Convergence Theorems for Two Iterative Methods. A stationary iterative method for solving the linear system: (1.1)

Similar documents
A general N-dimensional vector consists of N values. They can be arranged as a column or a row and can be real or complex.

Having a glimpse of some of the possibilities for solutions of linear systems, we move to methods of finding these solutions. The basic idea we shall

UNIVERSITY OF IOANNINA DEPARTMENT OF ECONOMICS. M.Sc. in Economics MICROECONOMIC THEORY I. Problem Set II

Minimum Spanning Trees

Spanning Tree. Preview. Minimum Spanning Tree. Minimum Spanning Tree. Minimum Spanning Tree. Minimum Spanning Tree 10/17/2017.

H-matrix theory and applications

Fundamentals of Continuum Mechanics. Seoul National University Graphics & Media Lab

SAMPLE CSc 340 EXAM QUESTIONS WITH SOLUTIONS: part 2

Ch 1.2: Solutions of Some Differential Equations

PH427/PH527: Periodic systems Spring Overview of the PH427 website (syllabus, assignments etc.) 2. Coupled oscillations.

minimize c'x subject to subject to subject to

Math 656 March 10, 2011 Midterm Examination Solutions

Chapter Newton-Raphson Method of Solving a Nonlinear Equation

Filter Design Techniques

External Equivalent. EE 521 Analysis of Power Systems. Chen-Ching Liu, Boeing Distinguished Professor Washington State University

(A) the function is an eigenfunction with eigenvalue Physical Chemistry (I) First Quiz

Basic Electrical Engineering for Welding [ ] --- Introduction ---

Extension Formulas of Lauricella s Functions by Applications of Dixon s Summation Theorem

8. Linear Contracts under Risk Neutrality

GAUSS ELIMINATION. Consider the following system of algebraic linear equations

Preview. Graph. Graph. Graph. Graph Representation. Graph Representation 12/3/2018. Graph Graph Representation Graph Search Algorithms

INTEGRALS. Chapter 7. d dx. 7.1 Overview Let d dx F (x) = f (x). Then, we write f ( x)

Floating Point Number System -(1.3)

Floating Point Number System -(1.3)

Linear Prediction Analysis of

22.615, MHD Theory of Fusion Systems Prof. Freidberg Lecture 2: Derivation of Ideal MHD Equation

8. Computing Eigenvalues in Parallel

Walk Like a Mathematician Learning Task:

Chapter Newton-Raphson Method of Solving a Nonlinear Equation

Linear Prediction Analysis of Speech Sounds

Heisenberg Model. Sayed Mohammad Mahdi Sadrnezhaad. Supervisor: Prof. Abdollah Langari

The Mathematics of Harmonic Oscillators

MM1. Introduction to State-Space Method

Dennis Bricker, 2001 Dept of Industrial Engineering The University of Iowa. MDP: Taxi page 1

Principle Component Analysis

Lecture 4: Piecewise Cubic Interpolation

Partially Observable Systems. 1 Partially Observable Markov Decision Process (POMDP) Formalism

ENGI 3424 Engineering Mathematics Five Tutorial Examples of Partial Fractions

A Probabilistic Characterization of Simulation Model Uncertainties

Lecture 1. Functional series. Pointwise and uniform convergence.

GUC (Dr. Hany Hammad) 9/28/2016

Lecture 08 Multiple View Geometry 2. Prof. Dr. Davide Scaramuzza

The Hyperelastic material is examined in this section.

Inner Product Spaces INNER PRODUCTS

1 Introduction to Modulo 7 Arithmetic

A Solution for multi-evaluator AHP

Wave Phenomena Physics 15c

Minimum Spanning Trees

Linear Momentum. Center of Mass.

PHYS 2421 Fields and Waves

DCDM BUSINESS SCHOOL NUMERICAL METHODS (COS 233-8) Solutions to Assignment 3. x f(x)

Remember: Project Proposals are due April 11.

Lucas Test is based on Euler s theorem which states that if n is any integer and a is coprime to n, then a φ(n) 1modn.

STAT 309: MATHEMATICAL COMPUTATIONS I FALL 2018 LECTURE 16

1) They represent a continuum of energies (there is no energy quantization). where all values of p are allowed so there is a continuum of energies.

Special Random Variables: Part 1

First day August 1, Problems and Solutions

CSE 373: More on graphs; DFS and BFS. Michael Lee Wednesday, Feb 14, 2018

Phy213: General Physics III 4/10/2008 Chapter 22 Worksheet 1. d = 0.1 m

Section 3: Antiderivatives of Formulas

Bayesian belief networks: Inference

A New Refinement of Jacobi Method for Solution of Linear System Equations AX=b

4D SIMPLICIAL QUANTUM GRAVITY

Multi-Section Coupled Line Couplers

The University of Sydney MATH 2009

Jens Siebel (University of Applied Sciences Kaiserslautern) An Interactive Introduction to Complex Numbers

Math 3B Final Review

8-node quadrilateral element. Numerical integration

(2) If we multiplied a row of B by λ, then the value is also multiplied by λ(here lambda could be 0). namely

Grand Canonical Ensemble

Katholieke Universiteit Leuven Department of Computer Science

Linear Algebra. Definition The inverse of an n by n matrix A is an n by n matrix B where, Properties of Matrix Inverse. Minors and cofactors

STEADY STATE ERRORS IN UNITY FEEDBACK CONTROL SYSTEMS

Relate p and T at equilibrium between two phases. An open system where a new phase may form or a new component can be added

. Double-angle formulas. Your answer should involve trig functions of θ, and not of 2θ. cos(2θ) = sin(2θ) =.

Quantitative Genomics and Genetics BTRY 4830/6830; PBSB

IX. EMPIRICAL ORTHOGONAL FUNCTIONS

Constructing Free Energy Approximations and GBP Algorithms

ON THE COMPLEXITY OF K-STEP AND K-HOP DOMINATING SETS IN GRAPHS

Outline. 1 Introduction. 2 Min-Cost Spanning Trees. 4 Example

Statistics and Probability Letters

4.1 Interval Scheduling. Chapter 4. Greedy Algorithms. Interval Scheduling: Greedy Algorithms. Interval Scheduling. Interval scheduling.

ρ some λ THE INVERSE POWER METHOD (or INVERSE ITERATION) , for , or (more usually) to

Matrix Eigenvalues and Eigenvectors September 13, 2017

6 Roots of Equations: Open Methods

h Summary Chapter 7.

INTRODUCTION TO COMPLEX NUMBERS

M/G/1/GD/ / System. ! Pollaczek-Khinchin (PK) Equation. ! Steady-state probabilities. ! Finding L, W q, W. ! π 0 = 1 ρ

A Note on Estimability in Linear Models

Smart Motorways HADECS 3 and what it means for your drivers

Cycles and Simple Cycles. Paths and Simple Paths. Trees. Problem: There is No Completely Standard Terminology!

Last time: introduced our first computational model the DFA.

The Matrix Exponential

UNIFORM CONVERGENCE. Contents 1. Uniform Convergence 1 2. Properties of uniform convergence 3

On Hamiltonian Tetrahedralizations Of Convex Polyhedra

. Double-angle formulas. Your answer should involve trig functions of θ, and not of 2θ. sin 2 (θ) =

Variable time amplitude amplification and quantum algorithms for linear algebra. Andris Ambainis University of Latvia

Math 360: A primitive integral and elementary functions

6.6 The Marquardt Algorithm

total If no external forces act, the total linear momentum of the system is conserved. This occurs in collisions and explosions.

Transcription:

Conrgnc Thors for Two Itrt Mthods A sttonry trt thod for solng th lnr syst: Ax = b (.) ploys n trton trx B nd constnt ctor c so tht for gn strtng stt x of x for = 2... x Bx c + = +. (.2) For such n trton to conrg to th soluton x t ust b consstnt wth th orgnl lnr syst nd t ust conrg. To b consstnt w sply nd for x to b fxd pont tht s: x = Bx + c. (.3) Snc tht s qulnt to ( I Bx ) = c th consstnc condton cn b sttd ndpndnt of x by syng AI ( B) c= b. (.4) Th sst wy to dlop consstnt sttonry trt thod s to splt th trx A : A = M + N (.5) thn rwrt Ax = b s Th trton wll thn b Mx Mx= Nx+ b. (.6) + = Nx + b. (.7) Rcstng ths n th for bo w h B M = N nd c= M b. It s sy to show tht ths trton s consstnt for ny splttng s long s M s nonsngulr. Obously to b prctcl th trx M ust b slctd so tht th syst My = d s sly sold. Populr chocs for M r dgonl trcs (s n th Jcob thod) lowr trngulr trcs (s n th Guss-Sdl nd SOR thods) nd trdgonl trcs. Conrgnc: Thus constructng consstnt trtons s sy th dffcult ssu s constructng conrgnt consstnt trtons. Howr notc tht f s quton (.3) subtrctd fro quton (.2) w obtn whr s th rror x x. + = B Our frst rsult on conrgnc follows dtly fro ths. Thor : (.8)

Th sttonry trt thod for solng th lnr syst: x + = Bx + c for = 2... conrgs for ny ntl crtor ctor nor x f B < for so trx nor tht s consstnt wth Proof: Lt. b trx nor consstnt wth ctor nor. nd such tht B <. W thn h + = B B nd spl nduct rgunt shows tht n gnrl Snc B < ust conrg to zro (nd thus (.9) B. (.) x conrg to x ) ndpndnt of. Ths thor prods suffcnt condton for conrgnc. Wthout proof w offr ths thor tht prods both ncssry nd suffcnt condtons for conrgnc. It ploys th spctrl rdus of trx: ρ ( A) = th bsolut lu of th lrgst gnlu of A n bsolut lu. Thor 2: Th sttonry trt thod for solng th lnr syst: x + = Bx + c for = 2... conrgs for ny ntl ctor x f nd only f ( B) ρ <. Th sst wy to pro ths uss th Jordn Norl For of th trx B. Notc tht th thor dos not sy tht f ρ( B) th trton wll not conrg. It sys tht f ρ( B) th trton wll not conrg for so ntl ctor x. In prctcl trs though th dffrnc s nor: th only wy to h conrgnc wth ρ( B) s to h n ntl rror hng no coponnt n ny drcton of n gnctor of B corrspondng to n gnlu t lst on n bsolut lu. Ths s probblty zro nt. Th followng thor uss Thor to show th Jcob trton conrgs f th trx s strctly row dgonlly donnt. Rcll tht Jcob trton s + x = ( b x for = 2...n (.) )/ nd tht strct row dgonl donnc sys tht < for = 2... n. (.2)

Th splttng for th Jcob thod s A = D+ ( L+ U) whr DL nd U r th dgonl strct lowr trngl nd strct uppr trngl of th trx rspctly. Thus th trton trx s D ( L+ U). Thor 3: Th Jcob trt thod + x = ( b x for = 2... n )/ for solng th lnr syst Ax = b conrgs for ny ntl ctor strctly row dgonlly donnt. x f th trx A s Proof: Lt. ndct th nfnty ctor nor s wll s ts subordnt trx nor. To pro th thor t suffcs to show D ( L+ U) bsolut lus of th trx D ( L+ U). Ths r <. To tht nd consdr th row sus n but proprty (.2) gur- nts tht ths s strctly lss thn on. Th xu of th row sus n bsolut lu s lso strctly lss thn on so D ( L+ U) < s wll. Th nxt thor uss Thor 2 to show th Guss-Sdl trton lso conrgs f th trx s strctly row dgonlly donnt. Rcll tht Guss-Sdl trton s x b x x = (.3) + = ( + )/ for 2...n < > Th splttng for th Guss-Sdl thod s A= ( L+ D) + U. Thus th trton trx s ( L+ D) U.

Thor 4: Proof: Th Guss-Sdl trt thod x + = ( b x + x )/ for = 2... n < > for solng th lnr syst Ax= b conrgs for ny ntl ctor strctly row dgonlly donnt. Accordng to Thor 2 t suffcs to show gnctor corrspondng to n gnlu λ so = ρ( ( L+ D) U). W shll show x f th trx A s ρ( ( L+ D) U) <. To tht nd lt b ny λ of ( L+ D) U such λ < nd thus ρ ( (L+ D) U) <. W h U = λ( L + D) (.4) ( L+ D) U= λ. (.5) In coponnt fshon ths sys = λ. (.6) > Lt dnot n ndx of corrspondng to th lrgst coponnt n bsolut lu. Tht s = x{ } (.7) so W lso h for row n prtculr. (.8) > > = λ Ddng by th ncssrly post lus nd = λ + < λ < λ w h <

so But snc nd λ λ > > < < It s sy to show tht < t follows tht > < λ > < > = + < > > λ <. <. (.9) (.2) < x so th bound on th spctrl rdus trton trx of th Guss-Sdl thod s strctly lss thn th bound of th nfnty nor of th trton trx of th Jcob thod. Tht dos not gurnt tht th Guss-Sdl trton lwys conrgs fstr thn th Jcob trton. Howr t s oftn obsrd n prctc tht Guss-Sdl trton conrgs bout twc s fst s th Jcob trton. To s ths gn tht. Cll ths quntty 2 θ. W h θ > nd f θ s sll thn > < > < 4θ. Yt ( θ ) + ( θ) = 2θ nd f w gn for x 2 2 thn our bound for th nor of th Jcob trton trx s 2θ whl our bound on th spctrl rdus trton trx of th Guss-Sdl thod s 4θ.

Notc tht f th trton conrgs s to so tolrnc ε rqurs lu of σ for so fctor of bout σ thn to rduc ln ε. If lnσ σ thn ln ε ln ε lnε ln σ ( σ ) so w stt bout stps. Wth Jcob w h ( σ ) σ 2θ ln ε lnε but wth Guss-Sdl w h whch ustfs th cl tht Jcob conrgs twc s σ 4θ fst. Lstly wthout proof w stt nothr thor for conrgnc of th Guss-Sdl trton. Thor 5: Th Guss-Sdl trt thod x + = ( b x + x )/ for = 2... n < > for solng th lnr syst Ax sytrc nd post dfnt. = b conrgs for ny ntl ctor x f th trx A s