollary D.3 (The closed upper boundary and row characters). Let 3 ≤ D ≤ n 3\le D\le n 3 ≤ D ≤ n , let λ ⊢ D \lambda\vdash D λ ⊢ D , and put a = λ 1 a=\lambda_1 a = λ 1 . A joint row-and-label constituent of positive moment homology satisfies
n − D + 3 j ≤ μ 1 ≤ min { ⌊ D + a 2 ⌋ − j , ⌊ D + a − 3 2 ⌋ } . (116) n-D+3j\le\mu_1\le\min\left\{\left\lfloor\frac{D+a}{2}\right\rfloor-j,\left\lfloor\frac{D+a-3}{2}\right\rfloor\right\}. \tag*{(116)} n − D + 3 j ≤ μ 1 ≤ min { ⌊ 2 D + a ⌋ − j , ⌊ 2 D + a − 3 ⌋ } . (116) In particular the row-λ \lambda λ Koszul Euler character is the actual row-λ \lambda λ character of ( R n , k ) D (R_{n,k})_D ( R n , k ) D whenever
n ≥ max { D , D − 2 + ⌊ D + a − 3 2 ⌋ } . (117) n\ge\max\left\{D,D-2+\left\lfloor\frac{D+a-3}{2}\right\rfloor\right\}. \tag*{(117)} n ≥ max { D , D − 2 + ⌊ 2 D + a − 3 ⌋ } . (117) For a single joint row-and-label component, equality of the Euler and actual characters already holds, for 3 ≤ D ≤ n 3\le D\le n 3 ≤ D ≤ n , if
μ 1 ≥ ⌊ D + a − 1 2 ⌋ . (118) \mu_1\ge\left\lfloor\frac{D+a-1}{2}\right\rfloor. \tag*{(118)} μ 1 ≥ ⌊ 2 D + a − 1 ⌋ . (118) Proof. The j j j -dependent bounds are Theorem D.1 . If D + a D+a D + a is odd, they already give the additional upper bound in (D.10). If D + a D+a D + a is even, the only additional case to exclude is j = 1 j=1 j = 1 and t = μ 1 = ( D + a − 2 ) / 2 t=\mu_1=(D+a-2)/2 t = μ 1 = ( D + a − 2 ) /2 . Branching gives an S t S_t S t -fixed vector. At that value, Lemma D.2 excludes the row unless λ = ( a , 2 κ ) \lambda=(a,2\kappa) λ = ( a , 2 κ ) . In the latter case, with m = ∣ κ ∣ = ( D − a ) / 2 m=|\kappa|=(D-a)/2 m = ∣ κ ∣ = ( D − a ) /2 , the hypothesis n ≥ D n\ge D n ≥ D gives
r = n − t ≥ D − t = m + 1 ≥ ℓ ( κ ) + 1 = ℓ ( λ ) . r=n-t\ge D-t=m+1\ge\ell(\kappa)+1=\ell(\lambda). r = n − t ≥ D − t = m + 1 ≥ ℓ ( κ ) + 1 = ℓ ( λ ) . The same lemma therefore excludes the remaining case as well. This proves the strip. Write C = ⌊ ( D + a − 3 ) / 2 ⌋ C=\lfloor(D+a-3)/2\rfloor C = ⌊( D + a − 3 ) /2 ⌋ . Under (D.11), positive homology would require μ 1 ≥ n − D + 3 ≥ C + 1 \mu_1\ge n-D+3\ge C+1 μ 1 ≥ n − D + 3 ≥ C + 1 and μ 1 ≤ C \mu_1\le C μ 1 ≤ C , a contradiction. Thus zeroth equals Koszul Euler character in that row. The final assertion follows directly from the upper bound and C + 1 = ⌊ ( D + a − 1 ) / 2 ⌋ C+1=\lfloor(D+a-1)/2\rfloor C + 1 = ⌊( D + a − 1 ) /2 ⌋ . □
For the widest row a = D a=D a = D , the strip gives n ≥ max { D , 2 D − 4 } n \ge\max\{D,2D-4\} n ≥ max { D , 2 D − 4 } ; for bounded a a a its threshold is 3 2 D + O a ( 1 ) \frac{3}{2}D+O_a(1) 2 3 D + O a ( 1 ) , and balanced three-row partitions have threshold 5 3 D + O ( 1 ) \frac{5}{3}D+O(1) 3 5 D + O ( 1 ) . These finer rowwise conclusions are independent of the shorter proof of the uniform theorem.
The odd moment differential algebra
The two-column reduction has a simple differential-algebra model. It makes the ring duality explicit and identifies the minimal syzygy-generating layer of Proposition 10.15 as values of triple Massey products. Throughout, m = n − 1 m=n-1 m = n − 1 and E \mathcal{E} E , ω \omega ω , L L L , c c c , and P e P_e P e have the meanings of Section 10.7.
Totalization and the Heisenberg algebra
E.1. Totalization and the Heisenberg algebra. On a purely odd two-dimensional row space, the only moment generators are u = ∑ i θ i u=\sum_i\theta_i u = ∑ i θ i , v = ∑ i ϕ i v=\sum_i\phi_i v = ∑ i ϕ i , and q = ∑ i θ i ϕ i q=\sum_i\theta_i\phi_i q = ∑ i θ i ϕ i . Let ξ \xi ξ , ζ \zeta ζ , η \eta η be their Koszul generators. For internal degree D D D and homological degree j j j , assign cohomological degree D − j D-j D − j and twist the product by
a ⋆ b = ( − 1 ) D ( a ) j ( b ) a b . (119) a\star b=(-1)^{D(a)j(b)}ab. \tag*{(119)} a ⋆ b = ( − 1 ) D ( a ) j ( b ) ab . (119)
The original commutation sign ( − 1 ) D ( a ) D ( b ) + j ( a ) j ( b ) (-1)^{D(a)D(b)+j(a)j(b)} ( − 1 ) D ( a ) D ( b ) + j ( a ) j ( b ) becomes the cohomological commutation sign, and the differential satisfies the corresponding Leibniz rule without modification.
Now ξ , ζ \xi,\zeta ξ , ζ have cohomological degree zero, with d ξ = u d\xi=u d ξ = u and d ζ = v d\zeta=v d ζ = v . Centering the odd variables and making the G L ( U ) GL(U) G L ( U ) -equivariant change
ω = q − u v n , η ′ = η − ξ ⋆ v − ζ ⋆ u 2 n \omega=q-\frac{uv}{n},\qquad\eta'=\eta-\frac{\xi\star v-\zeta\star u}{2n} ω = q − n uv , η ′ = η − 2 n ξ ⋆ v − ζ ⋆ u
gives d η ′ = ω d\eta'=\omega d η ′ = ω . The two factors ( K [ ξ ] ⊗ Λ ( u ) , d ξ = u ) (K[\xi]\otimes\Lambda(u),d\xi=u) ( K [ ξ ] ⊗ Λ ( u ) , d ξ = u ) and ( K [ ζ ] ⊗ Λ ( v ) , d ζ = v ) (K[\zeta]\otimes\Lambda(v),d\zeta=v) ( K [ ζ ] ⊗ Λ ( v ) , d ζ = v ) have only constant cohomology in characteristic zero. Their joint augmentation is a quasi-isomorphism onto
A m = ( Λ ( θ 1 , … , θ m , ϕ 1 , … , ϕ m , η ) , d η = ω ) . (120) A_m=\left(\Lambda(\theta_1,\ldots,\theta_m,\phi_1,\ldots,\phi_m,\eta),d\eta=\omega\right). \tag*{(120)} A m = ( Λ ( θ 1 , … , θ m , ϕ 1 , … , ϕ m , η ) , d η = ω ) . (120)
This is the Chevalley–Eilenberg algebra of the ( 2 m + 1 ) (2m+1) ( 2 m + 1 ) -dimensional Heisenberg Lie algebra, up to the sign convention for its central generator. Its cohomology is classical; see [10 ] . Here it is the actual complete odd specialization of the moment complex, with its product.
Put R = E / ( ω ) R=\mathcal{E}/(\omega) R = E / ( ω ) and I = η Ann E ( ω ) I=\eta\operatorname{Ann}_{\mathcal{E}}(\omega) I = η Ann E ( ω ) . Cycles and boundaries in (E.2) give the ring identity
H ∗ ( A m ) = R ⊕ I , I 2 = 0. (121) H^*(A_m)=R\oplus I,\qquad I^2=0. \tag*{(121)} H ∗ ( A m ) = R ⊕ I , I 2 = 0. (121)
The exterior top-product pairing identifies I j I^j I j with the dual of R 2 m + 1 − j R^{2m+1-j} R 2 m + 1 − j , tensored with the orientation line ( det U ) m + 1 ⊗ ( det W ) 2 (\det U)^{m+1}\otimes(\det W)^2 ( det U ) m + 1 ⊗ ( det W ) 2 . The orientation class is η ω m / m ! \eta\omega^m/m! η ω m / m ! , of cohomological degree 2 n − 1 2n-1 2 n − 1 and internal degree 2 n 2n 2 n . Thus (E.3) is a Poincaré-duality trivial extension. By Proposition 10.15 and this pairing, Soc ( R ) = R m = P m \operatorname{Soc}(R)=R_m=P_m Soc ( R ) = R m = P m ; its type is the Catalan number 1 n + 1 ( 2 n n ) \frac{1}{n+1}\binom{2n}{n} n + 1 1 ( n 2 n ) , so R R R is not Gorenstein when n ≥ 2 n\ge2 n ≥ 2 . Also,
dim R e = ( 2 m e ) − ( 2 m e − 2 ) ( 0 ≤ e ≤ m ) , dim H ∗ ( A m ) = ( 2 n n ) . \dim R_e=\binom{2m}{e}-\binom{2m}{e-2}\quad(0\le e\le m),\qquad\dim H^*(A_m)=\binom{2n}{n}. dim R e = ( e 2 m ) − ( e − 2 2 m ) ( 0 ≤ e ≤ m ) , dim H ∗ ( A m ) = ( n 2 n ) .
Fixing ω \omega ω gives an action of S p 2 m Sp_{2m} S p 2 m on the differential algebra. Here P 0 = K P_0=K P 0 = K and P e P_e P e is the primitive fundamental representation for 1 ≤ e ≤ m 1\le e\le m 1 ≤ e ≤ m . This larger symmetry mixes row components; a fixed G L ( U ) GL(U) G L ( U ) highest-weight space need not be stable.
E.2. Massey products and all positive homology. For β ∈ U ⊗ W \beta\in U\otimes W β ∈ U ⊗ W , set ι β = [ c , ε β ] \iota_\beta=[c,\varepsilon_\beta] ι β = [ c , ε β ] and T β = ε β ι β T_\beta=\varepsilon_\beta\iota_\beta T β = ε β ι β . In Darboux coordinates, ι θ i = ∂ ϕ i \iota_{\theta_i}=\partial_{\phi_i} ι θ i = ∂ ϕ i and ι ϕ i = − ∂ θ i \iota_{\phi_i}=-\partial_{\theta_i} ι ϕ i = − ∂ θ i . The operator T β T_\beta T β is the rank-one symplectic infinitesimal transformation, acting as a derivation on E \mathcal{E} E .
Theorem E.1 (Massey generation of positive moment homology). For m ≥ 1 m\ge1 m ≥ 1 , α ∈ P m \alpha\in P_m α ∈ P m , and β ∈ U ⊗ W \beta\in U\otimes W β ∈ U ⊗ W , the triple Massey product is defined, has zero indeterminacy, and equals
⟨ [ α ] , [ β ] , [ β ] ⟩ = { ( − 1 ) m [ η T β α ] } . (122) \langle[\alpha],[\beta],[\beta]\rangle=\{(-1)^m[\eta T_\beta\alpha]\}. \tag*{(122)} ⟨[ α ] , [ β ] , [ β ]⟩ = {( − 1 ) m [ η T β α ]} . (122) Its values span the minimal generator space η P m \eta P_m η P m and generate all positive moment homology as an R R R -module.
Proof. Since L α = 0 L\alpha=0 Lα = 0 , the identity [ L , ι β ] = ε β [L,\iota_\beta]=\varepsilon_\beta [ L , ι β ] = ε β gives L ι β α = β α L\iota_\beta\alpha=\beta\alpha L ι β α = β α . Use the convention that d x = a b dx=ab d x = ab , d y = b c dy=bc d y = b c represents the triple product by a y + ( − 1 ) ∣ a ∣ + 1 x c ay+(-1)^{|a|+1}xc a y + ( − 1 ) ∣ a ∣ + 1 x c . Taking x = ( − 1 ) m η ι β α x=(-1)^m\eta\iota_\beta\alpha x = ( − 1 ) m η ι β α and y = 0 y=0 y = 0 proves (E.4). Its indeterminacy is [ α ] H 1 ( A m ) + H m ( A m ) [ β ] = 0 [\alpha]H^1(A_m)+H^m(A_m)[\beta]=0 [ α ] H 1 ( A m ) + H m ( A m ) [ β ] = 0 : these groups have no η \eta η part, and their indicated products lie in R m + 1 = 0 R^{m+1}=0 R m + 1 = 0 .
Polarizing T β T_\beta T β identifies their span with s p 2 m ≅ Sym 2 ( U ⊗ W ) \mathfrak{sp}_{2m}\cong\operatorname{Sym}^2(U\otimes W) sp 2 m ≅ Sym 2 ( U ⊗ W ) . The middle primitive module P m P_m P m is nontrivial and irreducible, so s p 2 m P m = P m \mathfrak{sp}_{2m}P_m=P_m sp 2 m P m = P m . Proposition 10.15 now gives the final assertion. □
For example,
⟨ [ θ 1 ⋯ θ m ] , [ ϕ 1 ] , [ ϕ 1 ] ⟩ = { η θ 2 ⋯ θ m ϕ 1 } \left\langle[\theta_1\cdots\theta_m],[\phi_1],[\phi_1]\right\rangle=\{\eta\theta_2\cdots\theta_m\phi_1\} ⟨ [ θ 1 ⋯ θ m ] , [ ϕ 1 ] , [ ϕ 1 ] ⟩ = { η θ 2 ⋯ θ m ϕ 1 }
is nonzero. Consequently the odd moment differential algebra is nonformal for every n ≥ 2 n\ge2 n ≥ 2 . For n = 1 n=1 n = 1 the reduced algebra is Λ ( η ) \Lambda(\eta) Λ ( η ) and is formal. These assertions concern this odd specialization, not every bosonic specialization.
The complete ternary operation
Define the Lefschetz partial inverse J J J by
J ( L j f ) = L j − 1 f ( j > 0 ) , J ( f ) = 0 ( f ∈ P e ) , π = 1 − L J , Q = 1 − J L . J(L^j f)=L^{j-1}f\quad(j>0),\qquad J(f)=0\quad(f\in P_e),\qquad\pi=1-LJ,\qquad Q=1-JL. J ( L j f ) = L j − 1 f ( j > 0 ) , J ( f ) = 0 ( f ∈ P e ) , π = 1 − L J , Q = 1 − J L .
The operators π \pi π and Q Q Q project to initial and terminal string vectors, respectively. Identify R R R with π ( E ) \pi(\mathcal{E}) π ( E ) .
Proposition E.2 (A minimal model with only binary and ternary operations). There is a minimal A ∞ A_\infty A ∞ model on R ⊕ I R\oplus I R ⊕ I , with its usual cohomology product, for which
m 3 ( a , b , c ) = η Q J ( a b ) c − a J ( b c ) ( a , b , c ∈ R ) . (123) m_3(a,b,c)=\eta QJ(ab)c-aJ(bc)\qquad(a,b,c\in R). \tag*{(123)} m 3 ( a , b , c ) = η Q J ( ab ) c − a J ( b c ) ( a , b , c ∈ R ) . (123) It has m 3 = 0 m_3=0 m 3 = 0 if an input lies in I I I , and m r = 0 m_r=0 m r = 0 for every r ≥ 4 r\ge4 r ≥ 4 .
Proof. There is an internal-degree-preserving deformation retract with h ( a ) = η J ( a ) h(a)=\eta J(a) h ( a ) = η J ( a ) , h ( η b ) = 0 h(\eta b)=0 h ( η b ) = 0 and p ( a + η b ) = π ( a ) + η Q ( b ) p(a+\eta b)=\pi(a)+\eta Q(b) p ( a + η b ) = π ( a ) + η Q ( b ) . If i i i includes the chosen representatives, then d h + h d = 1 − i p dh+hd=1-ip d h + h d = 1 − i p . Apply homotopy transfer, with the cohomological conventions of [22 Section 3] :
∑ r + s + t = N ( − 1 ) r + s t m r + 1 + t ( id ⊗ r ⊗ m s ⊗ id ⊗ t ) = 0. \sum_{r+s+t=N}(-1)^{r+st}m_{r+1+t}(\operatorname{id}^{\otimes r}\otimes m_s\otimes\operatorname{id}^{\otimes t})=0. r + s + t = N ∑ ( − 1 ) r + s t m r + 1 + t ( id ⊗ r ⊗ m s ⊗ id ⊗ t ) = 0.
The tensor products of maps carry their Koszul signs. The two ternary transfer terms are p ( h ( a b ) c − ( − 1 ) ∣ a ∣ a h ( b c ) ) p(h(ab)c-(-1)^{|a|}ah(bc)) p ( h ( ab ) c − ( − 1 ) ∣ a ∣ ah ( b c )) ; moving a a a across η \eta η gives (E.5). Every transferred operation preserves internal degree. Since ∣ m r ∣ = 2 − r |m_r|=2-r ∣ m r ∣ = 2 − r , its output moment-homological index is ∑ j i n + r − 2 \sum j_{\mathrm{in}}+r-2 ∑ j in + r − 2 . Only indices zero and one survive, which proves the vanishing assertions. □
In this convention m 3 ( α , β , β ) = − η T β α m_3(\alpha,\beta,\beta)=-\eta T_\beta\alpha m 3 ( α , β , β ) = − η T β α for α ∈ P m \alpha\in P_m α ∈ P m . The Massey convention in (E.4) differs by ( − 1 ) m + 1 (-1)^{m+1} ( − 1 ) m + 1 , as it must. Formula (E.5) also specifies the operation when an ordinary Massey product is not defined. The cutoff concerns this chosen bigraded minimal model, not all possible higher ordinary Massey products.
Acknowledgment
The author used ChatGPT for mathematical exploration and proof development, literature searches, computational experiments, and code for numerical certificates. Proofs developed with this assistance were carefully checked, adapted, and rewritten by the author, who also rewrote the exposition. The author takes full responsibility for the mathematical content, computations, and any errors.
Email address: eren321@gmail.com