Numerical Linear Algebra - KOCWcontents.kocw.net › ... › 2016 › pusan › kimhyunmin ›...

Post on 05-Jul-2020

13 views 0 download

Transcript of Numerical Linear Algebra - KOCWcontents.kocw.net › ... › 2016 › pusan › kimhyunmin ›...

Schedule Prerequisite Preliminaries Errors and Algorithms

Numerical Linear Algebra

Kim, Hyun-Min

Department of Mathematics, Pusan National UniversityE-mail:hyunmin@pusan.ac.kr

Phone: 510-1060, 2596, 010-3833-8200Office: 612

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Schedule for Mid-term Exam

I Introduction to Numerical Linear Algebra

- What is the numerical linear algebra?- Why?- How?

I Linear Systems

- What are linear systems?- Existence and uniqueness of solution- Why?- How?

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

I Direct methods

- Gaussian elimination- LU, QR Factorizations- Special matrices

I Iterative Methods

- Norms of vectors and matrices- Iterative methods- Iterative methods for special matrices

I Decompositions

- Diagonalization- Jordan canonical form- Schur decompositions- Singular decomposition

I Mid-term Exam

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Linear Algebra

Math.Linear Algebra by S. Friedberg, A. Insel and L. SpenceChapters 1, 2, 3, 4, 5, 6, 7

Eng.Elementary Linear Algebra by H. Anton and C. RorresChapters 1, 2, 3, 4, 5, 6, 7, 8

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Linear Algebra

Math.Linear Algebra by S. Friedberg, A. Insel and L. SpenceChapters 1, 2, 3, 4, 5, 6, 7

Eng.Elementary Linear Algebra by H. Anton and C. RorresChapters 1, 2, 3, 4, 5, 6, 7, 8

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Mathematical Programming

Math.Mathematica, Maple, MATLAB

Eng.Fortran, C, MATLAB

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Mathematical Programming

Math.Mathematica, Maple, MATLAB

Eng.Fortran, C, MATLAB

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Numerical Analysis

I What is the mathematics?

I What is the numerical analysis?

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Definition

I The study of quantitative approximations to the solutions ofmathematical problems including consideration of the errorsand bounds to the errors involved.- Webster’s New Collegiate Dictionary (1973)

I The study of methods of approximation and their accuracy,etc.- Chambers 20th Century Dictionary (1983)

I The branch of mathematics concerned with the developmentand use of numerical methods for solving problems- Concise Oxford Dictionary 10th Edition (1999)

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

New Definition

I Numerical analysis is the study of rounding errors. bad one

I Numerical analysis is the study of algorithms for the problemsof continuous mathematics. good definition- Lloyd N. Trefethen (1993)

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

New Definition

I Numerical analysis is the study of rounding errors. bad one

I Numerical analysis is the study of algorithms for the problemsof continuous mathematics. good definition- Lloyd N. Trefethen (1993)

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Errors

Definitionx: the true valuex∗: an approximation to x

I |x− x∗|: Absolute Error

I|x− x∗||x|

(x 6= 0): Relative Error

Which one is better?

Why?

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Errors

Definitionx: the true valuex∗: an approximation to x

I |x− x∗|: Absolute Error

I|x− x∗||x|

(x 6= 0): Relative Error

Which one is better?

Why?

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Errors

Definitionx: the true valuex∗: an approximation to x

I |x− x∗|: Absolute Error

I|x− x∗||x|

(x 6= 0): Relative Error

Which one is better?

Why?

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Sources of Errors

I Errors in mathematical modelling: Simplifying andAssumptions

I Blunders (Prgramming Errors): Large programmes,Subprogrammes

I Errors in input: Errors in data transfer, uncertaintiesassociated with measurements

I Machine errors by computer (Floating point arithmetic):Rounding and Chopping, Underflow and Overflow

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Sources of Errors

I Errors in mathematical modelling: Simplifying andAssumptions

I Blunders

(Prgramming Errors): Large programmes,Subprogrammes

I Errors in input: Errors in data transfer, uncertaintiesassociated with measurements

I Machine errors by computer (Floating point arithmetic):Rounding and Chopping, Underflow and Overflow

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Sources of Errors

I Errors in mathematical modelling: Simplifying andAssumptions

I Blunders (Prgramming Errors): Large programmes,Subprogrammes

I Errors in input: Errors in data transfer, uncertaintiesassociated with measurements

I Machine errors by computer (Floating point arithmetic):Rounding and Chopping, Underflow and Overflow

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Sources of Errors

I Errors in mathematical modelling: Simplifying andAssumptions

I Blunders (Prgramming Errors): Large programmes,Subprogrammes

I Errors in input: Errors in data transfer, uncertaintiesassociated with measurements

I Machine errors by computer (Floating point arithmetic):Rounding and Chopping, Underflow and Overflow

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Sources of Errors

I Errors in mathematical modelling: Simplifying andAssumptions

I Blunders (Prgramming Errors): Large programmes,Subprogrammes

I Errors in input: Errors in data transfer, uncertaintiesassociated with measurements

I Machine errors by computer (Floating point arithmetic):Rounding and Chopping, Underflow and Overflow

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Arithmetic

In 1985 IEEE(Institute for Electrical and Electronic Engineers)report: Binary Floating Point Arithmetic Standard 754.

Single, Double and Extended Precisions

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Arithmetic

In 1985 IEEE(Institute for Electrical and Electronic Engineers)report: Binary Floating Point Arithmetic Standard 754.

Single, Double and Extended Precisions

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Algorithms

Examining approximation procedures involving finite sequenceof calculations.

I Stable: Small changes in the initial data

I Unstable: Otherwise

I Conditionally Stable: Stable only for certain of initial data

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Algorithms

Examining approximation procedures involving finite sequenceof calculations.

I Stable: Small changes in the initial data

I Unstable: Otherwise

I Conditionally Stable: Stable only for certain of initial data

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Algorithms

Examining approximation procedures involving finite sequenceof calculations.

I Stable: Small changes in the initial data

I Unstable: Otherwise

I Conditionally Stable: Stable only for certain of initial data

Kim, Hyun-Min Numerical Linear Algebra

Schedule Prerequisite Preliminaries Errors and Algorithms

Convergence

{αn}∞n=1, {βn}∞n=1: sequences

limn→∞

βn = 0, limn→∞

αn = α.

If ∃ K > 0 s. t. |αn−α| ≤ |βn| for large n then {αn}∞n=1 convergesto α with the rate of convergence O(βn).

αn = α+O(βn)

Kim, Hyun-Min Numerical Linear Algebra