on semiring complexity of schur polynomialsdima grigoriev (cnrs) complexity of symmetric functions...

56
On Semiring Complexity of Schur Polynomials Dima Grigoriev (Lille, CNRS) jointly with S. Fomin, D. Nogneng, E. Schost 15/06/2017, Nice Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 1/8

Upload: others

Post on 25-Jan-2021

1 views

Category:

Documents


0 download

TRANSCRIPT

  • On Semiring Complexity of Schur Polynomials

    Dima Grigoriev (Lille, CNRS)

    jointly with S. Fomin, D. Nogneng, E. Schost

    15/06/2017, Nice

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 1 / 8

  • Elementary symmetric functionsek ,n =

    ∑1≤i1

  • Elementary symmetric functionsek ,n =

    ∑1≤i1

  • Elementary symmetric functionsek ,n =

    ∑1≤i1

  • Elementary symmetric functionsek ,n =

    ∑1≤i1

  • Elementary symmetric functionsek ,n =

    ∑1≤i1

  • Elementary symmetric functionsek ,n =

    ∑1≤i1

  • Elementary symmetric functionsek ,n =

    ∑1≤i1

  • Elementary symmetric functionsek ,n =

    ∑1≤i1

  • Monomial symmetric functions

    For a partition λ = {λ1 ≥ λ2 ≥ · · · ≥ λn ≥ 0} definemλ =

    ∑π∈Sn x

    λ1π(1) · · · x

    λnπ(n),

    where the summation ranges over all the permutations π.Golomb ruler is a sequence of integers a1 > · · · > as such that thedifferences ai − aj , i < j are pairwise distinct.

    Theorem

    For a prime p the sequence λp−i+1 := 2pi + {i2 mod p}, 1 ≤ i ≤ p,where 0 ≤ i2 mod p < p, is a Golomb ruler (Erdös-Turán, 1941).

    TheoremIf λ is a Golomb ruler then the complexity C+,×(mλ) ≥ Ω(cn), c > 1.(G.-Koshevoy).

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 3 / 8

  • Monomial symmetric functions

    For a partition λ = {λ1 ≥ λ2 ≥ · · · ≥ λn ≥ 0} definemλ =

    ∑π∈Sn x

    λ1π(1) · · · x

    λnπ(n),

    where the summation ranges over all the permutations π.Golomb ruler is a sequence of integers a1 > · · · > as such that thedifferences ai − aj , i < j are pairwise distinct.

    Theorem

    For a prime p the sequence λp−i+1 := 2pi + {i2 mod p}, 1 ≤ i ≤ p,where 0 ≤ i2 mod p < p, is a Golomb ruler (Erdös-Turán, 1941).

    TheoremIf λ is a Golomb ruler then the complexity C+,×(mλ) ≥ Ω(cn), c > 1.(G.-Koshevoy).

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 3 / 8

  • Monomial symmetric functions

    For a partition λ = {λ1 ≥ λ2 ≥ · · · ≥ λn ≥ 0} definemλ =

    ∑π∈Sn x

    λ1π(1) · · · x

    λnπ(n),

    where the summation ranges over all the permutations π.Golomb ruler is a sequence of integers a1 > · · · > as such that thedifferences ai − aj , i < j are pairwise distinct.

    Theorem

    For a prime p the sequence λp−i+1 := 2pi + {i2 mod p}, 1 ≤ i ≤ p,where 0 ≤ i2 mod p < p, is a Golomb ruler (Erdös-Turán, 1941).

    TheoremIf λ is a Golomb ruler then the complexity C+,×(mλ) ≥ Ω(cn), c > 1.(G.-Koshevoy).

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 3 / 8

  • Monomial symmetric functions

    For a partition λ = {λ1 ≥ λ2 ≥ · · · ≥ λn ≥ 0} definemλ =

    ∑π∈Sn x

    λ1π(1) · · · x

    λnπ(n),

    where the summation ranges over all the permutations π.Golomb ruler is a sequence of integers a1 > · · · > as such that thedifferences ai − aj , i < j are pairwise distinct.

    Theorem

    For a prime p the sequence λp−i+1 := 2pi + {i2 mod p}, 1 ≤ i ≤ p,where 0 ≤ i2 mod p < p, is a Golomb ruler (Erdös-Turán, 1941).

    TheoremIf λ is a Golomb ruler then the complexity C+,×(mλ) ≥ Ω(cn), c > 1.(G.-Koshevoy).

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 3 / 8

  • Monomial symmetric functions

    For a partition λ = {λ1 ≥ λ2 ≥ · · · ≥ λn ≥ 0} definemλ =

    ∑π∈Sn x

    λ1π(1) · · · x

    λnπ(n),

    where the summation ranges over all the permutations π.Golomb ruler is a sequence of integers a1 > · · · > as such that thedifferences ai − aj , i < j are pairwise distinct.

    Theorem

    For a prime p the sequence λp−i+1 := 2pi + {i2 mod p}, 1 ≤ i ≤ p,where 0 ≤ i2 mod p < p, is a Golomb ruler (Erdös-Turán, 1941).

    TheoremIf λ is a Golomb ruler then the complexity C+,×(mλ) ≥ Ω(cn), c > 1.(G.-Koshevoy).

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 3 / 8

  • Lower bound on the complexity C+,×For a polynomial P denote by mon(P) the set of its monomials.

    LemmaLet P be a homogeneous polynomial in n variables. If for anyhomogeneous polynomials Q,R such that mon(P) ⊃ mon(Q · R) and13 · deg(P) ≤ deg(Q), deg(R) ≤

    23 · deg(P) we have

    |mon(P)| > cn1 · |mon(Q · R)|, c1 > 1, then C+,×(P) ≥ Ω(cn2), c2 > 1.(Schnorr, 1976, Valiant, 1980)

    LemmaLet λ be a Golomb ruler and let mon(Q · R) ⊂ mon(mλ) withhomogeneous Q,R. Then there is a subset S ⊂ {1, . . . ,n} such thatQ = Q′(xi , i ∈ S) ·M(xj , j 6∈ S), R = N(xi , i ∈ S) · R′(xj , j 6∈ S) whereM,N are monomials in variables xj , j 6∈ S and in xi , i ∈ S, respectively.(G.-Koshevoy).

    Thus, one can separate the variables in such a way that Q,R dependon disjoint subsets of variables up to monomial factors.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 4 / 8

  • Lower bound on the complexity C+,×For a polynomial P denote by mon(P) the set of its monomials.

    LemmaLet P be a homogeneous polynomial in n variables. If for anyhomogeneous polynomials Q,R such that mon(P) ⊃ mon(Q · R) and13 · deg(P) ≤ deg(Q), deg(R) ≤

    23 · deg(P) we have

    |mon(P)| > cn1 · |mon(Q · R)|, c1 > 1, then C+,×(P) ≥ Ω(cn2), c2 > 1.(Schnorr, 1976, Valiant, 1980)

    LemmaLet λ be a Golomb ruler and let mon(Q · R) ⊂ mon(mλ) withhomogeneous Q,R. Then there is a subset S ⊂ {1, . . . ,n} such thatQ = Q′(xi , i ∈ S) ·M(xj , j 6∈ S), R = N(xi , i ∈ S) · R′(xj , j 6∈ S) whereM,N are monomials in variables xj , j 6∈ S and in xi , i ∈ S, respectively.(G.-Koshevoy).

    Thus, one can separate the variables in such a way that Q,R dependon disjoint subsets of variables up to monomial factors.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 4 / 8

  • Lower bound on the complexity C+,×For a polynomial P denote by mon(P) the set of its monomials.

    LemmaLet P be a homogeneous polynomial in n variables. If for anyhomogeneous polynomials Q,R such that mon(P) ⊃ mon(Q · R) and13 · deg(P) ≤ deg(Q), deg(R) ≤

    23 · deg(P) we have

    |mon(P)| > cn1 · |mon(Q · R)|, c1 > 1, then C+,×(P) ≥ Ω(cn2), c2 > 1.(Schnorr, 1976, Valiant, 1980)

    LemmaLet λ be a Golomb ruler and let mon(Q · R) ⊂ mon(mλ) withhomogeneous Q,R. Then there is a subset S ⊂ {1, . . . ,n} such thatQ = Q′(xi , i ∈ S) ·M(xj , j 6∈ S), R = N(xi , i ∈ S) · R′(xj , j 6∈ S) whereM,N are monomials in variables xj , j 6∈ S and in xi , i ∈ S, respectively.(G.-Koshevoy).

    Thus, one can separate the variables in such a way that Q,R dependon disjoint subsets of variables up to monomial factors.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 4 / 8

  • Lower bound on the complexity C+,×For a polynomial P denote by mon(P) the set of its monomials.

    LemmaLet P be a homogeneous polynomial in n variables. If for anyhomogeneous polynomials Q,R such that mon(P) ⊃ mon(Q · R) and13 · deg(P) ≤ deg(Q), deg(R) ≤

    23 · deg(P) we have

    |mon(P)| > cn1 · |mon(Q · R)|, c1 > 1, then C+,×(P) ≥ Ω(cn2), c2 > 1.(Schnorr, 1976, Valiant, 1980)

    LemmaLet λ be a Golomb ruler and let mon(Q · R) ⊂ mon(mλ) withhomogeneous Q,R. Then there is a subset S ⊂ {1, . . . ,n} such thatQ = Q′(xi , i ∈ S) ·M(xj , j 6∈ S), R = N(xi , i ∈ S) · R′(xj , j 6∈ S) whereM,N are monomials in variables xj , j 6∈ S and in xi , i ∈ S, respectively.(G.-Koshevoy).

    Thus, one can separate the variables in such a way that Q,R dependon disjoint subsets of variables up to monomial factors.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 4 / 8

  • Lower bound on the complexity C+,×For a polynomial P denote by mon(P) the set of its monomials.

    LemmaLet P be a homogeneous polynomial in n variables. If for anyhomogeneous polynomials Q,R such that mon(P) ⊃ mon(Q · R) and13 · deg(P) ≤ deg(Q), deg(R) ≤

    23 · deg(P) we have

    |mon(P)| > cn1 · |mon(Q · R)|, c1 > 1, then C+,×(P) ≥ Ω(cn2), c2 > 1.(Schnorr, 1976, Valiant, 1980)

    LemmaLet λ be a Golomb ruler and let mon(Q · R) ⊂ mon(mλ) withhomogeneous Q,R. Then there is a subset S ⊂ {1, . . . ,n} such thatQ = Q′(xi , i ∈ S) ·M(xj , j 6∈ S), R = N(xi , i ∈ S) · R′(xj , j 6∈ S) whereM,N are monomials in variables xj , j 6∈ S and in xi , i ∈ S, respectively.(G.-Koshevoy).

    Thus, one can separate the variables in such a way that Q,R dependon disjoint subsets of variables up to monomial factors.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 4 / 8

  • Lower bound on the complexity C+,×For a polynomial P denote by mon(P) the set of its monomials.

    LemmaLet P be a homogeneous polynomial in n variables. If for anyhomogeneous polynomials Q,R such that mon(P) ⊃ mon(Q · R) and13 · deg(P) ≤ deg(Q), deg(R) ≤

    23 · deg(P) we have

    |mon(P)| > cn1 · |mon(Q · R)|, c1 > 1, then C+,×(P) ≥ Ω(cn2), c2 > 1.(Schnorr, 1976, Valiant, 1980)

    LemmaLet λ be a Golomb ruler and let mon(Q · R) ⊂ mon(mλ) withhomogeneous Q,R. Then there is a subset S ⊂ {1, . . . ,n} such thatQ = Q′(xi , i ∈ S) ·M(xj , j 6∈ S), R = N(xi , i ∈ S) · R′(xj , j 6∈ S) whereM,N are monomials in variables xj , j 6∈ S and in xi , i ∈ S, respectively.(G.-Koshevoy).

    Thus, one can separate the variables in such a way that Q,R dependon disjoint subsets of variables up to monomial factors.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 4 / 8

  • Lower bound on the complexity C+,×For a polynomial P denote by mon(P) the set of its monomials.

    LemmaLet P be a homogeneous polynomial in n variables. If for anyhomogeneous polynomials Q,R such that mon(P) ⊃ mon(Q · R) and13 · deg(P) ≤ deg(Q), deg(R) ≤

    23 · deg(P) we have

    |mon(P)| > cn1 · |mon(Q · R)|, c1 > 1, then C+,×(P) ≥ Ω(cn2), c2 > 1.(Schnorr, 1976, Valiant, 1980)

    LemmaLet λ be a Golomb ruler and let mon(Q · R) ⊂ mon(mλ) withhomogeneous Q,R. Then there is a subset S ⊂ {1, . . . ,n} such thatQ = Q′(xi , i ∈ S) ·M(xj , j 6∈ S), R = N(xi , i ∈ S) · R′(xj , j 6∈ S) whereM,N are monomials in variables xj , j 6∈ S and in xi , i ∈ S, respectively.(G.-Koshevoy).

    Thus, one can separate the variables in such a way that Q,R dependon disjoint subsets of variables up to monomial factors.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 4 / 8

  • Lower bound on the complexity C+,×For a polynomial P denote by mon(P) the set of its monomials.

    LemmaLet P be a homogeneous polynomial in n variables. If for anyhomogeneous polynomials Q,R such that mon(P) ⊃ mon(Q · R) and13 · deg(P) ≤ deg(Q), deg(R) ≤

    23 · deg(P) we have

    |mon(P)| > cn1 · |mon(Q · R)|, c1 > 1, then C+,×(P) ≥ Ω(cn2), c2 > 1.(Schnorr, 1976, Valiant, 1980)

    LemmaLet λ be a Golomb ruler and let mon(Q · R) ⊂ mon(mλ) withhomogeneous Q,R. Then there is a subset S ⊂ {1, . . . ,n} such thatQ = Q′(xi , i ∈ S) ·M(xj , j 6∈ S), R = N(xi , i ∈ S) · R′(xj , j 6∈ S) whereM,N are monomials in variables xj , j 6∈ S and in xi , i ∈ S, respectively.(G.-Koshevoy).

    Thus, one can separate the variables in such a way that Q,R dependon disjoint subsets of variables up to monomial factors.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 4 / 8

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Complete symmetric polynomials

    hk (x1, . . . , xn) :=∑

    1≤i1≤···≤ik≤n xi1 · · · xikis homogeneous of degree k . Denote h̃k := hk (x21 , . . . , x

    2n ). Then∑

    0≤m

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Schur polynomialsConsider a matrix with n rows and the infinite number of columnshaving (i , j)-entry equal x j−1i . For I = {i1 > · · · > in ≥ 1} denote by ∆Ithe n × n minor with the columns from I. In particular,∆n,n−1,...,1 =

    ∏i>j(xi − xj) is the Vandermonde minor.

    For a partition λ = {λ1 ≥ · · · ≥ λn ≥ 0} the Schur polynomialsλ := ∆λ1+n,λ2+n−1,...,λn+1/∆n,n−1,...,1i. e. a symmetric polynomial with non-negative integer (Kostka)coefficients of the degree |λ| = λ1 + · · ·+ λn. Computing of Kostkacoefficients is #P-hard. Note that for λ = {λ1,0, . . . ,0} we havesλ = hλ1 . From the definition we get C+,−,×,/(sλ) ≤ O(n3 · log |λ|).

    Theorem(V.Strassen, 1973). For any polynomial f it holdsC+,−,×(f ) ≤ O(C+,−,×,/(f ) · deg(f ) · log deg(f ))

    Corollary

    C+,−,×(sλ) ≤ O(n3 · |λ| · log2 |λ|).Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 6 / 8

  • Complexity of Schur polynomials

    Theorem

    (S.Fomin-G.-Koshevoy). C+,×,/(sλ) ≤ O((n + |λ|)3).

    The proof is based on the cluster transformations.

    Theorem

    (Demmel-Koev). C+,×(sλ) ≤ O(exp(|λ|) · n2).

    Theorem

    (S.Fomin-G.-Nogneng-Schost). C+,×(sλ) ≤ O(exp(n2) · log |λ|).

    Question. C+,×(sλ) ≤ (n + |λ|)O(1)?

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 7 / 8

  • Complexity of Schur polynomials

    Theorem

    (S.Fomin-G.-Koshevoy). C+,×,/(sλ) ≤ O((n + |λ|)3).

    The proof is based on the cluster transformations.

    Theorem

    (Demmel-Koev). C+,×(sλ) ≤ O(exp(|λ|) · n2).

    Theorem

    (S.Fomin-G.-Nogneng-Schost). C+,×(sλ) ≤ O(exp(n2) · log |λ|).

    Question. C+,×(sλ) ≤ (n + |λ|)O(1)?

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 7 / 8

  • Complexity of Schur polynomials

    Theorem

    (S.Fomin-G.-Koshevoy). C+,×,/(sλ) ≤ O((n + |λ|)3).

    The proof is based on the cluster transformations.

    Theorem

    (Demmel-Koev). C+,×(sλ) ≤ O(exp(|λ|) · n2).

    Theorem

    (S.Fomin-G.-Nogneng-Schost). C+,×(sλ) ≤ O(exp(n2) · log |λ|).

    Question. C+,×(sλ) ≤ (n + |λ|)O(1)?

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 7 / 8

  • Complexity of Schur polynomials

    Theorem

    (S.Fomin-G.-Koshevoy). C+,×,/(sλ) ≤ O((n + |λ|)3).

    The proof is based on the cluster transformations.

    Theorem

    (Demmel-Koev). C+,×(sλ) ≤ O(exp(|λ|) · n2).

    Theorem

    (S.Fomin-G.-Nogneng-Schost). C+,×(sλ) ≤ O(exp(n2) · log |λ|).

    Question. C+,×(sλ) ≤ (n + |λ|)O(1)?

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 7 / 8

  • Complexity of Schur polynomials

    Theorem

    (S.Fomin-G.-Koshevoy). C+,×,/(sλ) ≤ O((n + |λ|)3).

    The proof is based on the cluster transformations.

    Theorem

    (Demmel-Koev). C+,×(sλ) ≤ O(exp(|λ|) · n2).

    Theorem

    (S.Fomin-G.-Nogneng-Schost). C+,×(sλ) ≤ O(exp(n2) · log |λ|).

    Question. C+,×(sλ) ≤ (n + |λ|)O(1)?

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 7 / 8

  • Complexity of Schur polynomials

    Theorem

    (S.Fomin-G.-Koshevoy). C+,×,/(sλ) ≤ O((n + |λ|)3).

    The proof is based on the cluster transformations.

    Theorem

    (Demmel-Koev). C+,×(sλ) ≤ O(exp(|λ|) · n2).

    Theorem

    (S.Fomin-G.-Nogneng-Schost). C+,×(sλ) ≤ O(exp(n2) · log |λ|).

    Question. C+,×(sλ) ≤ (n + |λ|)O(1)?

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 7 / 8

  • Semi-standard Young tableaux

    The proof of the bound on C+,×(sλ) relies on the obtained bound onC+,×(hk ) and on the formula expressing sλ via the semi-standardYoung tableaux T with the shape λ:T = (ti,j), 1 ≤ ti,j ≤ n, 1 ≤ i ≤ n, 1 ≤ j ≤ λi ,ti+1,j > ti,j (strictly increasing down) andti,j+1 ≥ ti,j (non-decreasing to the right).Monomial xT :=

    ∏i,j∈T xti,j . Then sλ =

    ∑T x

    T where the summationranges over all the semi-standard tableaux with the shape λ.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 8 / 8

  • Semi-standard Young tableaux

    The proof of the bound on C+,×(sλ) relies on the obtained bound onC+,×(hk ) and on the formula expressing sλ via the semi-standardYoung tableaux T with the shape λ:T = (ti,j), 1 ≤ ti,j ≤ n, 1 ≤ i ≤ n, 1 ≤ j ≤ λi ,ti+1,j > ti,j (strictly increasing down) andti,j+1 ≥ ti,j (non-decreasing to the right).Monomial xT :=

    ∏i,j∈T xti,j . Then sλ =

    ∑T x

    T where the summationranges over all the semi-standard tableaux with the shape λ.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 8 / 8

  • Semi-standard Young tableaux

    The proof of the bound on C+,×(sλ) relies on the obtained bound onC+,×(hk ) and on the formula expressing sλ via the semi-standardYoung tableaux T with the shape λ:T = (ti,j), 1 ≤ ti,j ≤ n, 1 ≤ i ≤ n, 1 ≤ j ≤ λi ,ti+1,j > ti,j (strictly increasing down) andti,j+1 ≥ ti,j (non-decreasing to the right).Monomial xT :=

    ∏i,j∈T xti,j . Then sλ =

    ∑T x

    T where the summationranges over all the semi-standard tableaux with the shape λ.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 8 / 8

  • Semi-standard Young tableaux

    The proof of the bound on C+,×(sλ) relies on the obtained bound onC+,×(hk ) and on the formula expressing sλ via the semi-standardYoung tableaux T with the shape λ:T = (ti,j), 1 ≤ ti,j ≤ n, 1 ≤ i ≤ n, 1 ≤ j ≤ λi ,ti+1,j > ti,j (strictly increasing down) andti,j+1 ≥ ti,j (non-decreasing to the right).Monomial xT :=

    ∏i,j∈T xti,j . Then sλ =

    ∑T x

    T where the summationranges over all the semi-standard tableaux with the shape λ.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 8 / 8

  • Semi-standard Young tableaux

    The proof of the bound on C+,×(sλ) relies on the obtained bound onC+,×(hk ) and on the formula expressing sλ via the semi-standardYoung tableaux T with the shape λ:T = (ti,j), 1 ≤ ti,j ≤ n, 1 ≤ i ≤ n, 1 ≤ j ≤ λi ,ti+1,j > ti,j (strictly increasing down) andti,j+1 ≥ ti,j (non-decreasing to the right).Monomial xT :=

    ∏i,j∈T xti,j . Then sλ =

    ∑T x

    T where the summationranges over all the semi-standard tableaux with the shape λ.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 8 / 8

  • Semi-standard Young tableaux

    The proof of the bound on C+,×(sλ) relies on the obtained bound onC+,×(hk ) and on the formula expressing sλ via the semi-standardYoung tableaux T with the shape λ:T = (ti,j), 1 ≤ ti,j ≤ n, 1 ≤ i ≤ n, 1 ≤ j ≤ λi ,ti+1,j > ti,j (strictly increasing down) andti,j+1 ≥ ti,j (non-decreasing to the right).Monomial xT :=

    ∏i,j∈T xti,j . Then sλ =

    ∑T x

    T where the summationranges over all the semi-standard tableaux with the shape λ.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 8 / 8

  • Semi-standard Young tableaux

    The proof of the bound on C+,×(sλ) relies on the obtained bound onC+,×(hk ) and on the formula expressing sλ via the semi-standardYoung tableaux T with the shape λ:T = (ti,j), 1 ≤ ti,j ≤ n, 1 ≤ i ≤ n, 1 ≤ j ≤ λi ,ti+1,j > ti,j (strictly increasing down) andti,j+1 ≥ ti,j (non-decreasing to the right).Monomial xT :=

    ∏i,j∈T xti,j . Then sλ =

    ∑T x

    T where the summationranges over all the semi-standard tableaux with the shape λ.

    Dima Grigoriev (CNRS) Complexity of Symmetric Functions 15.06.17 8 / 8