DOCSLIB.ORG

15. Quasi-Newton Methods

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
15. Quasi-Newton Methods L. Vandenberghe ECE236C (Spring 2020) 15. Quasi-Newton methods variable metric methods • quasi-Newton methods • BFGS update • limited-memory quasi-Newton methods • 15.1 Newton method for unconstrained minimization minimize f x ¹ º f convex, twice continously differentiable Newton method 2 1 x = x t f x − f x k+1 k − kr ¹ kº r ¹ kº advantages: fast convergence, robustness, affine invariance • disadvantages: requires second derivatives and solution of linear equation • can be too expensive for large scale applications Quasi-Newton methods 15.2 Variable metric methods 1 x = x t H− f x k+1 k − k k r ¹ kº the positive definite matrix Hk is an approximatian of the Hessian at xk, chosen to: avoid calculation of second derivatives • simplify computation of search direction • ‘Variable metric’ interpretation (236B, lecture 10, page 11) 1 ∆x = H− f x − r ¹ º is the steepest descent direction at x for the quadratic norm 1 2 z = zT Hz / k kH Quasi-Newton methods 15.3 Quasi-Newton methods given: starting point x dom f , H 0 0 2 0 for k = 0;1;::: 1. compute quasi-Newton direction ∆x = H 1 f x k − k− r ¹ kº 2. determine step size tk (e.g., by backtracking line search) 3. compute xk+1 = xk + tk∆xk 4. compute Hk+1 different update rules exist for H in step 4 • k+1 can also propagate H 1 or a factorization of H to simplify calculation of ∆x • k− k k Quasi-Newton methods 15.4 Broyden–Fletcher–Goldfarb–Shanno (BFGS) update BFGS update yyT H ssT H H = H + k k k+1 k T T y s − s Hk s where s = x x ; y = f x f x k+1 − k r ¹ k+1º − r ¹ kº Inverse update T T T 1 sy 1 ys ss H− = I H− I + k+1 − yT s k − yT s yT s note that yT s > 0 for strictly convex f ; see page 1.8 • cost of update or inverse update is O n2 operations • ¹ º Quasi-Newton methods 15.5 Positive definiteness if yT s > 0, BFGS update preserves positive definitess of H • k this ensures that ∆x = H 1 f x is a descent direction • − k− r ¹ kº Proof: from inverse update formula, T T T T 2 T 1 s v 1 s v s v v H− v = v y H− v y + ¹ º k+1 − sT y k − sT y yT s if H 0, both terms are nonnegative for all v • k second term is zero only if sT v = 0; then first term is zero only if v = 0 • Quasi-Newton methods 15.6 Secant condition the BFGS update satisfies the secant condition Hk+1s = y where s = x x and y = f x f x k+1 − k r ¹ k+1º − r ¹ kº Interpretation: we define a quadratic approximation of f around xk+1 T 1 T f˜ x = f x + + f x + x x + + x x + H + x x + ¹ º ¹ k 1º r ¹ k 1º ¹ − k 1º 2¹ − k 1º k 1¹ − k 1º by construction f˜ x = f x • r ¹ k+1º r ¹ k+1º secant condition implies that also f˜ x = f x : • r ¹ kº r ¹ kº f˜ x = f x + H x x r ¹ kº r ¹ k+1º k+1¹ k − k+1º = f x r ¹ kº Quasi-Newton methods 15.7 Secant method for f : R R, BFGS with unit step size gives the secant method ! f 0 xk f 0 xk f 0 xk 1 xk+1 = xk ¹ º; Hk = ¹ º − ¹ − º − Hk xk xk 1 − − xk 1 xk xk+1 − f˜ x 0( ) f x 0( ) Quasi-Newton methods 15.8 Convergence Global result if f is strongly convex, BFGS with backtracking line search (EE236B, lecture 10-6) converges from any x , H 0 0 0 Local convergence if f is strongly convex and 2 f x is Lipschitz continuous, local convergence is r ¹ º superlinear: for sufficiently large k, x x? c x x? k k+1 − k2 ≤ k k k − k2 where c 0 k ! (cf., quadratic local convergence of Newton method) Quasi-Newton methods 15.9 Example m T X T minimize c x log bi ai x − i=1 ¹ − º n = 100, m = 500 Newton BFGS 103 103 100 100 ? ? f 3 f 3 10− 10− ) − ) − k k x 6 x 6 ( 10− ( 10− f f 9 9 10− 10− 10 12 10 12 − 0 2 4 6 8 10 12 − 0 50 100 150 k k cost per Newton iteration: O n3 plus computing 2 f x • ¹ º r ¹ º cost per BFGS iteration: O n2 • ¹ º Quasi-Newton methods 15.10 Square root BFGS update T to improve numerical stability, propagate Hk in factored form Hk = Lk Lk if H = L LT then H = L LT with • k k k k+1 k+1 k+1 αy˜ s˜ s˜T Lk+1 = Lk I + ¹ − º ; s˜T s˜ where T 1 2 1 T s˜ s˜ / y˜ = L− y; s˜ = L s; α = k k yT s if L is triangular, cost of reducing L to triangular form is O n2 • k k+1 ¹ º Quasi-Newton methods 15.11 Optimality of BFGS update X = Hk+1 solves the convex optimization problem minimize tr H 1X log det H 1X n ¹ k− º − ¹ k− º − subject to Xs = y cost function is nonnegative, equal to zero only if X = H • k also known as relative entropy between densities N 0; X , N 0; H • ¹ º ¹ kº BFGS update is a least-change secant update • optimality result follows from KKT conditions: X = Hk+1 satisfies 1 1 1 T T X− = H− sν + νs ; Xs = y; X 0 k − 2¹ º with yT 1y! ! 1 1 Hk− ν = 2H− y 1 + s sT y k − yT s Quasi-Newton methods 15.12 Davidon–Fletcher–Powell (DFP) update switch Hk and X in objective on previous page minimize tr H X 1 log det H X 1 n ¹ k − º − ¹ k − º − subject to Xs = y minimize relative entropy between N 0; H and N 0; X • ¹ kº ¹ º problem is convex in X 1 (with constraint written as s = X 1y) • − − solution is ‘dual’ of BFGS formula • ysT syT yyT Hk+1 = I Hk I + − sT y − sT y sT y (known as DFP update) predates BFGS update, but is less often used Quasi-Newton methods 15.13 Limited memory quasi-Newton methods 1 main disadvantage of quasi-Newton method is need to store Hk, Hk− , or Lk Limited-memory BFGS 1 (L-BFGS): do not store Hk− explicitly instead we store up to m (e.g., m = 30) values of • s = x x ; y = f x f x j j+1 − j j r ¹ j+1º − r ¹ jº we evaluate ∆x = H 1 f x recursively, using • k k− r ¹ kº T ! T ! T sj y yj s sj s 1 j 1 j j H− = I H− I + j+1 − T j − T T yj sj yj sj yj sj for j = k 1;:::; k m, assuming, for example, Hk m = I − − − an alternative is to restart after m iterations • cost per iteration is O nm , storage is O nm • ¹ º ¹ º Quasi-Newton methods 15.14 Interpretation of CG as restarted BFGS method first two iterations of BFGS (page 15.5) if H0 = I: 1 x = x t f x ; x = x t H− f x 1 0 − 0r ¹ 0º 2 1 − 1 1 r ¹ 1º where H is computed from s = x x and y = f x f x via 1 1 − 0 r ¹ 1º − r ¹ 0º T T T T 1 y y ss ys + sy H− = I + 1 + 1 ¹ sT y ºyT s − yT s if t is determined by exact line search, then f x T s = 0 • 0 r ¹ 1º quasi-Newton step in second iteration simplifies to • T 1 y f x1 H− f x1 = f x1 + r ¹ ºs − 1 r ¹ º −r ¹ º yT s this is the Hestenes–Stiefel conjugate gradient update nonlinear CG can be interpreted as L-BFGS with m = 1 Quasi-Newton methods 15.15 References J. E. Dennis, Jr., and R. B. Schabel, Numerical Methods for Unconstrained Optimization and • Nonlinear Equations (1996), chapter 9. C. T. Kelley, Iterative Methods for Optimization (1999), chapter 4. • J. Nocedal and S. J. Wright, Numerical Optimization (2006), chapter 6 and section 7.2. • Quasi-Newton methods 15.16.
Load more

Recommended publications
  • Lec9p1, ORF363/COS323
    Lec9p1, ORF363/COS323 Instructor: Amir Ali Ahmadi Fall 2014 This lecture: • Multivariate Newton's method TAs: Y. Chen, G. Hall, • Rates of convergence J. Ye • Modifications for global convergence • Nonlinear least squares ○ The Gauss-Newton algorithm • In the previous lecture, we saw the general framework of descent algorithms, with several choices for the step size and the descent direction. We also discussed convergence issues associated with these methods and provided some formal definitions for studying rates of convergence. Our focus before was on gradient descent methods and variants, which use only first order information (first order derivatives). These algorithms achieved a linear rate of convergence. • Today, we see a wonderful descent method with superlinear (in fact quadratic) rate of convergence: the Newton algorithm. This is a generalization of what we saw a couple of lectures ago in dimension one for root finding and function minimization. • The Newton's method is nothing but a descent method with a specific choice of a descent direction; one that iteratively adjusts itself to the local geometry of the function to be minimized. • In practice, Newton's method can converge with much fewer iterations than gradient methods. For example, for quadratic functions, while we saw that gradient methods can zigzag for a long time (depending on the underlying condition number), Newton's method will always get the optimal solution in a single step. • The cost that we pay for fast convergence is the need to (i) access second order information (i.e., derivatives of first and second order), and (ii) solve a linear system of equations at every step of the algorithm.
    [Show full text]
  • Use of the NLPQLP Sequential Quadratic Programming Algorithm to Solve Rotorcraft Aeromechanical Constrained Optimisation Problems
    NASA/TM—2016–216632 Use of the NLPQLP Sequential Quadratic Programming Algorithm to Solve Rotorcraft Aeromechanical Constrained Optimisation Problems Jane Anne Leyland Ames Research Center Moffett Field, California April 2016 This page is required and contains approved text that cannot be changed. NASA STI Program ... in Profile Since its founding, NASA has been dedicated • CONFERENCE PUBLICATION. to the advancement of aeronautics and Collected papers from scientific and space science. The NASA scientific and technical conferences, symposia, technical information (STI) program plays a seminars, or other meetings sponsored key part in helping NASA maintain this or co-sponsored by NASA. important role. • SPECIAL PUBLICATION. Scientific, The NASA STI program operates under the technical, or historical information from auspices of the Agency Chief Information NASA programs, projects, and Officer. It collects, organizes, provides for missions, often concerned with subjects archiving, and disseminates NASA’s STI. The having substantial public interest. NASA STI program provides access to the NTRS Registered and its public interface, • TECHNICAL TRANSLATION. the NASA Technical Reports Server, thus English-language translations of foreign providing one of the largest collections of scientific and technical material aeronautical and space science STI in the pertinent to NASA’s mission. world. Results are published in both non- NASA channels and by NASA in the NASA Specialized services also include STI Report Series, which includes the organizing and publishing research following report types: results, distributing specialized research announcements and feeds, providing • TECHNICAL PUBLICATION. Reports of information desk and personal search completed research or a major significant support, and enabling data exchange phase of research that present the results services.
    [Show full text]
  • A Sequential Quadratic Programming Algorithm with an Additional Equality Constrained Phase
    A Sequential Quadratic Programming Algorithm with an Additional Equality Constrained Phase Jos´eLuis Morales∗ Jorge Nocedal † Yuchen Wu† December 28, 2008 Abstract A sequential quadratic programming (SQP) method is presented that aims to over- come some of the drawbacks of contemporary SQP methods. It avoids the difficulties associated with indefinite quadratic programming subproblems by defining this sub- problem to be always convex. The novel feature of the approach is the addition of an equality constrained phase that promotes fast convergence and improves performance in the presence of ill conditioning. This equality constrained phase uses exact second order information and can be implemented using either a direct solve or an iterative method. The paper studies the global and local convergence properties of the new algorithm and presents a set of numerical experiments to illustrate its practical performance. 1 Introduction Sequential quadratic programming (SQP) methods are very effective techniques for solv- ing small, medium-size and certain classes of large-scale nonlinear programming problems. They are often preferable to interior-point methods when a sequence of related problems must be solved (as in branch and bound methods) and more generally, when a good estimate of the solution is available. Some SQP methods employ convex quadratic programming sub- problems for the step computation (typically using quasi-Newton Hessian approximations) while other variants define the Hessian of the SQP model using second derivative informa- tion, which can lead to nonconvex quadratic subproblems; see [27, 1] for surveys on SQP methods. ∗Departamento de Matem´aticas, Instituto Tecnol´ogico Aut´onomode M´exico, M´exico. This author was supported by Asociaci´onMexicana de Cultura AC and CONACyT-NSF grant J110.388/2006.
    [Show full text]
  • Strongly Polynomial Algorithm for the Intersection of a Line with a Polymatroid Jean Fonlupt, Alexandre Skoda
    Strongly polynomial algorithm for the intersection of a line with a polymatroid Jean Fonlupt, Alexandre Skoda To cite this version: Jean Fonlupt, Alexandre Skoda. Strongly polynomial algorithm for the intersection of a line with a polymatroid. Research Trends in Combinatorial Optimization, Springer Berlin Heidelberg, pp.69-85, 2009, 10.1007/978-3-540-76796-1_5. hal-00634187 HAL Id: hal-00634187 https://hal.archives-ouvertes.fr/hal-00634187 Submitted on 20 Oct 2011 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Strongly polynomial algorithm for the intersection of a line with a polymatroid Jean Fonlupt1 and Alexandre Skoda2 1 Equipe Combinatoire et Optimisation. CNRS et Universit´eParis6 [email protected] 2 Equipe Combinatoire et Optimisation. CNRS et Universit´eParis6 France Telecom R&D. Sophia Antipolis [email protected] Summary. We present a new algorithm for the problem of determining the inter- N section of a half-line Δu = {x ∈ IR | x = λu for λ ≥ 0} with a polymatroid. We then propose a second algorithm which generalizes the first algorithm and solves a parametric linear program. We prove that these two algorithms are strongly poly- nomial and that their running time is O(n8 + γn7)whereγ is the time for an oracle call.
    [Show full text]
  • Hybrid Algorithms for On-Line and Combinatorial Optimization Problems
    Hybrid Algorithms for On-Line Search and Combinatorial Optimization Problems. Yury V. Smirnov May 28, 1997 CMU-CS-97-171 Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy School of Computer Science Carnegie Mellon University Pittsburgh, PA 15213 Copyright c 1997 Yury V. Smirnov All rights reserved This research is sponsored in part by the National Science Foundation under grant number IRI-9502548. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the official policies, either expressed or implied, of the sponsoring organizations or the U.S. government. Keywords: Hybrid Algorithms, Cross-Fertilization, Agent-Centered Search, Limited Lookahead, Pigeonhole Principle, Linear Programming Relaxation. “ Each problem I solved became a pattern, that I used later to solve other problems.” Ren´e Descartes To my wife, Tanya, and my sons, Nick and Michael ii Abstract By now Artificial Intelligence (AI), Theoretical Computer Science (CS theory) and Opera- tions Research (OR) have investigated a variety of search and optimization problems. However, methods from these scientific areas use different problem descriptions, models, and tools. They also address problems with particular efficiency requirements. For example, approaches from CS theory are mainly concerned with the worst-case scenarios and are not focused on empirical performance. A few efforts have tried to apply methods across areas. Usually a significant amount of work is required to make different approaches “talk the same language,” be suc- cessfully implemented, and, finally, solve the actual same problem with an overall acceptable efficiency. This thesis presents a systematic approach that attempts to advance the state of the art in the transfer of knowledge across the above mentioned areas.
    [Show full text]
  • Mechanical Identification of Materials and Structures with Optical Methods
    materials Article Mechanical Identification of Materials and Structures with Optical Methods and Metaheuristic Optimization Elisa Ficarella 1, Luciano Lamberti 1,* and Sadik Ozgur Degertekin 2 1 Dipartimento di Meccanica, Matematica e Management, Politecnico di Bari, 70126 Bari, Italy 2 Department of Civil Engineering, Dicle University, 21280 Diyarbakır, Turkey * Correspondence: [email protected]; Tel.: +39-080-5962774 Received: 27 May 2019; Accepted: 24 June 2019; Published: 2 July 2019 Abstract: This study presents a hybrid framework for mechanical identification of materials and structures. The inverse problem is solved by combining experimental measurements performed by optical methods and non-linear optimization using metaheuristic algorithms. In particular, we develop three advanced formulations of Simulated Annealing (SA), Harmony Search (HS) and Big Bang-Big Crunch (BBBC) including enhanced approximate line search and computationally cheap gradient evaluation strategies. The rationale behind the new algorithms—denoted as Hybrid Fast Simulated Annealing (HFSA), Hybrid Fast Harmony Search (HFHS) and Hybrid Fast Big Bang-Big Crunch (HFBBBC)—is to generate high quality trial designs lying on a properly selected set of descent directions. Besides hybridizing SA/HS/BBBC metaheuristic search engines with gradient information and approximate line search, HS and BBBC are also hybridized with an enhanced 1-D probabilistic search derived from SA. The results obtained in three inverse problems regarding composite and transversely isotropic hyperelastic materials/structures with up to 17 unknown properties clearly demonstrate the validity of the proposed approach, which allows to significantly reduce the number of structural analyses with respect to previous SA/HS/BBBC formulations and improves robustness of metaheuristic search engines.
    [Show full text]
  • Sequential Quadratic Programming Methods∗
    SEQUENTIAL QUADRATIC PROGRAMMING METHODS∗ Philip E. GILLy Elizabeth WONGy UCSD Department of Mathematics Technical Report NA-10-03 August 2010 Abstract In his 1963 PhD thesis, Wilson proposed the first sequential quadratic programming (SQP) method for the solution of constrained nonlinear optimization problems. In the intervening 48 years, SQP methods have evolved into a powerful and effective class of methods for a wide range of optimization problems. We review some of the most prominent developments in SQP methods since 1963 and discuss the relationship of SQP methods to other popular methods, including augmented Lagrangian methods and interior methods. Given the scope and utility of nonlinear optimization, it is not surprising that SQP methods are still a subject of active research. Recent developments in methods for mixed-integer nonlinear programming (MINLP) and the minimization of functions subject to differential equation constraints has led to a heightened interest in methods that may be \warm started" from a good approximate solution. We discuss the role of SQP methods in these contexts. Key words. Large-scale nonlinear programming, SQP methods, nonconvex pro- gramming, quadratic programming, KKT systems. AMS subject classifications. 49J20, 49J15, 49M37, 49D37, 65F05, 65K05, 90C30 1. Introduction This paper concerns the formulation of methods for solving the smooth nonlinear pro- grams that arise as subproblems within a method for mixed-integer nonlinear programming (MINLP). In general, this subproblem has both linear and nonlinear constraints, and may be written in the form minimize f(x) n x2R 8 9 < x = (1.1) subject to ` ≤ Ax ≤ u; : c(x) ; where f(x) is a linear or nonlinear objective function, c(x) is a vector of m nonlinear constraint functions ci(x), A is a matrix, and l and u are vectors of lower and upper bounds.
    [Show full text]
  • An Efficient Sequential Quadratic Programming Algorithm For
    View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Elsevier - Publisher Connector Journal of Computational and Applied Mathematics 175 (2005) 447–464 www.elsevier.com/locate/cam An efficient sequential quadratic programming algorithm for nonlinear programming Zhibin Zhu∗ Department of Computational Science and Mathematics, Guilin Institute of Electronic Technology, No. 1, JinJi Road, Guilin 541004, PR China Received 23 November 2003; received in revised form 27 June 2004 Abstract In this paper, a new feasible sequential quadratic programming (FSQP) algorithm is proposed to solve the nonlinear programming, where a feasible descent direction is obtained by solving only one QP subproblem.In order to avoid Maratos effect, a high-order revised direction is computed by solving a linear system with involving some “active” constraints.The theoretical analysis shows that global and superlinear convergence can be deduced. © 2004 Elsevier B.V. All rights reserved. Keywords: Inequality constrained optimization; Method of feasible direction; SQP algorithm; Global convergence; Superlinear convergence rate 1. Introduction Consider the following inequality constrained nonlinear programming problem: min f(x) s.t. g(x)0, (1.1) where f : Rn → R,g : Rn → Rm are continuously differentiable.Such problems arise in a variety of applications in science and engineering.The method of feasible directions (MFD) is widely acknowledged to be among the most successful algorithms available for solving problem (1.1). ∗ Tel.: 07735608612; fax: 07735601519. E-mail address: [email protected] (Z.Zhu). 0377-0427/$ - see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.cam.2004.07.001 448 Z.
    [Show full text]
  • NORTHWESTERN UNIVERSITY Inexact Sequential Quadratic
    NORTHWESTERN UNIVERSITY Inexact Sequential Quadratic Programming Methods for Large-Scale Nonlinear Optimization A DISSERTATION SUBMITTED TO THE GRADUATE SCHOOL IN PARTIAL FULFILLMENT OF THE REQUIREMENTS for the degree DOCTOR OF PHILOSOPHY Field of Industrial Engineering and Management Sciences By Frank Edward Curtis EVANSTON, ILLINOIS June 2007 2 c Copyright by Frank Edward Curtis 2007 ⃝ All rights reserved 3 ABSTRACT Inexact Sequential Quadratic Programming Methods for Large-Scale Nonlinear Optimization Frank Edward Curtis This thesis concerns the development of robust algorithms for large-scale nonlinear pro- gramming. Despite recent advancements in high-performance computing power, classes of problems exist that continue to challenge the practical limits of contemporary optimization methods. The focus of this dissertation is the design and analysis of algorithms intended to achieve economy of computation for the solution of such problems, where, for ease of presentation, the discussion is framed in the context of equality constrained optimization. The first part of this thesis concerns the development of a globally convergent inexact Sequential Quadratic Programming (SQP) framework. The novelty of the approach is that it is matrix-free (i.e., only mechanisms for computing products of vectors with Jacobian and Hessian matrices are required), thereby avoiding a need for the explicit formation of the arising iteration matrices. Iterative linear algebra techniques, for example, can then be used in place of factorization methods, allowing the introduction of inexactness into the step computation process to achieve further savings in computation. The algorithm automatically determines when a given inexact SQP step makes sufficient progress toward a solution of the nonlinear program, as measured by an exact penalty function, in order to ensure global convergence to a first order optimal point.
    [Show full text]
  • Subgradient Based Multiple-Starting-Point Algorithm for Non-Smooth Optimization of Analog Circuits
    Subgradient Based Multiple-Starting-Point Algorithm for Non-Smooth Optimization of Analog Circuits Wenlong Lv1, Fan Yang1∗, Changhao Yan1, Dian Zhou1,2 and Xuan Zeng1∗ 1State Key Lab of ASIC & System, School of Microelectronics, Fudan University, Shanghai, P. R. China 2Department of Electrical Engineering, University of Texas at Dallas, Richardson, TX, U.S.A. Abstract—Starting from a set of starting points, the multiple- is to transform the general polynomial optimization problem starting-point optimization searches the local optimums by to convex programming by Semidefinite-Programming (SDP) gradient-guided local search. The global optimum is selected relaxations [4]. The advantage of model based algorithms from these local optimums. The region-hit property of the multiple-starting-point optimization makes the multiple-starting- is that given the circuit performance models, the optimal point approach more likely to reach the global optimum. solutions can be found very efficiently even for large-scale However, for non-smooth objective functions, e.g., worst-case problems. But what limits their usage is the generation of optimization, the traditional gradient based local search methods accurate performance models. Even with a large number of may stuck at non-smooth points, even if the objective function simulations, the accuracy of the generated performance models is smooth “almost everywhere”. In this paper, we propose a subgradient based multiple-starting-point algorithm for non- still cannot be guaranteed for the whole design space. smooth optimization of analog circuits. Subgradients instead of The simulation based analog circuit optimization methods traditional gradients are used to guide the local search of the non- require no explicit analytical performance models.
    [Show full text]
  • Modified Nonmonotone Armijo Line Search for Descent Method
    Numer Algor (2011) 57:1–25 DOI 10.1007/s11075-010-9408-7 ORIGINAL PAPER Modified nonmonotone Armijo line search for descent method Zhenjun Shi · Shengquan Wang Received: 23 February 2009 / Accepted: 20 June 2010 / Published online: 3 July 2010 © Springer Science+Business Media, LLC 2010 Abstract Nonmonotone line search approach is a new technique for solving optimization problems. It relaxes the line search range and finds a larger step-size at each iteration, so as to possibly avoid local minimizer and run away from narrow curved valley. It is helpful to find the global minimizer of optimization problems. In this paper we develop a new modification of matrix- free nonmonotone Armijo line search and analyze the global convergence and convergence rate of the resulting method. We also address several approaches to estimate the Lipschitz constant of the gradient of objective functions that would be used in line search algorithms. Numerical results show that this new modification of Armijo line search is efficient for solving large scale unconstrained optimization problems. Keywords Unconstrained optimization · Nonmonotone line search · Convergence Mathematics Subject Classifications (2010) 90C30 · 65K05 · 49M37 The work was supported in part by Rackham Faculty Research Grant of University of Michigan, USA. Z. Shi (B) Department of Mathematics and Computer Science, Central State University, Wilberforce, OH 45384, USA e-mail: [email protected] S. Wang Department of Computer and Information Science, The University of Michigan, Dearborn, MI 48128, USA e-mail: [email protected] 2 Numer Algor (2011) 57:1–25 1 Introduction Let Rn be an n-dimensional Euclidean space and f : Rn → R1 be a continu- ously differentiable function, where n is a positive integer.
    [Show full text]
  • Coordinate Descent Heuristics for the Irregular Strip Packing Problem of Rasterized Shapes
    Coordinate descent heuristics for the irregular strip packing problem of rasterized shapes Shunji Umetani,∗ Shohei Murakami† April 13, 2021 Abstract We consider the irregular strip packing problem of rasterized shapes, where a given set of pieces of irregular shapes represented in pixels should be placed into a rectangular container without overlap. The rasterized shapes enable us to check overlap without any exceptional handling due to geometric issues, while they often require much memory and computational effort in high-resolution. We develop an efficient algorithm to check overlap using a pair of scanlines that reduces the complexity of rasterized shapes by merging consecutive pixels in each row and column into strips with unit width, respectively. Based on this, we develop coordinate descent heuristics that repeat a line search in the horizontal and vertical direc- tions alternately. Computational results for test instances show that the proposed algorithm obtains sufficiently dense layouts of rasterized shapes in high-resolution within a reasonable computation time. keyword: packing, irregular strip packing problem, nesting problem, raster model, coordi- nate descent heuristics 1 Introduction The irregular strip packing problem (ISP), or often called the nesting problem, is the one of the representative cutting and packing problems that emerges in a wide variety of industrial applications, such as garment manufacturing, sheet metal cutting, furniture making and shoe manufacturing [Alvarez-Valdes et al., 2018, Scheithauer, 2018]. This problem is categorized as the two-dimensional, irregular open dimensional problem in Dyckhoff [1990], W¨ascher et al. [2007]. Given a set of pieces of irregular shapes and a rectangular container with a fixed width and a variable length, this problem asks a feasible layout of the pieces into the container such that no pair of pieces overlaps with each other and the container length is minimized.
    [Show full text]