Curves of descent

DOI10.1137/130920216MaRDI QIDQ2810053zbMATH OpenOpenAlexFDO

Authors Alexander D. Ioffe, D. Drusvyatskiy, A. S. Lewis

Publication date 31 May 2016

Published in SIAM Journal on Control and Optimization (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1212.1231

zbMATH Keywords

descent curvessemialgebraic functionsslopessubdifferentialssubgradient dynamical systems

Mathematics Subject Classification ID

Real-analytic and semi-analytic sets (14P15) Inequalities involving derivatives and differential and integral operators (26D10) Semi-analytic sets, subanalytic sets, and generalizations (32B20) Gradient-like behavior; isolated (locally maximal) invariant sets; attractors, repellers for topological dynamical systems (37B35) Nonsmooth analysis (49J52)

Abstract: Steepest descent is central in variational mathematics. We present a new transparent existence proof for curves of near-maximal slope --- an influential notion of steepest descent in a nonsmooth setting. We moreover show that for semi-algebraic functions --- prototypical nonpathological functions in nonsmooth optimization --- such curves are precisely the solutions of subgradient dynamical systems.

Review text

This paper is dedicated to unveiling new facts about the notion of steepest descent, some of which quite surprising. An alternate strategy for constructing near-steepest descent curves is proposed, exploiting a discretization of the range of the involved function and then building a piecewise linear curve by projecting iterates onto successive sublevel sets. Connections to the theory of metric regularity are mentioned, too. The existence of curves of near-steepest descent is proven under reasonable hypotheses in the third section of the paper, while in the next one relations between curves of near-steepest descent and solutions of subgradient dynamical systems in Euclidean spaces are presented or recalled, respectively, by means of variational analysis. In the last section, descent curves for semialgebraic functions are considered and it is shown that such functions always admit nontrivial descent curves emanating from any nonminimizing point.

Recommendations

• When Does Water Find the Shortest Path Downhill? The Geometry of Steepest Descent Curves
• Plane R-curves and their steepest descent properties I
• scientific article; zbMATH DE number 4150771
• Steepest descent with curvature dynamical system
• Maximum length of steepest descent curves for quasi-convex functions

Cited in

(35)

• A stochastic subgradient method for distributionally robust non-convex and non-smooth learning
• Nonsmooth optimization using Taylor-like models: error bounds, convergence, and termination criteria
• Convergence of a stochastic subgradient method with averaging for nonsmooth nonconvex constrained optimization
• Determination of convex functions via subgradients of minimal norm
• Stochastic subgradient method converges on tame functions
• Perturbation of error bounds
• Conservative and semismooth derivatives are equivalent for semialgebraic maps
• scientific article; zbMATH DE number 6385415 (Why is no real title available?)
• scientific article; zbMATH DE number 3884146 (Why is no real title available?)
• scientific article; zbMATH DE number 4150771 (Why is no real title available?)
• scientific article; zbMATH DE number 1354288 (Why is no real title available?)
• When Does Water Find the Shortest Path Downhill? The Geometry of Steepest Descent Curves
• A Stochastic Subgradient Method for Nonsmooth Nonconvex Multilevel Composition Optimization
• Spectral operators of matrices: semismoothness and characterizations of the generalized Jacobian
• A determination theorem in terms of the metric slope
• Global convergence of the gradient method for functions definable in o-minimal structures
• Lyapunov stability of the subgradient method with constant step size
• Certifying the Absence of Spurious Local Minima at Infinity
• Mini-Batch Risk Forms
• Global stability of first-order methods for coercive tame functions
• Descent modulus and applications
• A functional model method for nonconvex nonsmooth conditional stochastic optimization
• Metric compatibility and determination in complete metric spaces
• Steepest geometric descent for regularized quasiconvex functions
• Error bounds, PL condition, and quadratic growth for weakly convex functions, and linear convergences of proximal point methods
• Subdifferential determination of a primal lower regular function on a Banach space
• Fast convergence to non-isolated minima: four equivalent conditions for \({\mathrm{C}^2}\) functions
• A slope generalization of Attouch theorem
• The complexity of first-order optimization methods from a metric perspective
• Identifiability, the KL property in metric spaces, and subgradient curves
• Slopes and the Moreau-Rockafellar theorem
• The gradient's limit of a definable family of functions admits a variational stratification
• Descent: an optimization point of view on different fields
• Transversality and alternating projections for nonconvex sets
• Error bounds and Hölder metric subregularity

This page was built for publication: Curves of descent

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2810053)