Minimization of the quadratic test function

Similar documents
( ) 1 Comparison Functions. α is strictly increasing since ( r) ( r ) α = for any positive real number c. = 0. It is said to belong to

Lecture 24: Observability and Constructibility

2012 GCE A Level H2 Maths Solution Paper Let x,

THE ANALYSIS OF SOME MODELS FOR CLAIM PROCESSING IN INSURANCE COMPANIES

12.6 Sequential LMMSE Estimation

MATH 10550, EXAM 3 SOLUTIONS

On composite conformal mapping of an annulus to a plane with two holes

Using Difference Equations to Generalize Results for Periodic Nested Radicals

Conjugate Gradient Methods. Michael Bader. Summer term 2012

MATH Midterm Solutions

CHAPTER 5 : SERIES. 5.2 The Sum of a Series Sum of Power of n Positive Integers Sum of Series of Partial Fraction Difference Method

Technical Report: Bessel Filter Analysis

Multivector Functions

Minimal order perfect functional observers for singular linear systems

Introduction to Optimization Techniques. How to Solve Equations

A two-sided Iterative Method for Solving

Scientific Computing II

Iterative Techniques for Solving Ax b -(3.8). Assume that the system has a unique solution. Let x be the solution. Then x A 1 b.

SOLUTIONS TO EXAM 3. Solution: Note that this defines two convergent geometric series with respective radii r 1 = 2/5 < 1 and r 2 = 1/5 < 1.

By the end of this section you will be able to prove the Chinese Remainder Theorem apply this theorem to solve simultaneous linear congruences

Greatest term (numerically) in the expansion of (1 + x) Method 1 Let T

Z - Transform. It offers the techniques for digital filter design and frequency analysis of digital signals.

The Discrete Fourier Transform

Finite q-identities related to well-known theorems of Euler and Gauss. Johann Cigler

( ) ( ) ( ) ( ) Solved Examples. JEE Main/Boards = The total number of terms in the expansion are 8.

Received 17 August 2015; accepted 22 September 2015; published 25 September 2015

15.081J/6.251J Introduction to Mathematical Programming. Lecture 21: Primal Barrier Interior Point Algorithm

INVERSE CAUCHY PROBLEMS FOR NONLINEAR FRACTIONAL PARABOLIC EQUATIONS IN HILBERT SPACE

THE ANALYTIC LARGE SIEVE

Math 166 Week-in-Review - S. Nite 11/10/2012 Page 1 of 5 WIR #9 = 1+ r eff. , where r. is the effective interest rate, r is the annual

Solution of Linear Constant-Coefficient Difference Equations

KEY. Math 334 Midterm II Fall 2007 section 004 Instructor: Scott Glasgow

BINOMIAL THEOREM An expression consisting of two terms, connected by + or sign is called a

BINOMIAL THEOREM NCERT An expression consisting of two terms, connected by + or sign is called a

x 1 b 1 Consider the midpoint x 0 = 1 2

Differentiable Convex Functions

v = -!g(x 0 ) Ûg Ûx 1 Ûx 2 Ú If we work out the details in the partial derivatives, we get a pleasing result. n Ûx k, i x i - 2 b k

Advanced Higher Formula List

MATH /19: problems for supervision in week 08 SOLUTIONS

PROBLEM SET 5 SOLUTIONS 126 = , 37 = , 15 = , 7 = 7 1.

IN many scientific and engineering applications, one often

= 5! 3! 2! = 5! 3! (5 3)!. In general, the number of different groups of r items out of n items (when the order is ignored) is given by n!

DETERMINATION OF MECHANICAL PROPERTIES OF A NON- UNIFORM BEAM USING THE MEASUREMENT OF THE EXCITED LONGITUDINAL ELASTIC VIBRATIONS.

Math 220B Final Exam Solutions March 18, 2002

Notes on iteration and Newton s method. Iteration

FIXED POINT AND HYERS-ULAM-RASSIAS STABILITY OF A QUADRATIC FUNCTIONAL EQUATION IN BANACH SPACES

Section A assesses the Units Numerical Analysis 1 and 2 Section B assesses the Unit Mathematics for Applied Mathematics

The Method of Least Squares. To understand least squares fitting of data.

CHAPTER 10 INFINITE SEQUENCES AND SERIES

EDEXCEL NATIONAL CERTIFICATE UNIT 28 FURTHER MATHEMATICS FOR TECHNICIANS OUTCOME 2- ALGEBRAIC TECHNIQUES TUTORIAL 1 - PROGRESSIONS

The picture in figure 1.1 helps us to see that the area represents the distance traveled. Figure 1: Area represents distance travelled

CS 270 Algorithms. Oliver Kullmann. Growth of Functions. Divide-and- Conquer Min-Max- Problem. Tutorial. Reading from CLRS for week 2

INTRODUCTORY MATHEMATICAL ANALYSIS

Some Properties of the K-Jacobsthal Lucas Sequence

Chapter 4. Fourier Series

AIEEE 2004 (MATHEMATICS)

Topic 5 [434 marks] (i) Find the range of values of n for which. (ii) Write down the value of x dx in terms of n, when it does exist.

Some Integral Mean Estimates for Polynomials

Find a formula for the exponential function whose graph is given , 1 2,16 1, 6

a) The average (mean) of the two fractions is halfway between them: b) The answer is yes. Assume without loss of generality that p < r.

-ORDER CONVERGENCE FOR FINDING SIMPLE ROOT OF A POLYNOMIAL EQUATION

ECE-S352 Introduction to Digital Signal Processing Lecture 3A Direct Solution of Difference Equations

Markov Decision Processes

Strong Result for Level Crossings of Random Polynomials. Dipty Rani Dhal, Dr. P. K. Mishra. Department of Mathematics, CET, BPUT, BBSR, ODISHA, INDIA

Einstein Classes, Unit No. 102, 103, Vardhman Ring Road Plaza, Vikas Puri Extn., Outer Ring Road New Delhi , Ph. : ,

Counting Functions and Subsets

Other quantum algorithms

Consider unordered sample of size r. This sample can be used to make r! Ordered samples (r! permutations). unordered sample

Generating Functions. 1 Operations on generating functions

AP Calculus BC Review Applications of Derivatives (Chapter 4) and f,

Crosscorrelation of m-sequences, Exponential sums and Dickson

METHOD OF FUNDAMENTAL SOLUTIONS FOR HELMHOLTZ EIGENVALUE PROBLEMS IN ELLIPTICAL DOMAINS

Lyapunov Stability Analysis for Feedback Control Design

10.1 Sequences. n term. We will deal a. a n or a n n. ( 1) n ( 1) n 1 2 ( 1) a =, 0 0,,,,, ln n. n an 2. n term.

Math 113, Calculus II Winter 2007 Final Exam Solutions

Chapter 2 Sampling distribution

Integrable Functions. { f n } is called a determining sequence for f. If f is integrable with respect to, then f d does exist as a finite real number

September 2012 C1 Note. C1 Notes (Edexcel) Copyright - For AS, A2 notes and IGCSE / GCSE worksheets 1

E a 0,a 1,...,a k y i P k x i 2

Solving Fuzzy Differential Equations using Runge-Kutta third order method with modified contra-harmonic mean weights

NICK DUFRESNE. 1 1 p(x). To determine some formulas for the generating function of the Schröder numbers, r(x) = a(x) =

Problem. Consider the sequence a j for j N defined by the recurrence a j+1 = 2a j + j for j > 0

Most text will write ordinary derivatives using either Leibniz notation 2 3. y + 5y= e and y y. xx tt t

f(x i ; ) L(x; p) = i=1 To estimate the value of that maximizes L or equivalently ln L we will set =0, for i =1, 2,...,m p x i (1 p) 1 x i i=1

Lecture 20. Curve fitting II. m y= f (x)= c j. (x) (1) f j. n [ y i )] 2 (2) f (x i. MSE= 1 n i =1

Ma 530 Introduction to Power Series

Recursion. Algorithm : Design & Analysis [3]

2.4 - Sequences and Series

Taylor expansion: Show that the TE of f(x)= sin(x) around. sin(x) = x - + 3! 5! L 7 & 8: MHD/ZAH

Auchmuty High School Mathematics Department Sequences & Series Notes Teacher Version

Reccurent sequenses in solving the Schrödinger equation

SVD ( ) Linear Algebra for. A bit of repetition. Lecture: 8. Let s try the factorization. Is there a generalization? = Q2Λ2Q (spectral theorem!

A widely used display of protein shapes is based on the coordinates of the alpha carbons - - C α

Recurrence Relations

Løsningsførslag i 4M

Advances in Mathematics and Statistical Sciences. On Positive Definite Solution of the Nonlinear Matrix Equation

Using Counting Techniques to Determine Probabilities

NEW FAST CONVERGENT SEQUENCES OF EULER-MASCHERONI TYPE

Definitions and Theorems. where x are the decision variables. c, b, and a are constant coefficients.

Mathematics 3 Curs /Q1 - First exam. 30/10/14 Group M1 Lecturer: Yolanda Vidal

Transcription:

Miimizatio of the quadatic test fuctio A quadatic fom is a scala quadatic fuctio of a vecto with the fom f ( ) A b c with b R A R whee A is assumed to be SPD ad c is a scala costat Note: A symmetic mati A is SPD (symmetic positive defiite) if A > o equivaletly if all the eigevalues of A ae positive he miimize of the fuctio f is give as the poit whee the gadiet of the fuctio is equal to zeo Diect calculatio gives the optimality coditios: f A b he statioay poits ae obtaied by solvig the liea system A b Note: his shows that the same that miimizes f() also seves as the solutio to the liea system of equatios A b ad the uiqueess of the solutio is guaateed by the SPD coditio he Method of Steepest Descet he method of steepest descet geeates a sequece of iteates appoimatig a local miimize of f() At iteatio a give poit is assumed to be a appoimatio of the eact solutio he eo is give by the esidual vecto: A b We ae at the poit How do we each? he idea is to take the diectio i which f() deceases most quickly (ie the gadiet) so the diectio of steepest descet is give by: d -f( ) ( A b) his is the easo fo the ame steepest descet o gadiet method hus we seach fo a bette solutio i the diectiod ie we have to detemie the coefficiet R such that the value of the eal fuctio he miimum of the fuctio h is eached at h( ) f( f ( )) mi A ad thus the et appoimatio of the solutio is give by mif We epeat this fo evey step takig the gadiet of f () i the et poit ad by fidig a ew step legth Eample:

We apply the Method of Steepest Descet to the fuctio f ( ) 4 4 with iitial poit X We fist compute the steepest descet diectio fom 8 4 f ( X ) 4 4 to obtai 4 f ( X ) f 4 We the miimize the fuctio 4 h( ) f(x f ( X )) f( ) f ( 4 4 ) 4 h( ) 4(-4 ) 4(-4 )(-4 ) (-4 ) by computig h ( ) 64 - his fuctio has a global miimum wheh( ) o 5 We theefoe set 4 X 5 4 Cotiuig the pocess we have 4 f ( X) f 4 ad by defiig - 4 h( ) f(x f ( X)) f( ) f (4 4 ) 4 h( ) 4(4 ) 4(4 )(-4 ) (-4 ) we obtai h ( ) - We have h( ) whe We theefoe set - 4 4 X 4 6 Repeatig this pocess yields X We ca see that the Method of Steepest Descet poduces a sequece of iteates X that is covegig to the stict global miimize of f ) at ) () ( (

Algoithm of Steepest Descet: o summaize we have obtaied a iteative algoithm fo steepest descet with the followig update step: Algoithm Stat with a abitay iitial guess to the solutio to Fo i b A( ) (a) ( i) i ( i) ( i) ( i) (b) A ( i) ( i) ( i ) ( i) ( i) ( i) (c) A b Eample: Pefom two iteatios of the steepest descet method statig fom the oigi towads fidig a miimize ad also detemie the optimal solutio aalytically he fuctio of iteest is f ( ) Solutio: f ca be epessed as f ) X AX b X c whee ( A I fact we kow that the miimize is X A b ad b / c / / / / 4 We will ow test the steepest descet algoithm at least the fist two steps theeof he mati A is positive defiite ad symmetic Fo steepest descet we set the statig poit: X he gadiet is f ( X ) / AX - b Now fo the fist step of the algoithm we fid f ( X ) / so

he Repeatig the pocess we fid so ad / 5 / 4 / / A 5 5/ 6 X X - 6 / 5/ /6 f ( X) / / 6 / 6 / / / 6 / 6 / / A 5 6 5 9 5/ 6 5 / 6 5/ 7 X X - 5/ 9 / 5/8 Eample: Coside the system 5 A 5 b 5 Fid the appoimate solutios of the liea system usig the Steepest Descet method with Solutio: he mati A is positive defiite ad symmetic Fo the Steepest Descet Method fist set the statig poit ad the compute the steepest descet diectio A b b So the we set 6 A

6 Repeatig the pocess we fid 4 A b ad 5 8 A the we have 7 45 54

his is a collectio of outies compaig diffeet iteative schemes fo appoimatig the solutio of a system of liea equatios Iteative methods Jacobi method he followig M-file shows how to use Jacobi method i MALAB: (Jacobi_with_tolm) fuctio [it]=jacobi_with_tol(abmaitetol) Jacobi_with_tolm solves the liea system A=b usig the Jacobi Method A mati iitial guess vecto b ight had side vecto maite maimum umbe of iteatios tol eo toleace solutio vecto it umbe of iteatios pefomed ele = if; Relative eo it = ; D = diag(diag(a)); N = D - A; = ; while (ele > tol) & (it< maite) it = it+; _ew = D \ (N*+b); ele = om(_ew-if)/om(_ewif); =_ew; ed Ruig M-file i commad widow: he followig MALAB commads compute the fist (it) Jacobi iteatios with iitial guess = : Fist we to defie all iput paamete: >> A=[7 - ; 8-4; - 4 -; -4-6] >> b= [-;;-;]; >> eact=a\b he eact solutio

eact = - - >> =[;;;] >> tol=^-5; >> maite=; >> [it]=jacobi_with_tol(abmaitetol) he output: he appoimate solutio ad the umbe of iteatios pefomed = -9999 9999-9999 it = 7 Iteative methods - Gauss-Seidel method he followig M-file shows how to use Gauss-Seidl method i MALAB: (Gauss_Seidl _tolm) fuctio [it]=gauss_seidel_tol(abmaitetol) Gauss_Seidel_tolm solves the liea system A=b usig the Gauss- Seidel Method A mati iitial guess vecto b ight had side vecto maite maimum umbe of iteatios tol eo toleace solutio vecto it umbe of iteatios pefomed

ele = if; Relative eo M = til(a); M = L + D N = M - A; N = -U = M - A = ; it=; while (ele > tol) & (it< maite) it = it+; _ew = M \ (N*+b); ele = om(_ew-if)/om(_ewif); =_ew; ed he followig MALAB commads compute the fist (it) Gauss-Seidl iteatios with iitial guess = fo the pevious eample: >> [it] =Gauss_Seidel_tol(Abtol) he output: he appoimate solutio ad the umbe of iteatios pefomed = - - it = 4 Iteative methods Steepest descet o gadiet method he followig M-file shows how to use Steepest descet o gadiet method i MALAB: (gadm) fuctio [it]= gad(abmaitetol) Steepest descet o Gadiet method Gauss_Seidel_tolm solves the liea system A=b usig the Gauss- Seidel Method A mati iitial guess vecto b ight had side vecto maite maimum umbe of iteatios tol eo toleace solutio vecto

it umbe of iteatios pefomed = A\b; Eact solutio ele = if; Relative eo = ; it=; while (ele > tol) & (it< maite) = A* - b; = - (om()^ / dot(a*)) * ; ele = om(-if)/om(if); it=it+; ed he output: he appoimate solutio ad the umbe of iteatios pefomed >> [it]= gad(abmaitetol) = - - it = 65 Iteative methods Summay Method Solutio otal Iteatio umbe Jacobi method ( -9999 9999-9999) 7 Gauss Seidel method (- - ) 4 Gadiet method (- - ) 65