Network Systems Lab. Korea Advanced Institute of Science and Technology No.1 Ch. 3 Iterative Method for Nonlinear problems EE692 Parallel and Distribution.

Slides:



Advertisements
Similar presentations
5.4 Basis And Dimension.
Advertisements

Chapter 4 Euclidean Vector Spaces
Boyce/DiPrima 9th ed, Ch 2.8: The Existence and Uniqueness Theorem Elementary Differential Equations and Boundary Value Problems, 9th edition, by William.
Longest Common Subsequence
Optimization 吳育德.
Copyright © Cengage Learning. All rights reserved. CHAPTER 5 SEQUENCES, MATHEMATICAL INDUCTION, AND RECURSION SEQUENCES, MATHEMATICAL INDUCTION, AND RECURSION.
Network Systems Lab. Korea Advanced Institute of Science and Technology No.1 Some useful Contraction Mappings  Results for a particular choice of norms.
Induction and recursion
Infinite Horizon Problems
1 L-BFGS and Delayed Dynamical Systems Approach for Unconstrained Optimization Xiaohui XIE Supervisor: Dr. Hon Wah TAM.
Message Passing Algorithms for Optimization
Ch 7.3: Systems of Linear Equations, Linear Independence, Eigenvalues
Ch 3.3: Linear Independence and the Wronskian
Tutorial 10 Iterative Methods and Matrix Norms. 2 In an iterative process, the k+1 step is defined via: Iterative processes Eigenvector decomposition.
Linear and generalised linear models
Orthogonality and Least Squares
1 © 2012 Pearson Education, Inc. Matrix Algebra THE INVERSE OF A MATRIX.
Scientific Computing Matrix Norms, Convergence, and Matrix Condition Numbers.
Boyce/DiPrima 9th ed, Ch 7.3: Systems of Linear Equations, Linear Independence, Eigenvalues Elementary Differential Equations and Boundary Value Problems,
Copyright © Cengage Learning. All rights reserved. CHAPTER 11 ANALYSIS OF ALGORITHM EFFICIENCY ANALYSIS OF ALGORITHM EFFICIENCY.
Linear Equations in Linear Algebra
Linear Algebra Chapter 4 Vector Spaces.
Elementary Linear Algebra Anton & Rorres, 9th Edition
Advanced Counting Techniques CSC-2259 Discrete Structures Konstantin Busch - LSU1.
Simplex method (algebraic interpretation)
Linear Programming System of Linear Inequalities  The solution set of LP is described by Ax  b. Gauss showed how to solve a system of linear.
Network Systems Lab. Korea Advanced Institute of Science and Technology No.1 Appendix A. Mathematical Background EE692 Parallel and Distribution Computation.
Vectors CHAPTER 7. Ch7_2 Contents  7.1 Vectors in 2-Space 7.1 Vectors in 2-Space  7.2 Vectors in 3-Space 7.2 Vectors in 3-Space  7.3 Dot Product 7.3.
Pareto Linear Programming The Problem: P-opt Cx s.t Ax ≤ b x ≥ 0 where C is a kxn matrix so that Cx = (c (1) x, c (2) x,..., c (k) x) where c.
1 Markov Decision Processes Infinite Horizon Problems Alan Fern * * Based in part on slides by Craig Boutilier and Daniel Weld.
Vector Norms and the related Matrix Norms. Properties of a Vector Norm: Euclidean Vector Norm: Riemannian metric:
Colorado Center for Astrodynamics Research The University of Colorado 1 STATISTICAL ORBIT DETERMINATION ASEN 5070 LECTURE 11 9/16,18/09.
Relations, Functions, and Matrices Mathematical Structures for Computer Science Chapter 4 Copyright © 2006 W.H. Freeman & Co.MSCS Slides Relations, Functions.
Chap. 4 Vector Spaces 4.1 Vectors in Rn 4.2 Vector Spaces
I.4 Polyhedral Theory 1. Integer Programming  Objective of Study: want to know how to describe the convex hull of the solution set to the IP problem.
CompSci 102 Discrete Math for Computer Science March 1, 2012 Prof. Rodger Slides modified from Rosen.
Linear Program Set Cover. Given a universe U of n elements, a collection of subsets of U, S = {S 1,…, S k }, and a cost function c: S → Q +. Find a minimum.
Sets Definition: A set is an unordered collection of objects, called elements or members of the set. A set is said to contain its elements. We write a.
Chapter 8 Maximum Flows: Additional Topics All-Pairs Minimum Value Cut Problem  Given an undirected network G, find minimum value cut for all.
Assam Don Bosco University Waves in 3D Parag Bhattacharya Department of Basic Sciences School of Engineering and Technology.
Linear & Nonlinear Programming -- Basic Properties of Solutions and Algorithms.
Linear Programming Chapter 9. Interior Point Methods  Three major variants  Affine scaling algorithm - easy concept, good performance  Potential.
OR  Now, we look for other basic feasible solutions which gives better objective values than the current solution. Such solutions can be examined.
A function is a rule f that associates with each element in a set A one and only one element in a set B. If f associates the element b with the element.
Chapter 5. Section 5.1 Climbing an Infinite Ladder Suppose we have an infinite ladder: 1.We can reach the first rung of the ladder. 2.If we can reach.
Linear Programming Chap 2. The Geometry of LP  In the text, polyhedron is defined as P = { x  R n : Ax  b }. So some of our earlier results should.
Network Systems Lab. Korea Advanced Institute of Science and Technology No.1 Maximum Norms & Nonnegative Matrices  Weighted maximum norm e.g.) x1x1 x2x2.
Theory of Computational Complexity Probability and Computing Lee Minseon Iwama and Ito lab M1 1.
Matrices, Vectors, Determinants.
1 Proving Properties of Recursive List Functions CS 270 Math Foundations of CS Jeremy Johnson.
Advanced Algorithms Analysis and Design
Induction and recursion
Chapter 5. Optimal Matchings
Systems of First Order Linear Equations
Polyhedron Here, we derive a representation of polyhedron and see the properties of the generators. We also see how to identify the generators. The results.
Chap 3. The simplex method
Polyhedron Here, we derive a representation of polyhedron and see the properties of the generators. We also see how to identify the generators. The results.
Induction and recursion
3.5 Minimum Cuts in Undirected Graphs
§1-3 Solution of a Dynamical Equation
Chapter 5. The Duality Theorem
Affine Spaces Def: Suppose
I.4 Polyhedral Theory (NW)
5.4 T-joins and Postman Problems
I.4 Polyhedral Theory.
Advanced Analysis of Algorithms
Totally Asynchronous Iterative Algorithms
Variational Inequalities
Chapter 2. Simplex method
Presentation transcript:

Network Systems Lab. Korea Advanced Institute of Science and Technology No.1 Ch. 3 Iterative Method for Nonlinear problems EE692 Parallel and Distribution Computation | Prof. Song Chong

Network Systems Lab. Korea Advanced Institute of Science and Technology No.2 Nonlinear Problems  Nonlinear Problems to be covered  Systems of nonlinear equations  Optimization problems - constrained or unconstrained  Variational inequality - Viewed as a generalization of the aboves

Network Systems Lab. Korea Advanced Institute of Science and Technology No.3 Contraction Mappings  Several nonlinear iteration algorithms can be written as where T is a mapping from a set X of R n into itself and has the property : Here, is some vector norm, and is a constant belonging to [0,1)  Any vector satisfying is called a fixed point of T.  Show that contraction mappings are continuous.

Network Systems Lab. Korea Advanced Institute of Science and Technology No.4 Pseudo contraction  A mapping T: X → X has a fixed point and the property  Cleary, pseudocontraction condition is weaker than contraction condition  A pseudo contraction is not necessarily continuous x X 2 T(x) x X contraction pseudocontraction

Network Systems Lab. Korea Advanced Institute of Science and Technology No.5  A mapping T could be a contraction (or a pseudocontraction) for some choice of the vector norm, and fail to be a contraction ( or pseudocontraction) under a different choice of norm. → The proper choice of norm is crucial.  A nonnegative matrix M has iff it is a contraction mapping w.r.t some weighted maximum norm ( Cor. 6.1)  Proposition 1.1 Suppose that T : X → X is a contraction with modulus and that X is a closed subset of then, (a) (Existence and Uniqueness of fixed point) The mapping T has a unique fixed point (b) (Geometric Convergence) For every initial vector, the seq {x(t)} generated by x(t+1) = T{x(t)} converges to geometrically, In particular

Network Systems Lab. Korea Advanced Institute of Science and Technology No.6  Proof of prop 1.1 choose some By def. of contraction, therefore, for every and choose an arbitrary then, therefore, {x(t)} is a cauch seq, and hence must converge to a limit, say ( Prop. A.5) Furthermore, since X is closed,

Network Systems Lab. Korea Advanced Institute of Science and Technology No.7 For all t≥1, As Thus, T(x * ) = x *, i.e. is a fixed point of T. Suppose that y * (≠x * ) is another fixed point. Then, b.We have

Network Systems Lab. Korea Advanced Institute of Science and Technology No.8  Proposition 1.2  Suppose that and the mapping T: X -> X is a pseudo contraction with a fixed point and modulus Then, T has no other fixed points and the sequence {x(t)} generated by x(t+1) = T{x(t)} satisfies for every choice of the initial vector, In particular, {x(t)} converges to x *. Proof)  Uniqueness of the fixed point follows as in the proof of 1.1 By definition of pseudocontraction, By induction on t, we concluded. (*)

Network Systems Lab. Korea Advanced Institute of Science and Technology No.9 Contractions over Cartesian Product Sets  Block-maximum Norm  Assume that, where x i is a nonempty subset of and where n 1 +…+n m =n  Any vector is decomposed as x=(x 1,…,x m ) with  Given a norm || · || i on for each i, R n is endowed with the norm, termed block-maximum norm  Let T:X -> X be a contraction with modulus α under the above block- maximum norm (termed block contraction), and let T(x) = ( T 1 (x), …, T m (x) ) where T i :X -> X i is the i-th block-component of T Then, and Why?

Network Systems Lab. Korea Advanced Institute of Science and Technology No.10 Gauss-Seidel Methods  An iteration x(t+1) = T(x(t)) corresponds to updating all components of x simultaneously  A Gauss-Seidel mode of implementation of T -> block components of x are updated one at a time Define a mapping, corresponding to an update of the i- th block-component only, by Update all the block components of x, one at a time in increasing order is equivalent to applying the mapping S:X -> X defined by An equivalent definition of S is given by where S i :X -> X i is the i-th block-component of S

Network Systems Lab. Korea Advanced Institute of Science and Technology No.11 Gauss-Seidel Methods (Cont’d)  The mapping S is called Gauss-Seidel mapping based on T and the iteration x(t+1)=S(x(t)) is called Gauss-Seidel algorithm based on T.  Any fixed point of T is also a fixed point of S, and conversely.  Proposition 1.4 (Convergence of Gauss-Seidel Block-Contracting Iterations)  If T:X -> X is a block-contraction. Then the Gauss-Seidel mapping S is also a block-contraction with the same modulus as T.  In particular, if x is closed, the seq. of vectors generated by the Gauss-Seidel algorithm based on T converges to the unique fixed point of T geometrically.

Network Systems Lab. Korea Advanced Institute of Science and Technology No.12 Prop 1.4 & proof  Proof) For every and i = 1,…,m, An induction on i yields

Network Systems Lab. Korea Advanced Institute of Science and Technology No.13 Prop 1.4 & proof  Thus, S is a block-contraction with the same modulus as T  Therefore, by Prop1.1, the iteration converges to a unique fixed point of S in X geometrically.  Need to show that T and S have the same fixed point.  Suppose that is a unique fixed point of T

Network Systems Lab. Korea Advanced Institute of Science and Technology No.14 Prop 1.5  If a mapping has a fixed point and is a pseudo- contraction of modulus with respect to a block-maximum norm, then the same is true for the Gauss-Seidel mapping S, that is  In Particular, the sequence generated by converges to geometrically Pf.) simple

Network Systems Lab. Korea Advanced Institute of Science and Technology No.15 Component Solution Methods  Consider  Decompose this into m smaller systems  Algorithm that solves individual equations, the i-th equation for, while keeping the other components fixed.  Let be the set of all solutions of the i-th equation, defined by

Network Systems Lab. Korea Advanced Institute of Science and Technology No.16 Component Solution Methods * x )),(( 21 xxQ ))(( 2,1 xQx ),( 21 xxx  (),(()( 21 xQxQxQ   ),(| 2111 xxTxx   ),(| 2122 xxTxx  1 x 2 x

Network Systems Lab. Korea Advanced Institute of Science and Technology No.17 Prop 1.6 & proof  Suppose that X is closed and is a block contraction. Then, the set has exactly one element ( i.e., singleton) for each i and for each  Proof) For some i and some, and consider the mapping defined by Then is equal to the set of fixed points of. Let and. Then by block –contraction assumption on T Therefore, is a contraction w.r.t., which implies that it has a unique point in, Therefore, is a singleton. Q.E.D.

Network Systems Lab. Korea Advanced Institute of Science and Technology No.18 Component solution methods  Define a mapping by letting be equal to the unique element of. Then, the component solution method is  Gauss-Seidel algorithm based on the mapping Q is referred to as “Gauss-Seidel Component Solution Method” where the block- components of x are updated one at a time.

Network Systems Lab. Korea Advanced Institute of Science and Technology No.19 Gauss-Seidel Component Solution Method  Proposition 1.7 If T: X -> X is a block contraction, then Q is also a block-contraction with the same modulus as T. In particular, if X is closed, then the component solution method as well as Gauss-Seidel algorithm based on Q, converges to the unique fixed point of T geometrically.  ),(| 2122 xxTxx   ),(| 2111 xxTxx 

Network Systems Lab. Korea Advanced Institute of Science and Technology No.20 Gauss-Seidel Component Solution Method Pf. ) Let By definition of and using the block-contraction assumption on T, Thus, Q is a block-contraction with the same modulus as T. Since x is closed, Q has a unique fixed point, i.e. is equivalent to By definition of Thus, is the unique fixed point of T. Q.E.D

Network Systems Lab. Korea Advanced Institute of Science and Technology No.21 Gauss-Seidel Component Solution Method  Proposition 1.8 Suppose that T:X->X is continuous and a pseudocontraction w.r.t. a block- maximum norm. If each set X i is closed and convex, then the set R i (x) is nonempty for each i and for each x in X.  Proposition 1.9 Suppose that T: X -> X has a fixed point x * and is a pseudocontraction w.r.t. a block-maximum norm. Suppose that for every i and x in X, the set R i (x) is nonempty. Then, Q is also a pseudocontraction w.r.t. the same norm, and x * is its unique fixed point. In particular, the component solution method as well as the Gauss-Seidel algorithm based on Q, converges to x * geometrically.

Network Systems Lab. Korea Advanced Institute of Science and Technology No.22 Some useful Contraction Mappings  Consider a mapping T: X  R n whose i-th block component T i is of the form where is a fct. from into, is some scalar and G i is an invertible symmetric matrix of dimensions  Simple case

Network Systems Lab. Korea Advanced Institute of Science and Technology No.23 Some useful Contraction Mappings

Network Systems Lab. Korea Advanced Institute of Science and Technology No.24 Some useful Contraction Mappings  Proposition 1.10

Network Systems Lab. Korea Advanced Institute of Science and Technology No.25 Some useful Contraction Mappings

Network Systems Lab. Korea Advanced Institute of Science and Technology No.26 Some useful Contraction Mappings  Proposition 1.11