Tensor Invariants and Kronecker Coefficients
Jiarui Fei
University of California, Riverside
November 23, 2013
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Tensor Invariants and Kronecker Coefficients Jiarui Fei University - - PowerPoint PPT Presentation
Tensor Invariants and Kronecker Coefficients Jiarui Fei University of California, Riverside November 23, 2013 Jiarui Fei Tensor Invariants and Kronecker Coefficients Tensor Invariants By a (tri-)tensor of vector spaces ( U , V , W ), we mean
Jiarui Fei
University of California, Riverside
November 23, 2013
Jiarui Fei Tensor Invariants and Kronecker Coefficients
By a (tri-)tensor of vector spaces (U, V , W ), we mean the vector space U∗ ⊗ V ⊗ W ∗. The product of special linear groups G := SL(U) × SL(V ) × SL(W ) acts naturally on it. We are interested in the ring of invariants k[U∗ ⊗ V ⊗ W ∗]G = Sym(U ⊗ V ∗ ⊗ W )G. The vector space can be identified with the (n1, n2)-dimensional representation space of the m-arrow Kronecker quiver Km, where dim U = n1, dim V = n2, and dim W = m. The ring of invariants SI(n1,n2)(Km) := k[U∗ ⊗ V ⊗ W ∗]SL(U)×SL(V ) is generated by Schofield’s determinantal invariants (Derksen-Weyman, Schofield-Van den Bergh).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
By a (tri-)tensor of vector spaces (U, V , W ), we mean the vector space U∗ ⊗ V ⊗ W ∗. The product of special linear groups G := SL(U) × SL(V ) × SL(W ) acts naturally on it. We are interested in the ring of invariants k[U∗ ⊗ V ⊗ W ∗]G = Sym(U ⊗ V ∗ ⊗ W )G. The vector space can be identified with the (n1, n2)-dimensional representation space of the m-arrow Kronecker quiver Km, where dim U = n1, dim V = n2, and dim W = m. The ring of invariants SI(n1,n2)(Km) := k[U∗ ⊗ V ⊗ W ∗]SL(U)×SL(V ) is generated by Schofield’s determinantal invariants (Derksen-Weyman, Schofield-Van den Bergh).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
For simplicity, we will assume n1 = n2 = n. By a semi-invariants of degree d, we mean an element f ∈ SI(n,n)(Km) of weight (d, −d), i.e., for any x ∈ U∗ ⊗ V ⊗ W ∗ and (g1, g2) ∈ GL(U) × GL(V ) we have that f ((g1, g2)x) = det(g1)d det(g2)−df (x). We consider the determinant of the dn × dn matrix
m
Λk ⊗ Ak, where Λk = (λk
ij) is a d × d matrix and ⊗ is the Kronecker product.
Let c(D1, D2, . . . , Dm) be the coefficient of
i,j,k(λk ij)(Dk)ij in the
expansion.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
For simplicity, we will assume n1 = n2 = n. By a semi-invariants of degree d, we mean an element f ∈ SI(n,n)(Km) of weight (d, −d), i.e., for any x ∈ U∗ ⊗ V ⊗ W ∗ and (g1, g2) ∈ GL(U) × GL(V ) we have that f ((g1, g2)x) = det(g1)d det(g2)−df (x). We consider the determinant of the dn × dn matrix
m
Λk ⊗ Ak, where Λk = (λk
ij) is a d × d matrix and ⊗ is the Kronecker product.
Let c(D1, D2, . . . , Dm) be the coefficient of
i,j,k(λk ij)(Dk)ij in the
expansion.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Example (m = 2, d = 3)
λ1
11A1 + λ2 11A2
λ1
12A1 + λ2 12A2
λ1
13A1 + λ2 13A2
λ1
21A1 + λ2 21A2
λ1
22A1 + λ2 22A2
λ1
23A1 + λ2 23A2
λ1
31A1 + λ2 31A2
λ1
32A1 + λ2 32A2
λ1
33A1 + λ2 33A2
Lemma
{c(D1, D2, . . . , Dm)} linearly span the quiver invariants SId
(n,n)(Km).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Example (m = 2, d = 3)
λ1
11A1 + λ2 11A2
λ1
12A1 + λ2 12A2
λ1
13A1 + λ2 13A2
λ1
21A1 + λ2 21A2
λ1
22A1 + λ2 22A2
λ1
23A1 + λ2 23A2
λ1
31A1 + λ2 31A2
λ1
32A1 + λ2 32A2
λ1
33A1 + λ2 33A2
Lemma
{c(D1, D2, . . . , Dm)} linearly span the quiver invariants SId
(n,n)(Km).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
We view c(D1, D2, . . . , Dm) as formal variables, and denote F(C, Λ) :=
(λk
ij)(Dk)ij.
Since we have for any elementary matrix Fij = I + Eij that, det(
FijΛk ⊗ Ak) = det(
Λk ⊗ Ak), such an equality generates a set of relations among c(D1, D2, . . . , Dm), namely F(C, FijΛ) − F(C, Λ) = 0.
Theorem
The relations below span all relations among c(D1, D2, . . . , Dm)’s. F(C, FijΛ) − F(C, Λ) = 0, and F(C, ΛFij) − F(C, Λ) = 0.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
We view c(D1, D2, . . . , Dm) as formal variables, and denote F(C, Λ) :=
(λk
ij)(Dk)ij.
Since we have for any elementary matrix Fij = I + Eij that, det(
FijΛk ⊗ Ak) = det(
Λk ⊗ Ak), such an equality generates a set of relations among c(D1, D2, . . . , Dm), namely F(C, FijΛ) − F(C, Λ) = 0.
Theorem
The relations below span all relations among c(D1, D2, . . . , Dm)’s. F(C, FijΛ) − F(C, Λ) = 0, and F(C, ΛFij) − F(C, Λ) = 0.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
This set of generators behave well under the Lie algebra sl(W )
matrix Ekl ∈ sl(W ). Then we have the following formula ∂kl
−
In particular, c(D) is T-invariant iff. m|dn and
n
n
Dij = dn m (1, 1, . . . , 1). (In priori,
n
m
Dk =
n
m
Dk = n(1, 1, . . . , 1).)
Jiarui Fei Tensor Invariants and Kronecker Coefficients
This set of generators behave well under the Lie algebra sl(W )
matrix Ekl ∈ sl(W ). Then we have the following formula ∂kl
−
In particular, c(D) is T-invariant iff. m|dn and
n
n
Dij = dn m (1, 1, . . . , 1). (In priori,
n
m
Dk =
n
m
Dk = n(1, 1, . . . , 1).)
Jiarui Fei Tensor Invariants and Kronecker Coefficients
222 The degree 2 invariant is Cayley’s hyperdeterminant c(11)2 − 4c(20)c(02). 233 The degree 4 invariant is c(12)2c(21)2 − 4c(03)c(21)3 − 4c(12)3c(30) +18c(03)c(12)c(21)c(30) − 27c(03)2c(30)2. 223 The degree 3 invariant is c 000 001 100
001 000 010 010 100 000
000 001 100
100 000 010 010 001 000
−A2 −A3 A1 A2 −A1
224 The degree 2 invariant is c( 1000 0100
0010 0001 ) − c( 1000 0010 0100 0001 ),
whose square is the classical invariant of quaternary quadratic
basis S(14), and has equivariant free resolution S(24) ⊗ A(−4) → A,
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Proposition
The tensor invariants are generated by three algebraically independent elements. The degree 2 invariant is c( 111 000
000 111 ) + 4
100 002 ) + c( 002 010 010 200 ) + c( 200 001 001 020 )
The degree 3 invariant is det
−A2 −A3 A1 A2 −A1
The degree 4 invariant is the Aronhold invariant of ternary cubic form. The invariant ring SI(3,3)(K3) is free over Sym(S3 ⊕ S(23)) with basis S(33), and has equivariant free resolution S(63) ⊗ A(−6) → A, where A = Sym(S3 ⊕ S(23) ⊕ S(33)).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Proposition
The tensor invariants are generated by three algebraically independent elements. The degree 2 invariant is c( 111 000
000 111 ) + 4
100 002 ) + c( 002 010 010 200 ) + c( 200 001 001 020 )
The degree 3 invariant is det
−A2 −A3 A1 A2 −A1
The degree 4 invariant is the Aronhold invariant of ternary cubic form. The invariant ring SI(3,3)(K3) is free over Sym(S3 ⊕ S(23)) with basis S(33), and has equivariant free resolution S(63) ⊗ A(−6) → A, where A = Sym(S3 ⊕ S(23) ⊕ S(33)).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
We want to know in advance how many linearly independent invariants in each degree and the degree-bounds. The number of invariants are given by some special Kronecker coefficients. Kronecker coefficients are by definition the structure coefficient of the representation ring of the symmetric groups SµSν =
kλ
µ,νSλ.
By Schur-Weyl duality, it also appears as Sλ(V ⊗ W ) ∼ =
kλ
µ,νSµ(V ) ⊗ Sν(W ).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Consider the following quiver K p,q
m
−p
· · · −2 −1
m 1
2 · · · q
Let the weight σ = (σ−p, . . . , σ−1, −σ1, . . . , −σq), and µT = (ασ−1
−1 , . . . , ασ−p −p , ), νT = (ασ1 1 , . . . , ασq q ) with |µ| = |ν| = n.
Then standard Schubert calculus shows that SIσ
α(K p,q m ) =
kλ
µ,νSλW
In particular, kλ
µ,ν = dim SIσ α(K p,q m )U(W ) λ
.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Consider the following quiver K p,q
m
−p
· · · −2 −1
m 1
2 · · · q
Let the weight σ = (σ−p, . . . , σ−1, −σ1, . . . , −σq), and µT = (ασ−1
−1 , . . . , ασ−p −p , ), νT = (ασ1 1 , . . . , ασq q ) with |µ| = |ν| = n.
Then standard Schubert calculus shows that SIσ
α(K p,q m ) =
kλ
µ,νSλW
In particular, kλ
µ,ν = dim SIσ α(K p,q m )U(W ) λ
.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
On the other hand, dim SIσ
α(K p,q m ) =
η1,η2,...,ηmcν η1,η2,...,ηm
Assume λ has m parts. By Cauchy and Jacobi-Trudi formula, kλ
µ,ν =
sgn(ω)
cµ
η1,η2,...,ηmcν η1,η2,...,ηm.
Each component
cµ
η1,η2,...,ηmcν η1,η2,...,ηm is the dimension
α(K p,q m ). This leads to an algorithm.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
On the other hand, dim SIσ
α(K p,q m ) =
η1,η2,...,ηmcν η1,η2,...,ηm
Assume λ has m parts. By Cauchy and Jacobi-Trudi formula, kλ
µ,ν =
sgn(ω)
cµ
η1,η2,...,ηmcν η1,η2,...,ηm.
Each component
cµ
η1,η2,...,ηmcν η1,η2,...,ηm is the dimension
α(K p,q m ). This leads to an algorithm.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
λ, µ, ν, the stretched Kronecker coefficients kcλ
cµ,cν is a
quasi-polynomial in c, but not a polynomial in general. It does not have the saturation property even in certain weak sense.
isomorphism of invariant rings SIσ
α(Q) ∼
= SIµk(σ)
µk(α)(µk(Q)).
(1) So we obtain many equalities among Kronecker coefficients. (Question: how to find these equalities conceptually?) Moreover, the tensor invariants of U × V × W and U × V ′ × W are the same if dim U × dim W − dim V = dim V ′.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
λ, µ, ν, the stretched Kronecker coefficients kcλ
cµ,cν is a
quasi-polynomial in c, but not a polynomial in general. It does not have the saturation property even in certain weak sense.
isomorphism of invariant rings SIσ
α(Q) ∼
= SIµk(σ)
µk(α)(µk(Q)).
(1) So we obtain many equalities among Kronecker coefficients. (Question: how to find these equalities conceptually?) Moreover, the tensor invariants of U × V × W and U × V ′ × W are the same if dim U × dim W − dim V = dim V ′.
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Conjecture
The degree bounds of the graded ring
SIcσ
α (K p,q m )U cλ is
polynomial in max( α, m) (or equivalently polynomial in max(l(λ), l(µ), l(ν))). This conjecture is equivalent to the relaxed modular index hypothesis in GCT VI, which implies the relaxed saturation
we can decide whether kcλ
cµ,cν > 0 in poly(|λ|, |µ|, |ν|, c).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Conjecture
The degree bounds of the graded ring
SIcσ
α (K p,q m )U cλ is
polynomial in max( α, m) (or equivalently polynomial in max(l(λ), l(µ), l(ν))). This conjecture is equivalent to the relaxed modular index hypothesis in GCT VI, which implies the relaxed saturation
we can decide whether kcλ
cµ,cν > 0 in poly(|λ|, |µ|, |ν|, c).
Jiarui Fei Tensor Invariants and Kronecker Coefficients
Tensor Invariants and Kronecker Coefficients