Interior-Point and Augmented Lagrangian Algorithms for Optimization and Control

Total Page:16

File Type:pdf, Size:1020Kb

Interior-Point and Augmented Lagrangian Algorithms for Optimization and Control Interior-Point and Augmented Lagrangian Algorithms for Optimization and Control Stephen Wright University of Wisconsin-Madison May 2014 Wright (UW-Madison) Constrained Optimization May 2014 1 / 46 In This Section... My first talk was about optimization formulations, optimality conditions, and duality | for LP, QP, LCP, and nonlinear optimization. This section will review some algorithms, in particular: Primal-dual interior-point (PDIP) methods Augmented Lagrangian (AL) methods. Both are useful in control applications. We'll say something about PDIP methods for model-predictive control, and how they exploit the structure in that problem. Wright (UW-Madison) Constrained Optimization May 2014 2 / 46 Recapping Gradient Methods Considering unconstrained minimization min f (x) where f is smooth and convex, or the constrained version in which x is restricted to the set Ω, usually closed and convex. First-order or gradient methods take steps of the form xk+1 = xk − αk gk ; where αk 2 R+ is a steplength and gk is a search direction. dk is constructed from knowledge of the gradient rf (x) at the current iterate x = xk and possibly previous iterates xk−1; xk−2;::: . Can extend to nonsmooth f by using the subgradient @f (x). Extend to constrained minimization by projecting the search line onto the convex set Ω, or (similarly) minimizing a linear approximation to f over Ω. Wright (UW-Madison) Constrained Optimization May 2014 3 / 46 Prox Interpretation of Line Search Can view the gradient method step xk+1 = xk − αk gk as the minimization of a first-order model of f plus a \prox-term" which prevents the step from being too long: T 1 2 xk+1 = arg min f (xk ) + gk (x − xk ) + kx − xk k2: x 2αk Taking the gradient the quadratic and setting to zero, we obtain 1 gk + (xk+1 − xk ) = 0; αk which gives the formula for xk+1. This viewpoint is the key to several extensions. Wright (UW-Madison) Constrained Optimization May 2014 4 / 46 Extensions: Constraints When a constraint set Ω is present we can simply minimize the quadratic model function over Ω: T 1 2 xk+1 = arg min f (xk ) + gk (x − xk ) + kx − xk k2: x2Ω 2αk Gradient Projection has this form. We can replace the `2-norm measure of distance with some other measure φ(x; xk ): T 1 xk+1 = arg min f (xk ) + gk (x − xk ) + φ(x; xk ): x2Ω 2αk Could choose φ to \match" Ω. For example, a measure derived from the entropy function is a good match for the simplex Ω := fx j x ≥ 0; eT x = 1g: Wright (UW-Madison) Constrained Optimization May 2014 5 / 46 Extensions: Regularizers In many modern applications of optimization, f has the form f (x) = l(x) + τ (x): |{z} | {z } smooth function simple nonsmooth function Can extend the prox approach above by choosing gk to contain gradient information from l(x) only; including τ (x) explicitly in the subproblem. Subproblems are thus: T 1 2 xk+1 = arg min l(xk ) + gk (x − xk ) + kx − xk k + τ (x): x 2αk Wright (UW-Madison) Constrained Optimization May 2014 6 / 46 Extensions: Explicit Trust Regions Rather than penalizing distance moved from current xk , we can enforce an explicit constraint: a trust region. T xk+1 = arg min f (xk ) + g (x − xk ) + Ikx−x k≤∆ (x); x k k k where IΛ(x) denotes an indicator function with ( 0 if x 2 Λ IΛ(x) = 1 otherwise: Adjust trust-region radius ∆k to ensure progress e.g. descent in f . Wright (UW-Madison) Constrained Optimization May 2014 7 / 46 Extension: Proximal Point Could use the original f in the subproblem rather than a simpler model function: 1 2 xk+1 = arg min f (x) + kx − xk k2: x 2αk although the subproblem seems \just as hard" to solve as the original, the prox-term may make it easier by introducing strong convexity, and may stabilize progress. Can extend to constrained and regularized cases also. Wright (UW-Madison) Constrained Optimization May 2014 8 / 46 Quadratic Models: Newton's Method We can extend the iterative strategy further by adding a quadratic term to the model, instead of (or in addition to) the simple prox-term above. Taylor's Theorem suggests basing this term on the Hessian (second-derivative) matrix. That is, obtain the step from T 1 T 2 xk+1 := arg min f (xk ) + rf (xk ) (x − xk ) + (x − xk ) r f (xk )(x − xk ): x 2 Can reformulate to solve for the step dk : Then have xk+1 = xk + dk , where T 1 T 2 dk := arg min f (xk ) + rf (xk ) d + d r f (xk )d: d 2 2 See immediately that this model won't have a bounded solution if r f (xk ) is not positive definite. It usually is positive definite near a strict local solution x∗, but need something that works more globally. Wright (UW-Madison) Constrained Optimization May 2014 9 / 46 Practical Newton Method One \obvious" strategy is to add the prox-term to the quadratic model: T 1 T 2 1 dk := arg min f (xk ) + rf (xk ) d + d r f (xk ) + I d; d 2 αk choosing αk so that The quadratic term is positive definite; Some other desirable property holds, e.g. descent f (xk + dk ) < f (xk ). We can also impose the trust-region explicitly: T 1 T 2 dk := arg min f (xk ) + rf (xk ) d + d r f (xk )d + Ikdk≤∆ (x); d 2 k or alternatively: T 1 T 2 dk := arg min f (xk ) + rf (xk ) d + d r f (xk )d: d : kdk≤∆k 2 But this is equivalent: For any ∆k , there exists αk such that the solutions of the prox form and the trust-region form are identical. Wright (UW-Madison) Constrained Optimization May 2014 10 / 46 Quasi-Newton Methods Another disadvantage of Newton is that the Hessian may be difficult to evaluate or otherwise work with. The quadratic model is still useful when we use first-order information to learn about the Hessian. Key observation (from Taylor's theorem): the secant condition: 2 r f (xk )(xk+1 − xk ) ≈ rf (xk+1) − rf (xk ): The difference of gradients tells us how the Hessian behaves along the direction xk+1 − xk . By aggregating such information over multiple steps, we can build up an approximation to the Hessian than is valid along multiple directions. 2 Quasi-Newton Methods maintain an approximation Bk to r f (xk ) that respects the secant condition. The approximation may be implicit rather than explicit, and we may store an approximation to the inverse Hessian instead. Wright (UW-Madison) Constrained Optimization May 2014 11 / 46 L-BFGS A particularly populat quasi-Newton method, suitable for large-scale problems is the limited-memory BFGS method (L-BFGS) which stores the Hessian or inverse Hessian approximation implicitly. L-BFGS stores the last 5 − 10 update pairs: sj := xj+1 − xj ; yj := rf (xj+1) − rf (xj ): for j = k; k − 1; k − 2;:::; k − m. Can implicitly construct Hk+1 that satisfies Hk+1yj = sj : In fact, an efficient recursive formula is available for evaluating dk+1 := −Hk+1rf (xk+1) | the next search direction | directly from the (sj ; yj ) pairs and from some initial estimate of the form (1/αk+1)I . Wright (UW-Madison) Constrained Optimization May 2014 12 / 46 Newton for Nonlinear Equations There is also a variant of Newton's method for nonlinear equations: Find x such that F (x) = 0; n n where F : R ! R (n equations in n unknowns.) Newton's method forms a linear approximation to this system, based on another variant of Taylor's Theorem which says Z 1 F (x + d) = F (x) + J(x)d + [J(x + td) − J(x)]d dt; 0 where J(x) is the Jacobian matrix of first partial derivatives: 2 @F1 ··· @F1 3 @x1 @xn 6 . 7 J(x) = 4 . 5 (usually not symmetric). @Fn ··· @Fn @x1 @xn Wright (UW-Madison) Constrained Optimization May 2014 13 / 46 When F is continuously differentiable, we have F (xk + d) ≈ F (xk ) + J(xk )d; so the Newton step is the one that make the right-hand side zero: −1 dk := −J(xk ) F (xk ): The basis Newton method takes steps xk+1 := xk + dk . Its effectiveness can be improved by Doing a line search: xk+1 := xk + αk dk , for some αk > 0; −1 Levenberg strategy: add λI to J and set dk := −(J(xk ) + λI ) F (xk ). Guide progress via a merit function, usually 1 φ(x) := kF (x)k2: 2 2 Achtung! Can get stuck in a local min of φ that's not a solution of F (x) = 0. Wright (UW-Madison) Constrained Optimization May 2014 14 / 46 Homotopy Tries to avoid the local-min issue with the merit function. Start with an \easier" set of nonlinear equations, and gradually deform it to the system F (x) = 0, tracking changes to the solution as you go. F (x; λ) := λF (x) + (1 − λ)F0(x); λ 2 [0; 1]: Assume that F (x; 0) = F0(x) = 0 has solution x0. Homotopy methods trace the curve of solutions (x; λ) until λ = 1 is reached. The corresponding value of x then solves the original problem. Many variants. Some supporting theory. Typically more expensive than enhanced Newton methods, but better at finding solutions to F (x) = 0. We mention homotopy mostly because of its connection to interior-point methods. Wright (UW-Madison) Constrained Optimization May 2014 15 / 46 Interior-Point Methods n Recall the monotone LCP: Find z 2 R such that 0 ≤ z ? Mz + q ≥ 0; n×n n where M 2 R is positive semidefinite, and q 2 R .
Recommended publications
  • Using Ant System Optimization Technique for Approximate Solution to Multi- Objective Programming Problems
    International Journal of Scientific & Engineering Research, Volume 4, Issue 9, September-2013 ISSN 2229-5518 1701 Using Ant System Optimization Technique for Approximate Solution to Multi- objective Programming Problems M. El-sayed Wahed and Amal A. Abou-Elkayer Department of Computer Science, Faculty of Computers and Information Suez Canal University (EGYPT) E-mail: [email protected], Abstract In this paper we use the ant system optimization metaheuristic to find approximate solution to the Multi-objective Linear Programming Problems (MLPP), the advantageous and disadvantageous of the suggested method also discussed focusing on the parallel computation and real time optimization, it's worth to mention here that the suggested method doesn't require any artificial variables the slack and surplus variables are enough, a test example is given at the end to show how the method works. Keywords: Multi-objective Linear. Programming. problems, Ant System Optimization Problems Introduction The Multi-objective Linear. Programming. problems is an optimization method which can be used to find solutions to problems where the Multi-objective function and constraints are linear functions of the decision variables, the intersection of constraints result in a polyhedronIJSER which represents the region that contain all feasible solutions, the constraints equation in the MLPP may be in the form of equalities or inequalities, and the inequalities can be changed to equalities by using slack or surplus variables, one referred to Rao [1], Philips[2], for good start and introduction to the MLPP. The LP type of optimization problems were first recognized in the 30's by economists while developing methods for the optimal allocations of resources, the main progress came from George B.
    [Show full text]
  • An Annotated Bibliography of Network Interior Point Methods
    AN ANNOTATED BIBLIOGRAPHY OF NETWORK INTERIOR POINT METHODS MAURICIO G.C. RESENDE AND GERALDO VEIGA Abstract. This paper presents an annotated bibliography on interior point methods for solving network flow problems. We consider single and multi- commodity network flow problems, as well as preconditioners used in imple- mentations of conjugate gradient methods for solving the normal systems of equations that arise in interior network flow algorithms. Applications in elec- trical engineering and miscellaneous papers complete the bibliography. The collection includes papers published in journals, books, Ph.D. dissertations, and unpublished technical reports. 1. Introduction Many problems arising in transportation, communications, and manufacturing can be modeled as network flow problems (Ahuja et al. (1993)). In these problems, one seeks an optimal way to move flow, such as overnight mail, information, buses, or electrical currents, on a network, such as a postal network, a computer network, a transportation grid, or a power grid. Among these optimization problems, many are special classes of linear programming problems, with combinatorial properties that enable development of efficient solution techniques. The focus of this annotated bibliography is on recent computational approaches, based on interior point methods, for solving large scale network flow problems. In the last two decades, many other approaches have been developed. A history of computational approaches up to 1977 is summarized in Bradley et al. (1977). Several computational studies established the fact that specialized network simplex algorithms are orders of magnitude faster than the best linear programming codes of that time. (See, e.g. the studies in Glover and Klingman (1981); Grigoriadis (1986); Kennington and Helgason (1980) and Mulvey (1978)).
    [Show full text]
  • An Affine-Scaling Pivot Algorithm for Linear Programming
    AN AFFINE-SCALING PIVOT ALGORITHM FOR LINEAR PROGRAMMING PING-QI PAN¤ Abstract. We proposed a pivot algorithm using the affine-scaling technique. It produces a sequence of interior points as well as a sequence of vertices, until reaching an optimal vertex. We report preliminary but favorable computational results obtained with a dense implementation of it on a set of standard Netlib test problems. Key words. pivot, interior point, affine-scaling, orthogonal factorization AMS subject classifications. 65K05, 90C05 1. Introduction. Consider the linear programming (LP) problem in the stan- dard form min cT x (1.1) subject to Ax = b; x ¸ 0; where c 2 Rn; b 2 Rm , A 2 Rm£n(m < n). Assume that rank(A) = m and c 62 range(AT ). So the trivial case is eliminated where the objective value cT x is constant over the feasible region of (1.1). The assumption rank(A) = m is not substantial either to the proposed algorithm,and will be dropped later. As a LP problem solver, the simplex algorithm might be one of the most famous and widely used mathematical tools in the world. Its philosophy is to move on the underlying polyhedron, from a vertex to adjacent vertex, along edges until an optimal vertex is reached. Since it was founded by G.B.Dantzig [8] in 1947, the simplex algorithm has been very successful in practice, and occupied a dominate position in this area, despite its infiniteness in case of degeneracy. However, it turned out that the simplex algorithm may require an exponential amount of time to solve LP problems in the worst-case [29].
    [Show full text]
  • On Some Polynomial-Time Algorithms for Solving Linear Programming Problems
    Available online a t www.pelagiaresearchlibrary.com Pelagia Research Library Advances in Applied Science Research, 2012, 3 (5):3367-3373 ISSN: 0976-8610 CODEN (USA): AASRFC On Some Polynomial-time Algorithms for Solving Linear Programming Problems B. O. Adejo and H. S. Adaji Department of Mathematical Sciences, Kogi State University, Anyigba _____________________________________________________________________________________________ ABSTRACT In this article we survey some Polynomial-time Algorithms for Solving Linear Programming Problems namely: the ellipsoid method, Karmarkar’s algorithm and the affine scaling algorithm. Finally, we considered a test problem which we solved with the methods where applicable and conclusions drawn from the results so obtained. __________________________________________________________________________________________ INTRODUCTION An algorithm A is said to be a polynomial-time algorithm for a problem P, if the number of steps (i. e., iterations) required to solve P on applying A is bounded by a polynomial function O(m n,, L) of dimension and input length of the problem. The Ellipsoid method is a specific algorithm developed by soviet mathematicians: Shor [5], Yudin and Nemirovskii [7]. Khachian [4] adapted the ellipsoid method to derive the first polynomial-time algorithm for linear programming. Although the algorithm is theoretically better than the simplex algorithm, which has an exponential running time in the worst case, it is very slow practically and not competitive with the simplex method. Nevertheless, it is a very important theoretical tool for developing polynomial-time algorithms for a large class of convex optimization problems. Narendra K. Karmarkar is an Indian mathematician; renowned for developing the Karmarkar’s algorithm. He is listed as an ISI highly cited researcher.
    [Show full text]
  • An Affine-Scaling Interior-Point Method for Continuous Knapsack Constraints with Application to Support Vector Machines∗
    SIAM J. OPTIM. c 2011 Society for Industrial and Applied Mathematics Vol. 21, No. 1, pp. 361–390 AN AFFINE-SCALING INTERIOR-POINT METHOD FOR CONTINUOUS KNAPSACK CONSTRAINTS WITH APPLICATION TO SUPPORT VECTOR MACHINES∗ MARIA D. GONZALEZ-LIMA†, WILLIAM W. HAGER‡ , AND HONGCHAO ZHANG§ Abstract. An affine-scaling algorithm (ASL) for optimization problems with a single linear equality constraint and box restrictions is developed. The algorithm has the property that each iterate lies in the relative interior of the feasible set. The search direction is obtained by approximating the Hessian of the objective function in Newton’s method by a multiple of the identity matrix. The algorithm is particularly well suited for optimization problems where the Hessian of the objective function is a large, dense, and possibly ill-conditioned matrix. Global convergence to a stationary point is established for a nonmonotone line search. When the objective function is strongly convex, ASL converges R-linearly to the global optimum provided the constraint multiplier is unique and a nondegeneracy condition holds. A specific implementation of the algorithm is developed in which the Hessian approximation is given by the cyclic Barzilai-Borwein (CBB) formula. The algorithm is evaluated numerically using support vector machine test problems. Key words. interior-point, affine-scaling, cyclic Barzilai-Borwein methods, global convergence, linear convergence, support vector machines AMS subject classifications. 90C06, 90C26, 65Y20 DOI. 10.1137/090766255 1. Introduction. In this paper we develop an interior-point algorithm for a box-constrained optimization problem with a linear equality constraint: (1.1) min f(x) subject to x ∈ Rn, l ≤ x ≤ u, aTx = b.
    [Show full text]
  • Interior Point Method
    LECTURE 6: INTERIOR POINT METHOD 1. Motivation 2. Basic concepts 3. Primal affine scaling algorithm 4. Dual affine scaling algorithm Motivation • Simplex method works well in general, but suffers from exponential-time computational complexity. • Klee-Minty example shows simplex method may have to visit every vertex to reach the optimal one. • Total complexity of an iterative algorithm = # of iterations x # of operations in each iteration • Simplex method - Simple operations: Only check adjacent extreme points - May take many iterations: Klee-Minty example Question: any fix? • Complexity of the simplex method Worst case performance of the simplex method Klee-Minty Example: •Victor Klee, George J. Minty, “How good is the simplex algorithm?’’ in (O. Shisha edited) Inequalities, Vol. III (1972), pp. 159-175. Klee-Minty Example Klee-Minty Example Karmarkar’s (interior point) approach • Basic idea: approach optimal solutions from the interior of the feasible domain • Take more complicated operations in each iteration to find a better moving direction • Require much fewer iterations General scheme of an interior point method • An iterative method that moves in the interior of the feasible domain Interior movement (iteration) • Given a current interior feasible solution , we have > 0 An interior movement has a general format Key knowledge • 1. Who is in the interior? - Initial solution • 2. How do we know a current solution is optimal? - Optimality condition • 3. How to move to a new solution? - Which direction to move? (good feasible direction) - How far to go? (step-length) Q1 - Who is in the interior? • Standard for LP • Who is at the vertex? • Who is on the edge? • Who is on the boundary? • Who is in the interior? What have learned before Who is in the interior? • Two criteria for a point x to be an interior feasible solution: 1.
    [Show full text]
  • Generalized Affine Scaling Algorithms for Linear Programming Problems
    Generalized Affine Scaling Algorithms for Linear Programming Problems Md Sarowar Morshed and Md Noor-E-Alam Department of Mechanical & Industrial Engineering Northeastern University 360 Huntington Avenue, Boston, MA 02115, USA Email : [email protected] Abstract Interior Point Methods are widely used to solve Linear Programming problems. In this work, we present two primal Affine Scaling algorithms to achieve faster convergence in solving Linear Programming problems. In the first algorithm, we integrate Nesterov's restarting strategy in the primal Affine Scaling method with an extra parameter, which in turn generalizes the original primal Affine Scaling method. We provide the proof of convergence for the proposed generalized al- gorithm considering long step size. We also provide the proof of convergence for the primal and dual sequence without the degeneracy assumption. This con- vergence result generalizes the original convergence result for the Affine Scaling methods and it gives us hints about the existence of a new family of meth- ods. Then, we introduce a second algorithm to accelerate the convergence rate of the generalized algorithm by integrating a non-linear series transformation technique. Our numerical results show that the proposed algorithms outperform the original primal Affine Scaling method. Key words: Linear Programming, Affine Scaling, Nesterov Acceleration, Dikin Process, Shanks Series Transformation 1 Introduction The Affine Scaling (AFS) algorithm was introduced by Dikin [1], which re- mained unnoticed to the Operations Research (OR) community until the seminal work of Karmarkar [2]. Karmarkar's work transformed the research in Interior Point Methods (IPMs) and induced a significant development in the theory of arXiv:1804.07373v4 [math.OC] 12 Oct 2019 IPMs.
    [Show full text]
  • Generalization of the Primal and Dual Affine Scaling Algorithms
    Generalization of the primal and dual a±ne scaling algorithms F. G. M. CUNHAy, A. W. M. PINTOz, P. R. OLIVEIRA¤x and J. X. da C. NETO{ yDepartment of Mathematics, Cear¶aFederal Centre of Technological Education, Fortaleza, Brazil; zDepartment of Mathematics, Federal University of Amazonas, Manaus, Brazil; xDepartment of Systems and Computing Engineering, Rio de Janeiro Federal University, Rio de Janeiro, Brazil; {Department of Mathematics, Federal University of Piau¶³,Teresina, Brazil; (v1.0 ¯nished 23 April 2005) We obtain a new class of primal a±ne scaling algorithms for the linearly constrained convex program- ming. It is constructed through a family of metrics generated by ¡r power, r ¸ 1, of the diagonal iterate vector matrix. We prove the so-called weak convergence. It generalizes some known algo- rithms. Working in dual space, we generalize the dual a±ne scaling algorithm of Adler, Karmarkar, Resende and Veiga, similarly depending on a r-parameter and we give its global convergence proof for nondegenerate linear programs. With the purpose of observing the computational performance of the methods, we compare them with classical algorithms (when r = 1 or r = 2), implementing the proposed families and applying to some linear programs obtained from NETLIB library. In the case of primal family, we also apply it to some quadratic programming problems described in the Maros and M¶esz¶arosrepository. Keywords: Interior Point Algorithms; A±ne Scaling Algorithms; Linear Convex Programming 1 Introduction We are interested in a±ne scaling potential algorithms. If we restrict ourselves to primal interior point methods which is one class we are interested in, we mention the a±ne scaling algorithm for linear and quadratic cost functions [1] and the polynomial logarithmic barrier approach for convex self-concordant functions, as can be seen in [2].
    [Show full text]
  • Affine Scaling I Orithll1s
    7 Affine Scaling I orithll1s Since its introduction in 1984, Karmarkar's projective scaling algorithm has become the most notable interior-point method for solving linear programming problems. This pioneering work has stimulated a flurry of research activities in the field. Among all reported variants of Karmarkar' s original algorithm, the affine scaling approach especially attracted researchers' attention. This approach uses the simple affine transformation to replace Karmarkar's original projective transformation and allows people to work on the linear programming problems in standard form. The special simplex structure required by Karmarkar' s algorithm is relaxed. The basic affine scaling algorithm was first presented by I. I. Dikin, a Soviet mathematician, in 1967. Later, in 1985, the work was independently rediscovered by E. Barnes and R. Vanderbei, M. Meketon, and B. Freedman. They proposed using the (primal) affine scaling algorithm to solve the (primal) linear programs in standard form and established convergence proof of the algorithm. A similar algorithm, the so-called dual affine scaling algorithm, was designed and implemented by I. Adler, N. Karmarkar, M. G. C. Resende, and G. Veiga for solving (dual) linear programs in inequality form. Compared to the relatively cumbersome projective transformation, the implementation of both the primal and dual affine scaling algorithms become quite straightforward. These two algorithms are currently the variants subject to the widest experimentation and exhibit promising results, although the theoretical proof of polynomial-time complexity was lost in the simplified transformation. In fact, N. Megiddo and M. Shub's work indicated that the trajectory leading to the optimal solution provided by the basic affine scaling algorithms depends upon the starting solution.
    [Show full text]
  • Interior-Point Algorithms for Linear-Programming Decoding
    Interior-Point Algorithms for Linear-Programming Decoding0 Pascal O. Vontobel Hewlett–Packard Laboratories Palo Alto, CA 94304, USA Email: [email protected] Abstract— Interior-point algorithms constitute a very interest- for which it can easily be shown that all codewords in C are ing class of algorithms for solving linear-programming problems. vertices of P. In this paper we study efficient implementations of such algo- The same polytope P appeared also in papers by Koetter rithms for solving the linear program that appears in the linear- programming decoder formulation. and Vontobel [3], [4], [5], where message-passing iterative (MPI) decoders were analyzed and where this polytope P was I. INTRODUCTION called the fundamental polytope. The appearance of the same Consider a binary linear code C of length n that is used object in these two different contexts suggests that there is a for data transmission over a binary-input discrete memoryless tight connection between LP decoding and MPI decoding. channel. As was observed by Feldman et al. [1], [2], the ML The above codes Cj can be any codes of length n, however, decoder for this setup can be written as in the following we will focus on the case where these codes are codes of dimension n−1. For example, let H be an m×n T ˆxML = arg min hγ, xi, parity-check matrix for the code C and let h be the j-th row x∈C j of H.1 Then, defining where γ is a length-n vector that contains the log-likelihood x n h x ratios and where hγ, xi is the inner product (in R) of the Cj = ∈{0, 1} h j , i =0 (mod 2) vector γ with the vector x.
    [Show full text]
  • A New Class of Scaling Matrices for Scaled Trust Region Algorithms
    ISSN: 2377-3316 http://dx.doi.org/10.21174/josetap.v3i1. 44 A New Class of Scaling Matrices for Scaled Trust Region Algorithms Aydin Ayanzadeh*1, Shokoufeh Yazdanian2, Ehsan Shahamatnia3 1 Department of Applied Informatics, Informatics Institute, Istanbul Technical University, Istanbul, Turkey 2Department of Computer science, University of Tabriz, Tabriz, Iran 3Department of Computer and Electrical Engineering, Universidade Nova de Lisboa, Lisbon, Portugal Abstract— A new class of affine scaling matrices for the interior point Newton-type methods is considered to solve the nonlinear systems with simple bounds. We review the essential properties of a scaling matrix and consider several well- known scaling matrices proposed in the literature. We define a new scaling matrix that is the convex combination of these matrices. The proposed scaling matrix inherits those interesting properties of the individual matrices and satisfies additional desired requirements. The numerical experiments demonstrate the superiority of the new scaling matrix in solving several important test problems. Keywords— Diagonal Scaling; Bound-Constrained Equations; Scaling Matrices; Trust Region Methods; Affine Scaling. I. INTRODUCTION where FX: n is a continuously differentiable Diverse applications including signal processing mapping, X n is an open set containing the n- [36] and compressive sensing [3,4] incorporate many dimensional box ={x n | l x u }. The vectors convex nonlinear optimization problems. Although nn are lower and upper specialized algorithms have been developed for some of lu( { − }) , ( { + }) these problems, the interior-point methods are still the bounds on the variables such that has a nonempty main tool to tackle them. These methods require a interior. feasible initial point [19].
    [Show full text]
  • Affine-Scaling Method, 202–204, 206, 210 Poor Performance Of
    Index affine-scaling method, 202–204, 206, 210 blocking variable, 48 poor performance of, 203 breakpoint, 163 steplength choice, 203 approximation problems, 218–227 canonical form, 8, 16, 45, 117, 120, 123, 1-norm, 221–224 129, 130, 134, 138, 139, 150, 2-norm. See least-squares problems 151, 156, 195, 197 ∞-norm. See Chebyshev approximation dual of, 109 for linear equalities, 218–223 for problems with bound constraints, with hard constraints, 223–224, 227 130 arc, 12–14, 144 centering parameter, 206, 214 central path, 205 back substitution. See triangular substitution Chebyshev approximation, 219–221, 223 basic feasible solution, 118 Cholesky factorization, 209, 212 as iterates of revised simplex method, classification, 11–12, 230–235 123 labeled points, 230 association with basis matrix, 120–121 testing set, 235 association with vertex, 121–123 training set, 235 degenerate, 120 tuning set, 235 existence, 119–120 classifier, 230 initial, 129, 134–139 linear, 230 basic solution, 118 nonlinear, 233, 234 basis, 117, 126, 152 complementarity, 101–102, 178, 237 for problems with bound constraints, almost-, 178, 179 130 existence of strictly complementary initial, 125, 134–137 primal-dual solution, 114–115 optimal, 129, 152–156 strict, 101 basis matrix, 52, 118, 120, 123, 139, 140, complementary slackness. 154 See complementarity for network linear program, 149 concave function, 161, 244 LU factorization of, 139–142 constraints, 1 big M method, 110–112 active, 14 behavior on infeasible problems, 112 bounds, 74–75 motivation, 110–111 equality, 72–87 proof of effectiveness, 111 inactive, 14 specification, 111–112 inequality, 72–87 bimatrix game, 185 convex combination, 239, 241 Bland’s rule.
    [Show full text]