Geometric analysis of the condition of the convex feasibility problem

The focus of this paper is the homogeneous convex feasibility problem, which is the following question: Given an m-dimensional subspace of R, does this subspace intersect a fixed convex cone solely in the origin or are there further intersection points? This problem includes as special cases the linear, the second order, and the semidefinite feasibility problems, where one simply chooses the positive orthant, a product of Lorentz cones, or the cone of positive semidefinite matrices, respectively. An important role for the running time of algorithms solving the convex feasibility problem is played by Renegar’s condition number. The (inverse of the) condition of an input is basically the magnitude of the smallest perturbation, which changes the status of the input, i.e., which makes a feasible input infeasible, or the other way round. We will give an average analysis of this condition for several classes of convex cones, and one that is independent of the underlying convex cone. We will also describe a way of deriving smoothed analyses from our approach. We will achieve these results by adopting a purely geometric viewpoint leading to computations in the Grassmann manifold. Besides these main results about the random behavior of the condition of the convex feasibility problem, we will obtain a couple of byproduct results in the domain of spherical convex geometry. Kurzbeschreibung Den Mittelpunkt dieser Arbeit bildet das homogene konvexe Losbarkeitsproblem, welches die folgende Frage ist: Gegeben sei ein m-dimensionaler Unterraum des R; schneidet dieser Unterraum einen gegebenen konvexen Kegel nur im Ursprung, oder gibt es weitere Schnittpunkte? Dieses Problem umfasst als Spezialfalle das lineare, das quadratische, und das semidefinite Losbarkeitsproblem, wobei man als konvexen Kegel den positiven Orthanten, ein Produkt von Lorentzkegeln, bzw. den Kegel der positiv semidefiniten Matrizen wahlt. Fur die Laufzeit von Algorithmen, welche das konvexe Losbarkeitsproblem losen, spielt die Renegarsche Konditionszahl eine wichtige Rolle. Die Kondition einer Eingabe, bzw. ihr Inverses, ist gegeben durch die Grose einer kleinsten Storung, welche den Status der Eingabe von ‘feasible’ zu ‘infeasible’ bzw. von ‘infeasible’ zu ‘feasible’ andert. Wir werden eine Durchschnittsanalyse dieser Kondition fur verschiedene Klassen von konvexen Kegeln angeben, sowie eine, welche unabhangig ist von der Wahl des zugrunde gelegten konvexen Kegels. Wir werden desweiteren einen Weg beschreiben, auf dem auch geglattete Analysen mittels unseres Ansatzes erzielt werden konnen. Wir erreichen diese Ergebnisse mit Hilfe einer rein geometrischen Sichtweise, welche zu Berechnungen in der Grassmann-Mannigfaltigkeit fuhrt. Neben diesen Hauptergebnissen uber das zufallige Verhalten der Kondition des konvexen Losbarkeitsproblems werden wir auch einige Nebenergebnisse im Bereich der spharischen Konvexgeometrie erzielen.

[1]  C. Jordan Essai sur la géométrie à $n$ dimensions , 1875 .

[2]  I. Holopainen Riemannian Geometry , 1927, Nature.

[3]  T. Bonnesen,et al.  Theorie der Konvexen Körper , 1934 .

[4]  H. Weyl On the Volume of Tubes , 1939 .

[5]  J. G. Wendel A Problem in Geometric Probability. , 1962 .

[6]  Milton Abramowitz,et al.  Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables , 1964 .

[7]  L. Nachbin,et al.  The Haar integral , 1965 .

[8]  J. Munkres,et al.  Calculus on Manifolds , 1965 .

[9]  H. Fédérer Geometric Measure Theory , 1969 .

[10]  H. Schubert,et al.  O. D. Kellogg, Foundations of Potential Theory. (Die Grundlehren der mathematischen Wissenschaften in Einzeldarstellungen, Band 31). X + 384 S. m. 30 Fig. Berlin/Heidelberg/New York 1967. Springer‐Verlag. Preis geb. DM 32,– , 1969 .

[11]  R. Tyrrell Rockafellar,et al.  Convex Analysis , 1970, Princeton Landmarks in Mathematics and Physics.

[12]  W. J. Firey An integral-geometric meaning for lower order area functions of convex bodies , 1972 .

[13]  W. Boothby An introduction to differentiable manifolds and Riemannian geometry , 1975 .

[14]  M. Spivak A comprehensive introduction to differential geometry , 1979 .

[15]  S. Helgason Differential Geometry, Lie Groups, and Symmetric Spaces , 1978 .

[16]  J. Thorpe Elementary Topics in Differential Geometry , 1979 .

[17]  R. Muirhead Aspects of Multivariate Statistical Theory , 1982, Wiley Series in Probability and Statistics.

[18]  Gene H. Golub,et al.  Matrix computations , 1983 .

[19]  Donald St. P. Richards,et al.  Hypergeometric Functions of Scalar Matrix Argument are Expressible in Terms of Classical Hypergeometric Functions , 1985 .

[20]  A. Edelman Eigenvalues and condition numbers of random matrices , 1988 .

[21]  M. Abramowitz,et al.  Handbook of Mathematical Functions With Formulas, Graphs and Mathematical Tables (National Bureau of Standards Applied Mathematics Series No. 55) , 1965 .

[22]  E. Teufel Kinematische Berührformeln in Räumen konstanter Krümmung , 1988 .

[23]  R. Stanley Log‐Concave and Unimodal Sequences in Algebra, Combinatorics, and Geometry a , 1989 .

[24]  F. Brenti,et al.  Unimodal, log-concave and Pólya frequency sequences in combinatorics , 1989 .

[25]  Kinematische Berührformeln in Riemannschen homogenen Räumen , 1989 .

[26]  R. Schneider,et al.  CHAPTER 5.1 – Integral Geometry , 1993 .

[27]  A simple proof of an approximation theorem of H. Minkowski , 1993 .

[28]  J. Renegar Some perturbation theory for linear programming , 1994, Math. Program..

[29]  K. Ball CONVEX BODIES: THE BRUNN–MINKOWSKI THEORY , 1994 .

[30]  James Renegar,et al.  Incorporating Condition Measures into the Complexity Theory of Linear Programming , 1995, SIAM J. Optim..

[31]  James Renegar,et al.  Linear programming, complexity theory and elementary functional analysis , 1995, Math. Program..

[32]  Daniel A. Klain,et al.  Introduction to Geometric Probability , 1997 .

[33]  L. Trefethen,et al.  Numerical linear algebra , 1997 .

[34]  Steve Smale,et al.  Complexity theory and numerical analysis , 1997, Acta Numerica.

[35]  Alan Edelman,et al.  The Geometry of Algorithms with Orthogonality Constraints , 1998, SIAM J. Matrix Anal. Appl..

[36]  Lenore Blum,et al.  Complexity and Real Computation , 1997, Springer New York.

[37]  G. Pataki The Geometry of Semidefinite Programming , 2000 .

[38]  Javier Peña,et al.  Understanding the Geometry of Infeasible Perturbations of a Conic Linear System , 1999, SIAM J. Optim..

[39]  James Renegar,et al.  A mathematical view of interior-point methods in convex optimization , 2001, MPS-SIAM series on optimization.

[40]  Alexander Barvinok,et al.  A course in convexity , 2002, Graduate studies in mathematics.

[41]  Fuchang Gao,et al.  Intrinsic volumes and polar sets in spherical space , 2002 .

[42]  P. Absil,et al.  Riemannian Geometry of Grassmann Manifolds with a View on Algorithmic Computation , 2004 .

[43]  Zizhong Chen,et al.  Condition Numbers of Gaussian Random Matrices , 2005, SIAM J. Matrix Anal. Appl..

[44]  Stephen P. Boyd,et al.  Convex Optimization , 2004, Algorithms and Theory of Computation Handbook.

[45]  P. Koev,et al.  On the largest principal angle between random subspaces , 2006 .

[46]  F. Cucker,et al.  Smoothed analysis of complex conic condition numbers , 2006, math/0605635.

[47]  Felipe Cucker,et al.  General formulas for the smoothed analysis of condition numbers , 2006 .

[48]  Riemannian Geometry: Riemannian Volume , 2006 .

[49]  S. Ole Warnaar,et al.  The importance of the Selberg integral , 2007, 0710.3981.

[50]  Javier Peña,et al.  A primal-dual symmetric relaxation for homogeneous conic systems , 2007, J. Complex..

[51]  Terence Tao,et al.  The condition number of a randomly perturbed matrix , 2007, STOC '07.

[52]  Gregorio Malajovich,et al.  On the number of minima of a random polynomial , 2008, J. Complex..

[53]  Felipe Cucker,et al.  The probability that a slightly perturbed numerical analysis problem is difficult , 2006, Math. Comput..

[54]  採編典藏組 Society for Industrial and Applied Mathematics(SIAM) , 2008 .

[55]  Robert M. Freund,et al.  A geometric analysis of Renegar’s condition number, and its interplay with conic curvature , 2007, Math. Program..

[56]  Bernd Sturmfels,et al.  The algebraic degree of semidefinite programming , 2010, Math. Program..

[57]  F. Cucker,et al.  Coverage processes on spheres and condition numbers for linear programming , 2007, 0712.2816.

[58]  Peter Bürgisser,et al.  Smoothed Analysis of Condition Numbers , 2011 .

[59]  Felipe Cucker,et al.  Adversarial smoothed analysis , 2010, J. Complex..

[60]  Felipe Cucker,et al.  Smoothed Analysis of Moore-Penrose Inversion , 2010, SIAM J. Matrix Anal. Appl..

[61]  Peter Bürgisser,et al.  Robust smoothed analysis of a condition number for linear programming , 2012, Math. Program..