completely log concave polynomials and matroids
play

Completely log-concave polynomials and matroids Cynthia Vinzant - PowerPoint PPT Presentation

Completely log-concave polynomials and matroids Cynthia Vinzant (North Carolina State University) joint work with Nima Anari, KuiKui Liu, Shayan Oveis Gharan (Stanford) (U. Washington) (U. Washington) Cynthia Vinzant Completely


  1. Completely log-concave polynomials and matroids ⇒ Cynthia Vinzant (North Carolina State University) joint work with Nima Anari, KuiKui Liu, Shayan Oveis Gharan (Stanford) (U. Washington) (U. Washington) Cynthia Vinzant Completely log-concave polynomials and matroids

  2. Matroids A matroid on ground set [ n ] = { 1 , . . . , n } is a nonempty collection I of independent subsets of [ n ] satisfying: ◮ If S ⊆ T and T ∈ I , then S ∈ I . ◮ If S , T ∈ I and | T | > | S | , then ∃ i ∈ T \ S with S ∪ { i } ∈ I . Examples: ◮ linear independence of vectors v 1 , . . . , v n ∈ R d ◮ cyclic independence of n edges in a graph I ∈I y n −| I | � Independence poly. g M ( y , z 1 , . . . , z n ) = � i ∈ I z i Cynthia Vinzant Completely log-concave polynomials and matroids

  3. Results and other work Mason’s conjecture: Let I k = # indep. sets of matroid M of size k . (i) I 2 k ≥ I k − 1 · I k +1 (log-concavity) � k +1 (ii) I 2 � k ≥ · I k − 1 · I k +1 k � 2 � I k ≥ I k − 1 � · I k +1 (iii) (ultra log-concavity) � n � n � n � � k k − 1 k +1 Cynthia Vinzant Completely log-concave polynomials and matroids

  4. Results and other work Mason’s conjecture: Let I k = # indep. sets of matroid M of size k . (i) I 2 k ≥ I k − 1 · I k +1 (log-concavity) � k +1 (ii) I 2 � k ≥ · I k − 1 · I k +1 k � 2 � I k ≥ I k − 1 � · I k +1 (iii) (ultra log-concavity) � n � n � n � � k k − 1 k +1 Adiprasito, Huh, Katz use combinatorial Hodge theory to prove (i) Huh, Schr¨ oter, Wang use ↑ to prove (ii) Anari, Liu, Oveis Gharan, V. use complete log-concavity to prove (iii) Br¨ and´ en, Huh independently use Lorentz polynomials to prove (iii) Cynthia Vinzant Completely log-concave polynomials and matroids

  5. Complete log-concavity f ∈ R [ z 1 , . . . , z n ] is log-concave on R n > 0 if f ≡ 0 or f ( x ) ≥ 0 for all x ∈ R n log( f ) is concave on R n and > 0 . ≥ 0 Cynthia Vinzant Completely log-concave polynomials and matroids

  6. Complete log-concavity f ∈ R [ z 1 , . . . , z n ] is log-concave on R n > 0 if f ≡ 0 or f ( x ) ≥ 0 for all x ∈ R n log( f ) is concave on R n and > 0 . ≥ 0 For v = ( v 1 , . . . , v n ) ∈ R n , let D v = � n i =1 v i ∂ f ∂ z i . f ∈ R [ z 1 , . . . , z n ] is completely log-concave (CLC) on R n > 0 if for all k ∈ N , v 1 , . . . , v k ∈ R n ≥ 0 , D v 1 · · · D v k f is log-concave on R n ≥ 0 . Cynthia Vinzant Completely log-concave polynomials and matroids

  7. Complete log-concavity f ∈ R [ z 1 , . . . , z n ] is log-concave on R n > 0 if f ≡ 0 or f ( x ) ≥ 0 for all x ∈ R n log( f ) is concave on R n and > 0 . ≥ 0 For v = ( v 1 , . . . , v n ) ∈ R n , let D v = � n i =1 v i ∂ f ∂ z i . f ∈ R [ z 1 , . . . , z n ] is completely log-concave (CLC) on R n > 0 if for all k ∈ N , v 1 , . . . , v k ∈ R n ≥ 0 , D v 1 · · · D v k f is log-concave on R n ≥ 0 . log( f ) ′′ = � d Example: f = � d − 1 i =1 ( z + r i ) ⇒ ( z + r i ) 2 ≤ 0 i =1 Cynthia Vinzant Completely log-concave polynomials and matroids

  8. Example: stable polynomials f ∈ R [ z 1 , . . . , z n ] d is stable if f ( tv + w ) ∈ R [ t ] is real rooted for all v ∈ R n ≥ 0 , w ∈ R n . ⇒ f is completely log-concave Cynthia Vinzant Completely log-concave polynomials and matroids

  9. Example: stable polynomials f ∈ R [ z 1 , . . . , z n ] d is stable if f ( tv + w ) ∈ R [ t ] is real rooted for all v ∈ R n ≥ 0 , w ∈ R n . ⇒ f is completely log-concave Example: det( � n i =1 z i v i v T d ) det( v i : i ∈ I ) 2 � i ) = � i ∈ I z i I ∈ ( [ n ] Cynthia Vinzant Completely log-concave polynomials and matroids

  10. Example: stable polynomials f ∈ R [ z 1 , . . . , z n ] d is stable if f ( tv + w ) ∈ R [ t ] is real rooted for all v ∈ R n ≥ 0 , w ∈ R n . ⇒ f is completely log-concave Example: det( � n i =1 z i v i v T d ) det( v i : i ∈ I ) 2 � i ) = � i ∈ I z i I ∈ ( [ n ] Choe, Oxley, Sokal, Wagner: If f = � d ) c I � i ∈ I z i is stable, I ∈ ( [ n ] then supp ( f ) = { I : c I � = 0 } are the bases of a matroid on [ n ]. Br¨ and´ en: Fano matroid � = support of a stable polynomial f Cynthia Vinzant Completely log-concave polynomials and matroids

  11. Equivalent conditions and univariate characterization Gurvits: f is strongly log-concave (SLC) if ∂ z 1 ) α 1 · · · ( ∂ ∂ α f = ( ∂ ∂ z n ) α n f is log-concave on R n ≥ 0 . Cynthia Vinzant Completely log-concave polynomials and matroids

  12. Equivalent conditions and univariate characterization Gurvits: f is strongly log-concave (SLC) if ∂ z 1 ) α 1 · · · ( ∂ ∂ α f = ( ∂ ∂ z n ) α n f is log-concave on R n ≥ 0 . Theorem (ALOV): For f ∈ R [ z 1 , . . . , z n ] d , � ∂ α f is indecomposable for all | α | ≤ d − 2 f CLC ⇔ f SLC ⇔ and ∂ α f is CLC for all | α | = d − 2 Cynthia Vinzant Completely log-concave polynomials and matroids

  13. Equivalent conditions and univariate characterization Gurvits: f is strongly log-concave (SLC) if ∂ z 1 ) α 1 · · · ( ∂ ∂ α f = ( ∂ ∂ z n ) α n f is log-concave on R n ≥ 0 . Theorem (ALOV): For f ∈ R [ z 1 , . . . , z n ] d , � ∂ α f is indecomposable for all | α | ≤ d − 2 f CLC ⇔ f SLC ⇔ and ∂ α f is CLC for all | α | = d − 2 (d=2) f = z T Qz is CLC ⇔ Q ij ≥ 0 and Q has 1 pos. eig. value. Cynthia Vinzant Completely log-concave polynomials and matroids

  14. Equivalent conditions and univariate characterization Gurvits: f is strongly log-concave (SLC) if ∂ z 1 ) α 1 · · · ( ∂ ∂ α f = ( ∂ ∂ z n ) α n f is log-concave on R n ≥ 0 . Theorem (ALOV): For f ∈ R [ z 1 , . . . , z n ] d , � ∂ α f is indecomposable for all | α | ≤ d − 2 f CLC ⇔ f SLC ⇔ and ∂ α f is CLC for all | α | = d − 2 (d=2) f = z T Qz is CLC ⇔ Q ij ≥ 0 and Q has 1 pos. eig. value. Cor. (Gurvits/ALOV) � 2 n � a k ≥ a k − 1 � · a k +1 a k y n − k z k is CLC ⇔ � � n � n � n � � k − 1 k +1 k k =0 Cynthia Vinzant Completely log-concave polynomials and matroids

  15. Complete log-concavity for matroids I ∈I y n −| I | � Theorem. g M ( y , z 1 , . . . , z n ) = � i ∈I z i is CLC. (just check rank-two matroids M ) k =0 I k y n − k z k is CLC. Cor: g M ( y , z , . . . , z ) = � n Cor: {I k } k is ultra log-concave (Mason’s conjecture) Cynthia Vinzant Completely log-concave polynomials and matroids

  16. Other results Theorem: For any matroid M , the solution to the concave program n � p i log 1 1 τ = max p i + (1 − p i ) log 1 − p i p ∈P M i =1 can be computed in polynomial time and β = e τ satisfies 2 O ( − r ) β ≤ # bases of M ≤ β. Cynthia Vinzant Completely log-concave polynomials and matroids

  17. Other results Theorem: For any matroid M , the solution to the concave program n � p i log 1 1 τ = max p i + (1 − p i ) log 1 − p i p ∈P M i =1 can be computed in polynomial time and β = e τ satisfies 2 O ( − r ) β ≤ # bases of M ≤ β. Theorem: The natural Markov Chain P ( B , B ′ ) on the bases of any rank- r matroid on [ n ] mixes quickly: min { t ∈ N : || P t ( B , · ) − π || 1 ≤ ǫ } ≤ r 2 log( n /ǫ ) . Cynthia Vinzant Completely log-concave polynomials and matroids

  18. Sum up: completely log-concave polynomials ◮ log-concavity of polynomial as functions ⇒ log-concavity of coefficients ◮ many matroid polynomials are completely log-concave ◮ the theory of stable polynomials extends to CLC polynomials ⇒ Cynthia Vinzant Completely log-concave polynomials and matroids

  19. Sum up: completely log-concave polynomials ◮ log-concavity of polynomial as functions ⇒ log-concavity of coefficients ◮ many matroid polynomials are completely log-concave ◮ the theory of stable polynomials extends to CLC polynomials ⇒ Thanks! Cynthia Vinzant Completely log-concave polynomials and matroids

  20. References ◮ Karim Adiprasito, June Huh, Eric Katz, Hodge theory for combinatorial geometries , Annals of Mathematics 188(2), 2018. ◮ Nima Anari, Shayan Oveis Gharan, Cynthia Vinzant, Log-Concave Polynomials I: Entropy and a Deterministic Approximation Algorithm for Counting Bases of Matroids , arXiv:1807.00929 ◮ Nima Anari, KuiKui Liu, Shayan Oveis Gharan, Cynthia Vinzant, Log-Concave Polynomials II: High-Dimensional Walks and an FPRAS for Counting Bases of a Matroid , arXiv:1811.01816 ◮ Nima Anari, KuiKui Liu, Shayan Oveis Gharan, Cynthia Vinzant, Log-Concave Polynomials III: Mason’s Ultra-Log-Concavity Conjecture for Independent Sets of Matroids , arXiv:1811.01600 ◮ Petter Br¨ and´ en, Polynomials with the half-plane property and matroid theory , Advances in Mathematics 216(1), 2007. ◮ Petter Br¨ and´ en, June Huh, Hodge-Riemann relations for Potts model partition functions , arXiv:1811.01696 ◮ Young-Bin Choe, James Oxley, Alan Sokal, David Wagner, Homogeneous multivariate polynomials with the half-plane property , Advances in Applied Mathematics, 32(1-2), 2004. ◮ Leonid Gurvits, On multivariate Newton-like inequalities , Advances in combinatorial mathematics, 61–78, 2009. ◮ June Huh, Benjamin Schr¨ oter, Botong Wang, Correlation bounds for fields and matroids , arXiv:1806.02675 Cynthia Vinzant Completely log-concave polynomials and matroids

Recommend


More recommend