Multiplicative Approximations for Polynomial Optimization Over the Unit Sphere

We consider the following basic problem: given an n-variate degree-d homogeneous polynomial f with real coefficients, compute a unit vector x ∈ Rn that maximizes | f (x)|. Besides its fundamental nature, this problem arises in many diverse contexts ranging from tensor and operator norms to graph expansion to quantum information theory. The homogeneous degree 2 case is efficiently solvable as it corresponds to computing the spectral norm of an associated matrix, but the higher degree case is NP-hard. In this work, we give multiplicative approximation algorithms for this problem. Our algorithms leverage the tractability of the degree 2 case, and output the best solution among a carefully constructed set of quadratic polynomials. They offer a trade-off between the approximation ratio and running time, which is governed by the number of quadratic problems we search over. Specifically, in nO(q) time, we get an approximation within factor Od((n/q) d/2−1) for arbitrary polynomials, and Od((n/q) d/4−1/2) for polynomials with non-negative coefficients. The approximation guarantees are with respect to the optimum of the level-q SoS SDP relaxation of the problem, which the algorithm rounds to a unit vector. We also consider the case when f is random with independent ±1 coefficients, and prove that w.h.p the level-q SoS solution gives a certificate within factor Õd((n/q) d/4−1/2) of the optimum. We complement our algorithmic results with some polynomially large integrality gaps for d-levels of the SoS relaxation. For the random polynomial case, we show a gap of Ωd(n d/4−1/2), which precisely matches the exponent of our upper bound, and shows the necessity of our Ω(d) exponent in the approximation ratio for general polynomials. For polynomials with nonnegative coefficients, we show an Ω̃(n1/12) gap for the d = 4 case. To obtain our results, we develop general techniques which help analyze the approximation obtained by higher levels of the SoS hierarchy. We believe these techniques will also be useful in understanding polynomial optimization for other constrained settings. ∗Supported by NSF CCF-1422045 and CCF-1526092. vpb@cs.cmu.edu †Supported by NSF CCF-1254044 mghosh@ttic.edu ‡Supported in part by NSF grant CCF-1526092. guruswami@cmu.edu §Supported by a Samsung Fellowship, Simons award, and NSF CCF-1526092. euiwoonl@cs.cmu.edu ¶Supported by NSF CCF-1254044 madhurt@ttic.edu

[1]  Van H. Vu,et al.  A Large Deviation Result on the Number of Small Subgraphs of a Random Graph , 2001, Combinatorics, Probability and Computing.

[2]  Pravesh Kothari,et al.  A Nearly Tight Sum-of-Squares Lower Bound for the Planted Clique Problem , 2016, 2016 IEEE 57th Annual Symposium on Foundations of Computer Science (FOCS).

[3]  Venkatesan Guruswami,et al.  Certifying Random Polynomials over the Unit Sphere via Sum of Squares Hierarchy , 2016, ArXiv.

[4]  J. Lasserre.  Moments, Positive Polynomials And Their Applications , 2009 .

[5]  Tengyu Ma,et al.  Decomposing Overcomplete 3rd Order Tensors using Sum-of-Squares Algorithms , 2015, APPROX-RANDOM.

[6]  Siu On Chan,et al.  Approximation resistance from pairwise independent subgroups , 2013, STOC '13.

[7]  Yuan Zhou,et al.  Approximability and proof complexity , 2012, SODA.

[8]  Alan M. Frieze,et al.  A new approach to the planted clique problem , 2008, FSTTCS.

[9]  Shuzhong Zhang,et al.  Approximation algorithms for homogeneous polynomial optimization with quadratic constraints , 2010, Math. Program..

[10]  Lars Engebretsen,et al.  Clique Is Hard To Approximate Within , 2000 .

[11]  Santosh S. Vempala,et al.  Random Tensors and Planted Cliques , 2009, APPROX-RANDOM.

[12]  Yufei Zhao,et al.  On the variational problem for upper tails in sparse random graphs , 2014, Random Struct. Algorithms.

[13]  B. Simon.  The Classical Moment Problem as a Self-Adjoint Finite Difference Operator , 1998, math-ph/9906008.

[14]  Jeff Kahn,et al.  Upper tails for triangles , 2010, Random Struct. Algorithms.

[15]  Ryota Tomioka,et al.  Spectral norm of random tensors , 2014, 1407.1870.

[16]  Stephanie Wehner,et al.  Convergence of SDP hierarchies for polynomial optimization on the hypersphere , 2012, ArXiv.

[17]  David Steurer,et al.  Sum-of-squares proofs and the quest toward optimal algorithms , 2014, Electron. Colloquium Comput. Complex..

[18]  Avi Wigderson,et al.  Sum-of-squares Lower Bounds for Planted Clique , 2015, STOC.

[19]  Yuan Zhou,et al.  Hypercontractivity, sum-of-squares proofs, and their applications , 2012, STOC '12.

[20]  Prasad Raghavendra,et al.  Strongly refuting random CSPs below the spectral threshold , 2016, STOC.

[21]  Andrea Montanari,et al.  A statistical model for tensor PCA , 2014, NIPS.

[22]  David Steurer,et al.  Rounding sum-of-squares relaxations , 2013, Electron. Colloquium Comput. Complex..

[23]  Y. Nesterov.  Random walk in a simplex and quadratic optimization over convex polytopes , 2003 .

[24]  Subhash Khot,et al.  Linear Equations Modulo 2 and the L1 Diameter of Convex Bodies , 2008, 48th Annual IEEE Symposium on Foundations of Computer Science (FOCS'07).

[25]  Miklós Simonovits,et al.  Deterministic and randomized polynomial-time approximation of radii , 2001 .

[26]  Prasad Raghavendra,et al.  On the Integrality Gap of Degree-4 Sum of Squares for Planted Clique , 2016, SODA.

[27]  Anthony Man-Cho So,et al.  Deterministic approximation algorithms for sphere constrained homogeneous polynomial optimization problems , 2011, Math. Program..

[28]  Etienne de Klerk,et al.  The complexity of optimizing over a simplex, hypercube or sphere: a short survey , 2008, Central Eur. J. Oper. Res..

[29]  M. Laurent.  Sums of Squares, Moment Matrices and Optimization Over Polynomials , 2009 .

[30]  David Steurer,et al.  Dictionary Learning and Tensor Decomposition via the Sum-of-Squares Method , 2014, STOC.

[31]  Van H. Vu,et al.  Divide and conquer martingales and the number of triangles in a random graph , 2004, Random Struct. Algorithms.

[32]  Jonathan Shi,et al.  Tensor principal component analysis via sum-of-square proofs , 2015, COLT.

[33]  Pravesh Kothari,et al.  Sum of Squares Lower Bounds from Pairwise Independence , 2015, STOC.

[34]  Aram Wettroth Harrow,et al.  Quantum de Finetti Theorems Under Local Measurements with Applications , 2012, Communications in Mathematical Physics.

[35]  Pablo A. Parrilo,et al.  A PTAS for the minimization of polynomials of fixed degree over the simplex , 2006, Theor. Comput. Sci..

[36]  R. Varga,et al.  Block diagonally dominant matrices and generalizations of the Gerschgorin circle theorem , 1962 .

[37]  Jeff Kahn,et al.  Tight upper tail bounds for cliques , 2011, Random Struct. Algorithms.

[38]  Grant Schoenebeck,et al.  Linear Level Lasserre Lower Bounds for Certain k-CSPs , 2008, 2008 49th Annual IEEE Symposium on Foundations of Computer Science.

[39]  Etienne de Klerk,et al.  Convergence analysis for Lasserre’s measure-based hierarchy of upper bounds for polynomial optimization , 2017, Math. Program..

[40]  Srinivasan Venkatesh,et al.  On the advantage over a random assignment , 2004, Random Struct. Algorithms.

[41]  S. Janson,et al.  Upper tails for subgraph counts in random graphs , 2004 .

[42]  Sourav Chatterjee,et al.  The missing log in large deviations for triangle counts , 2010, Random Struct. Algorithms.

[43]  Madhur Tulsiani.  CSP gaps and reductions in the lasserre hierarchy , 2009, STOC '09.