QUARTIC SPECTRAHEDRA. Bernd Sturmfels UC Berkeley and MPI Bonn. Joint work with John Christian Ottem, Kristian Ranestad and Cynthia Vinzant

Similar documents
Quartic Spectrahedra

Quartic Spectrahedra

SPECTRAHEDRA. Bernd Sturmfels UC Berkeley

Gram Spectrahedra. Lynn Chua UC Berkeley. joint work with Daniel Plaumann, Rainer Sinn, and Cynthia Vinzant

SPECTRAHEDRA. Bernd Sturmfels UC Berkeley

Semidefinite Programming

The Geometry of Semidefinite Programming. Bernd Sturmfels UC Berkeley

The Algebraic Degree of Semidefinite Programming

Sixty-Four Curves of Degree Six

Unbounded Convex Semialgebraic Sets as Spectrahedral Shadows

Voronoi Cells of Varieties

Exact algorithms: from Semidefinite to Hyperbolic programming

CONVEX ALGEBRAIC GEOMETRY. Bernd Sturmfels UC Berkeley

Rational Quartic Symmetroids. Martin Helsø Master s Thesis, Spring 2016

HYPERBOLIC POLYNOMIALS, INTERLACERS AND SUMS OF SQUARES

Arithmetic applications of Prym varieties in low genus. Nils Bruin (Simon Fraser University), Tübingen, September 28, 2018

arxiv: v3 [math.ag] 2 May 2018

Combinatorics and geometry of E 7

Representations of Positive Polynomials: Theory, Practice, and

LMI MODELLING 4. CONVEX LMI MODELLING. Didier HENRION. LAAS-CNRS Toulouse, FR Czech Tech Univ Prague, CZ. Universidad de Valladolid, SP March 2009

Symbolic computation in hyperbolic programming

ALGEBRAIC DEGREE OF POLYNOMIAL OPTIMIZATION. 1. Introduction. f 0 (x)

ALGEBRA: From Linear to Non-Linear. Bernd Sturmfels University of California at Berkeley

THE ENVELOPE OF LINES MEETING A FIXED LINE AND TANGENT TO TWO SPHERES

ALGEBRAIC BOUNDARIES OF HILBERT S SOS CONES

NOTES ON HYPERBOLICITY CONES

An Algebraic and Geometric Perspective on Exponential Families

PROBLEMS FOR VIASM MINICOURSE: GEOMETRY OF MODULI SPACES LAST UPDATED: DEC 25, 2013

COURSE ON LMI PART I.2 GEOMETRY OF LMI SETS. Didier HENRION henrion

On Maps Taking Lines to Plane Curves

where m is the maximal ideal of O X,p. Note that m/m 2 is a vector space. Suppose that we are given a morphism

Tensor decomposition and tensor rank

arxiv: v2 [math.ag] 1 Jul 2010

Real Algebraic Geometry in Convex Optimization. Cynthia Vinzant

Convex algebraic geometry, optimization and applications

d A 0 + m t k A k 0 whenever λ min (B k (x)) t k λ max (B k (x)) for k = 1, 2,..., m x n B n (k).

Ranks of Real Symmetric Tensors

0. Introduction 1 0. INTRODUCTION

LIFTS OF CONVEX SETS AND CONE FACTORIZATIONS JOÃO GOUVEIA, PABLO A. PARRILO, AND REKHA THOMAS

Convex algebraic geometry, optimization and applications

Combinatorial Types of Tropical Eigenvector

arxiv: v2 [math.ag] 3 Dec 2016

The geometry of the statistical model for range-based localization

arxiv: v1 [math.ag] 18 Dec 2017

OLIVIER SERMAN. Theorem 1.1. The moduli space of rank 3 vector bundles over a curve of genus 2 is a local complete intersection.

ELLIPTIC CURVES BJORN POONEN

Computational Methods in Finite Geometry

Computational Problems Using Riemann Theta Functions in Sage

Density of rational points on Enriques surfaces

SPACES OF RATIONAL CURVES IN COMPLETE INTERSECTIONS

Max-Planck-Institut für Mathematik in den Naturwissenschaften Leipzig

6-1 The Positivstellensatz P. Parrilo and S. Lall, ECC

Secant varieties. Marin Petkovic. November 23, 2015

Tensors. Notes by Mateusz Michalek and Bernd Sturmfels for the lecture on June 5, 2018, in the IMPRS Ringvorlesung Introduction to Nonlinear Algebra

10. Smooth Varieties. 82 Andreas Gathmann

Explicit Real Cubic Surfaces

TORIC REDUCTION AND TROPICAL GEOMETRY A.

KUMMER S QUARTICS AND NUMERICALLY REFLECTIVE INVOLUTIONS OF ENRIQUES SURFACES

Contents Preliminaries Non-abelian curve groups A special curve Characteristic Varieties

STATE COUNCIL OF EDUCATIONAL RESEARCH AND TRAINING TNCF DRAFT SYLLABUS.

Point counting and real multiplication on K3 surfaces

On complete degenerations of surfaces with ordinary singularities in

Quadratic families of elliptic curves and degree 1 conic bundles

Hyperbolic Polynomials and Generalized Clifford Algebras

Binomial Exercises A = 1 1 and 1

INTERSECTION OF THREE QUADRIC IN THE SPECIAL POSITION Milada Kočandrlová

Chapter 1. Preliminaries

Resolution of Singularities in Algebraic Varieties

Multidimensional Geometry and its Applications

Hilbert s 17th Problem to Semidefinite Programming & Convex Algebraic Geometry

Congruences and Concurrent Lines in Multi-View Geometry

Inflection Points on Real Plane Curves Having Many Pseudo-Lines

Equivariant semidefinite lifts and sum of squares hierarchies

Lecture 1. Toric Varieties: Basics

Fall, 2003 CIS 610. Advanced geometric methods. Homework 3. November 11, 2003; Due November 25, beginning of class

12. Hilbert Polynomials and Bézout s Theorem

THE HESSE PENCIL OF PLANE CUBIC CURVES

Polynomials, Ideals, and Gröbner Bases

Notes on Complex Analysis

Open Problems in Algebraic Statistics

AN INTRODUCTION TO MODULI SPACES OF CURVES CONTENTS

BRILL-NOETHER THEORY. This article follows the paper of Griffiths and Harris, "On the variety of special linear systems on a general algebraic curve.

MATH 423 Linear Algebra II Lecture 3: Subspaces of vector spaces. Review of complex numbers. Vector space over a field.

DEGREES PhD: Doktor der Naturwissenschaften (summa cum laude) May 2014; Advisor: Claus Scheiderer Diploma Degree in Mathematics (grade 1.

David Eklund. May 12, 2017

Mathematics Syllabus UNIT I ALGEBRA : 1. SETS, RELATIONS AND FUNCTIONS

Local properties of plane algebraic curves

Geometry of moduli spaces

Enumerative Real Algebraic Geometry

Linear Algebra, Theory And Applications. Kenneth Kuttler

Theorem 6.1 The addition defined above makes the points of E into an abelian group with O as the identity element. Proof. Let s assume that K is

arxiv: v1 [math.nt] 15 Sep 2009

DISCRIMINANTS, SYMMETRIZED GRAPH MONOMIALS, AND SUMS OF SQUARES

HYPERBOLICITY CONES AND IMAGINARY PROJECTIONS

Linear Algebra, Theory And Applications. Kenneth Kuttler

The Pythagoras numbers of projective varieties.

Discriminants, resultants, and their tropicalization

Euclidean projective geometry: reciprocal polar varieties and focal loci

Semidefinite representation of convex hulls of rational varieties

OSCULATION FOR CONIC FIBRATIONS

Transcription:

QUARTIC SPECTRAHEDRA Bernd Sturmfels UC Berkeley and MPI Bonn Joint work with John Christian Ottem, Kristian Ranestad and Cynthia Vinzant 1 / 20

Definition A spectrahedron of degree n in R 3 is a convex body of the form S = { (x, y, z) R 3 : xa + yb + zc + D is positive semidefinite } where A, B, C, D are real symmetric matrices of format n n. n = 1: S is a closed half space. n = 2: S is a quadric cone. Finite intersections of spectrahedra are spectrahedra. 2 / 20

Toeplitz Spectrahedron S = { (x, y, z) R 3 : 1 x y z x 1 x y y x 1 x }. 0 z y x 1 is the convex hull of the trigonometric curve {( cos(θ), cos(2θ), cos (3θ) : θ [0, π] }. is the intersection of two quadratic cones. 3 / 20

The Pillow... is a quartic spectrahedron: Q: Is it reducible? 1 x 0 x x 1 y 0 0 y 1 z x 0 z 1 Q: Is the convex dual of a spectrahedron is a spectrahedron? 4 / 20

A familiar picture The set of 3 3-correlation matrices is the elliptope 1 x y x 1 z y z 1 Q: Does every cubic spectrahedron have four nodes? 5 / 20

Symmetroids Fix projective space P 3 with coordinates (x : y : z : w). A symmetroid S of degree n is a surface with equation det(xa + yb + zc + wd) = 0 where A, B, C, D are (complex) symmetric n n matrices. Proposition If A, B, C, D are generic then the singular locus of the symmetroid S consists of ( ) n+1 3 nodes. The web xa+yb+zc+wd contains no matrix of rank n 3. Those of rank n 2 are precisely the nodes. S is transversal symmetroid if this holds. A 3-dimensional spectrahedron is transversal if its algebraic boundary is a transversal symmetroid. Problem Study the geometry and combinatorics of transversal spectrahedra. For instance, how many of the ( ) n+1 3 nodes can lie on its boundary? 6 / 20

Cubic Spectrahedra Proposition Every non-empty irreducible 3-dimensional cubic spectrahedron S has either 4 or 2 nodes, and these are connected by edges on S. The cubic surface S in P 3 is a Cayley symmetroid. For an algebraic geometer, this is obtained by blowing up P 2 at six points that are the intersections of four lines defined over R. The two cases correspond to either four real lines, or two real lines and a pair of complex conjugate lines. protrusion_kl.jpg 440 448 pixels 9/25/13 5:48 AM 7 / 20

Cubic Spectrahedra Two combinatorial types for n = 3. How about n = 4? 8 / 20

Degtyarev-Itenberg Theorem For quartic spectrahedra, this question was answered in a 2010 paper by Alex Degtyarev and Ilia Itenberg: Theorem There exists a transversal quartic spectrahedron with β nodes in its boundary and σ real nodes in its symmetroid if and only if 0 β σ, 2 σ 10, and both β and σ are even. Their proof is extremely indirect: it rests on the Global Torelli Theorem for K3 surfaces, and on deep topological results of Kharlamov and Nikhulin for moduli spaces of real K3 surfaces. It is impossible to use this to construct matrices A, B, C, D. We give a new proof that is direct, computational and geometric, via projections onto P 2 from rank 2 nodes on S. 9 / 20

Twenty Types A B C D 3 4 1 4 11 0 2 2 17 3 2 9 9 3 9 3 (2, 2): 4 14 6 10 0 6 1 4 3 6 4 1 3 10 6 7 1 6 9 2 2 1 6 2 2 4 13 10 9 6 18 3 4 10 2 8 2 4 2 4 9 1 10 17 3 7 3 5 18 3 9 6 17 10 4 3 8 6 10 10 8 4 8 0 (4, 4): 3 5 1 3 10 14 1 3 6 18 6 15 4 10 4 0 9 1 13 7 4 1 5 4 10 6 14 9 8 4 8 0 6 3 7 6 3 3 4 6 10 15 9 22 0 0 0 0 10 8 2 6 11 6 10 9 6 2 6 5 8 6 2 2 (6, 6): 8 14 0 2 6 10 5 5 2 9 2 0 6 9 9 6 2 0 5 7 10 5 14 11 6 2 6 5 2 9 13 12 6 2 7 11 9 5 11 9 5 0 5 5 2 6 12 13 5 3 3 4 19 10 12 17 5 1 3 3 1 1 0 2 (8, 8): 3 6 3 2 10 14 10 7 1 5 7 1 1 1 0 2 3 3 6 4 12 10 10 11 3 7 22 7 0 0 4 4 4 2 4 4 17 7 11 17 3 1 7 10 2 2 4 8 18 6 6 6 4 6 6 4 1 0 3 0 9 3 0 0 (10, 10): 6 2 2 6 13 9 8 0 4 0 6 3 10 9 6 2 2 2 6 9 9 6 3 0 9 0 0 9 9 6 6 2 2 4 4 8 6 5 0 6 0 9 0 6 6 4 20 6 14 4 54 27 16 12 42 8 9 3 0 9 3 3 (2, 0): 6 18 3 12 27 18 2 15 8 10 5 11 9 9 6 6 14 3 17 2 16 2 20 10 9 5 29 7 3 6 3 3 4 12 2 8 12 15 10 21 3 11 7 29 3 6 3 3 9 4 1 1 6 1 3 4 8 2 6 4 4 4 2 2 (4, 2): 4 5 3 2 1 5 5 2 2 5 1 3 4 0 0 2 1 3 3 1 3 5 6 2 6 1 6 2 2 0 0 1 1 2 1 1 4 2 2 8 4 3 2 3 2 2 1 1... etc... etc... 10 / 20

The View from a Node Figure: A quartic spectrahedron and its projection from an outside node. The ramification curve consists of two cubics totally tangent to a conic. 11 / 20

The View from a Node Our main main technical tool is a result that goes back to Cayley. Theorem Let p be a node on an irreducible quartic surface S P 3. The following are equivalent: S is a symmetroid and p corresponds to a rank 2 matrix. The projection of S from p to P 2 is branched along two cubics C 1 and C 2 that are totally tangent to a common conic Q. The pair (p, S) is real if and only if the pair (C 1 C 2, Q) is real. If this holds, then C 1 and C 2 are both real if and only if p is not on the spectrahedron. Equivalently, p is a node on the spectrahedron if and only if the cubic curves C 1 and C 2 are complex conjugates. Proposition A symmetroid with nonempty spectahedron has no rank 3 nodes. 12 / 20

The Kummer Surface... is an irreducible quartic with 16 nodes: Q: Are the convex bodies seen here spectrahedra? 13 / 20

Kummer Spectrahedron The following symmetric matrix is found on page 143 of Arthur Coble s 1929 book Algebraic Geometry and Theta Functions. a 6 w 3a 5 w 3a 4 w + z a 3 w y 3wa 5 9a 4 w 2z 9a 3 w + y 3a 2 w + x 3a 4 w + z 9a 3 w + y 9a 2 w 2x 3a 1 w a 3 w y 3a 2 w + x 3a 1 w a 0 w It is associated to polynomials of degree 6 in one variable t: p(t) = a 0 6a 1 t + 15a 2 t 2 20a 3 t 3 + 15a 4 t 4 6a 5 t 5 + a 6 t 6 These symmetroids are the Kummer surfaces. They have 16 nodes, ten of rank 2 and six of rank 3. If a 0,..., a 6 Q then all 16 nodes have rational coordinates. The spectrahedron is non-empty if and only if p(t) 0 on R. Its points are the sum of squares representations of p(t). 14 / 20

Transversal Spectrahedra This quartic spectrahedron has β = σ = 10 nodes: 18x+4y +z+9w 6x 6y 3w 6x+6y 3z 6x+4y 6x 6y 3w 2x+13y +4z+10w 2x 9y +9w 2x 8y +6z 6w 6x+6y 3z 2x 9y +9w 2x+9y +9z+9w 2x+6y 6w 6x+4y 2x 8y +6z 6w 2x+6y 6w 4x+5y +9z+4w This quartic spectrahedron has β = 0 and σ = 10: 263x 160y 20z 187w 3x 132y+28z+78w 114x 30y+4z 76w 103x+244y+32z 192w 3x 132y+28z+78w 45x+28y 32z 32w 35x+40y 32z+24w 48x+20y 4z+88w 114x 30y+4z 76w 35x+40y 32z+24w 275x+25y+96z+80w 55x 40y 156z 192w 103x+244y+32z 192w 48x+20y 4z+88w 55x 40y 156z 192w 278x 132y+180z 80w To get bounded convex bodies, set x+y+z+w = 1. Q: What will semidefinite programming do on these matrices? What is the rank of the optimal matrix for random cost functions? 15 / 20

Transversal Spectrahedra Out[124]= (β, σ) = (2, 2), (8, 10), (0, 10) 16 / 20

Spectrahedral Shadows (β, σ) = (10, 10) Curve of degree 12 with 46 nodes. Problem: Which compact convex semialgebraic subsets of the plane R 2 are projections of 3-dimensional spectrahedra? 17 / 20

From Polytope to Spectrahedron Toblerone-of-Switzerland.jpg 6,142 2,457 pixels 9/26/13 6:35 AM 18 / 20

Sylvester Spectrahedron Given general linear forms l 0, l 1, l 2, l 3, l 4 in R[x, y, z, w], consider l 1 + l 0 l 0 l 0 l 0 l 0 l 2 + l 0 l 0 l 0 l 0 l 0 l 3 + l 0 l 0 l 0 l 0 l 0 l 4 + l 0 The inequalities l i 0 define the cone over a triangular prism. Sanyal [2013]: the spectrahedron is the Renegar derivative of. The header refers to Sylvester s Pentahedral Theorem: every cubic surface in P 3 admits a unique representation f = l 3 1 + l 3 2 + l 3 3 + l 3 4 + l 3 5 as a sum of five powers of linear forms. Our matrix is the Hessian [ 2 f / x i x j ] in suitable coordinates. The Sylvester symmetroid is transversal. Q: What are the 10 nodes? What are β and σ? 19 / 20

Conclusion Happy Birthday, Jean Bernard. Here is your picture postcard. Out[124]= 20 / 20