ME 310 Numerical Methods. Solving Systems of Linear Algebraic Equations



Similar documents
Newton-Raphson Method of Solving a Nonlinear Equation Autar Kaw

This circuit than can be reduced to a planar circuit

Vector Geometry for Computer Graphics

Multiplication and Division - Left to Right. Addition and Subtraction - Left to Right.

Line-source based X-ray Tomography

Resistive Network Analysis. The Node Voltage Method - 1

AREA OF A SURFACE OF REVOLUTION

Graphs on Logarithmic and Semilogarithmic Paper

Unit 6: Exponents and Radicals

FI GERPRI T VERIFICATIO USI G OPTICAL SHIFTED PHASE- E CODED JOI T TRA SFORM CORRELATIO

Use Geometry Expressions to create a more complex locus of points. Find evidence for equivalence using Geometry Expressions.

WHAT HAPPENS WHEN YOU MIX COMPLEX NUMBERS WITH PRIME NUMBERS?

e.g. f(x) = x domain x 0 (cannot find the square root of negative values)

Operations with Polynomials

n + d + q = 24 and.05n +.1d +.25q = 2 { n + d + q = 24 (3) n + 2d + 5q = 40 (2)

Mathematics. Vectors. hsn.uk.net. Higher. Contents. Vectors 128 HSN23100

Vectors. The magnitude of a vector is its length, which can be determined by Pythagoras Theorem. The magnitude of a is written as a.

So far circuit analysis has been performed on single-

Or more simply put, when adding or subtracting quantities, their uncertainties add.

Calculation of Sampling Weights

Linear Circuits Analysis. Superposition, Thevenin /Norton Equivalent circuits

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by

Polynomial Functions. Polynomial functions in one variable can be written in expanded form as ( )

Recurrence. 1 Definitions and main statements

WiMAX DBA Algorithm Using a 2-Tier Max-Min Fair Sharing Policy

PROF. BOYAN KOSTADINOV NEW YORK CITY COLLEGE OF TECHNOLOGY, CUNY

Appendix D: Completing the Square and the Quadratic Formula. In Appendix A, two special cases of expanding brackets were considered:

Lecture 2: Single Layer Perceptrons Kevin Swingler

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

1. Introduction to CFD

Reasoning to Solve Equations and Inequalities

Lecture 3 Gaussian Probability Distribution

LECTURE #05. Learning Objective. To describe the geometry in and around a unit cell in terms of directions and planes.

Factoring Polynomials

Section 5.4 Annuities, Present Value, and Amortization

Algebra Review. How well do you remember your algebra?

21 Vectors: The Cross Product & Torque

Helicopter Theme and Variations

Math 135 Circles and Completing the Square Examples

Irregular Repeat Accumulate Codes 1

Peak Inverse Voltage

Boolean Algebra. ECE 152A Winter 2012

Treatment Spring Late Summer Fall Mean = 1.33 Mean = 4.88 Mean = 3.

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Lecture 5. Inner Product

Lectures 8 and 9 1 Rectangular waveguides

EN3: Introduction to Engineering. Teach Yourself Vectors. 1. Definition. Problems

SIMPLE LINEAR CORRELATION

Version 001 Summer Review #03 tubman (IBII ) 1

Texas Instruments 30X IIS Calculator

ORIGIN DESTINATION DISAGGREGATION USING FRATAR BIPROPORTIONAL LEAST SQUARES ESTIMATION FOR TRUCK FORECASTING

(6)(2) (-6)(-4) (-4)(6) + (-2)(-3) + (4)(3) + (2)(-3) = = 0

The OC Curve of Attribute Acceptance Plans

THE METHOD OF LEAST SQUARES THE METHOD OF LEAST SQUARES

2 DIODE CLIPPING and CLAMPING CIRCUITS

Finite difference method

THE FLEXURE AND SHEAR DESIGN OF CORBEL (BRACKET)

Simple Interest Loans (Section 5.1) :

Vectors Recap of vectors

Using Series to Analyze Financial Situations: Present Value

Experiment 6: Friction

A Hadoop Job Scheduling Model Based on Uncategorized Slot

Fuzzy Clustering for TV Program Classification

Regular Sets and Expressions

What is Candidate Sampling

Causal, Explanatory Forecasting. Analysis. Regression Analysis. Simple Linear Regression. Which is Independent? Forecasting

Sequences and Series

Exponential and Logarithmic Functions

Basic Analysis of Autarky and Free Trade Models

1 Example 1: Axis-aligned rectangles

NPAR TESTS. One-Sample Chi-Square Test. Cell Specification. Observed Frequencies 1O i 6. Expected Frequencies 1EXP i 6

SPECIAL PRODUCTS AND FACTORIZATION

Integration. 148 Chapter 7 Integration

Optimal Pricing Scheme for Information Services

Types of Injuries. (20 minutes) LEARNING OBJECTIVES MATERIALS NEEDED

BERNSTEIN POLYNOMIALS

Faraday's Law of Induction

P.3 Polynomials and Factoring. P.3 an 1. Polynomial STUDY TIP. Example 1 Writing Polynomials in Standard Form. What you should learn

Chem 350 Jasperse Ch. 3 Handouts 1

LINEAR TRANSFORMATIONS AND THEIR REPRESENTING MATRICES

5.6 POSITIVE INTEGRAL EXPONENTS

Solution: Let i = 10% and d = 5%. By definition, the respective forces of interest on funds A and B are. i 1 + it. S A (t) = d (1 dt) 2 1. = d 1 dt.

Singularity-Free Dynamic Modeling Including Wheel Dynamics for an Omni-Directional Mobile Robot with Three Caster Wheels

1. Math 210 Finite Mathematics

Section 5.3 Annuities, Future Value, and Sinking Funds

3. Present value of Annuity Problems

An Alternative Way to Measure Private Equity Performance

8.4. Annuities: Future Value. INVESTIGATE the Math Annuities: Future Value

Econ 4721 Money and Banking Problem Set 2 Answer Key

9 CONTINUOUS DISTRIBUTIONS

Second Term MAT2060B 1. Supplementary Notes 3 Interchange of Differentiation and Integration

MA Lesson 16 Notes Summer 2016 Properties of Logarithms. Remember: A logarithm is an exponent! It behaves like an exponent!

Small Businesses Decisions to Offer Health Insurance to Employees

Implementation of Deutsch's Algorithm Using Mathcad

Review Problems for the Final of Math 121, Fall 2014

Transcription:

ME 3 Nmercl Methods Solng Sstems of Lner Algebrc Eqtons These presenttons re prepred b Dr. Cnet Sert Mechncl Engneerng Deprtment Mddle Est Techncl Unerst Ankr, Trke csert@met.ed.tr The cn not be sed wthot the permsson of the thor

Mth Introdcton f (,,..., n ) = f (,,..., n ) =... A generl set of eqtons. n eqtons, n nknowns. f n (,,..., n ) = Lner Algebrc Eqton: An eqton of the form f()= where f s polnoml wth lner terms. + +... + n n = b + +... + n n = b............ n + n +... + nn n = b n A generl set of lner lgebrc eqtons. n eqtons, n nknowns. Mtr Form: [A] {} = {b} [A] nn Coeffcent mtr {} n Unknown ector {b} n Rght-Hnd-Sde (RHS) ector

Reew of Mtrces [A] n n m m nm n m nd row Elements re ndcted b j row colmn Row ector: m th colmn r r rn n [ R] Colmn ector: [C] c c c m m Sqre mtr: [A] nm s sqre mtr f n=m. A sstem of n eqtons wth n nknonws hs sqre coeffcent mtr. Mn (prncple) dgonl: of [A] nn conssts of elements ; =,...,n Smmetrc mtr: If j = j [A] nn s smmetrc mtr Dgonl mtr: [A] nn s dgonl f j = for ll =,...,n ; j=,...,n nd j Identt mtr: [A] nn s n dentt mtr f t s dgonl wth = =,...,n. Shown s [I] 3

Reew of Mtrces (cont d) Upper trnglr mtr: [A] nn s pper trnglr f j = =,...,n ; j=,...,n nd >j Lower trnglr mtr: [A] nn s lower trnglr f j = =,...,n ; j=,...,n nd <j Bnded mtr: [A] nn s bnded mtr f j = for ll j > HBW or j > HBW, where HBW s the hlf bndwdth. Trdgonl mtr: s bnded mtr wth HBW=. Inerse of mtr: [A] - s the nerse of [A] nn f [A] - [A]=[I] Trnspose of mtr: [B] s the trnspose of [A] nn f b j = j Shown s [A] or [A] T Mtr mltplcton: [C] ps = [A] pr [B] rs c b =,...,p ; j=,...,s Note: [A][B] [B][A] j r k k kj Agmented mtr: s specl w of showng two mtrces together. b For emple A gmented wth the colmn ector B s b b b Determnnt of mtr: s sngle nmber. Determnnt of [A] s shown s A. 4

Grphcl Method for Solng Smll (n3) Set of Eqtons Consder set of eqtons + = b + = b Plot these on the Crtesn coordnte sstem wth es nd. b b For n=3, ech eqton wll be plne on 3D coordnte sstem. Solton s the pont where these plnes ntersect. For n>3, grphcl solton s not prctcl. 5

Grphcl Method (cont d) Grphcl Method s sefl to llstrte: () no solton () nfntel mn soltons (3) ll-condtoned sstem In sstem () nd (), eqtons re lnerl dependent. In sstem (3), the slopes of the lnes re er close to ech other. Mthemtcll Coeffcent mtrces of () & () re snglr. Ther determnnts re zero nd ther nerse do not est. Coeffcent mtr of (3) s lmost snglr. Its nerse s dffclt to tke. Ths sstem hs nqe solton, whch s not es to determne nmercll becse of ts etreme senstt to rond-off errors. 6

Crmer s Rle for Solng Set of Eqtons Determnnt of sstem Determnnt of mtr s D Determnnt of 33 mtr s D 3 3 33 3 3 33 3 3 3 Determnnt of generl nn mtr cn be clclted recrsel sng the boe pttern. Determnnt of snglr sstem s zero. - 3 = 5-6 = 7 Determnnt of n lll-condtoned sstem s close to zero. - 3 = 5 3.98-6 = 7 Wrnng: Mltpl both eqtons of the boe sstem wth 3 = 5 398-6 = 7 Ths sstem s s ll-condtoned s the preos one bt t hs determnnt tmes lrger. Therefore we need to scle sstem when we tlk bot the mgntde of ts determnnt. Detls wll come lter. 7

Crmer s Rle for Solng Set of Eqtons (Cont d) Crmer s Rle: Ech nknown s clclted s frcton of two determnnts. The denomntor s the determnnt of the sstem, D. The nmertor s the determnnt of modfed sstem obtned b replcng the colmn of coeffcents of the nknown beng clclted b the RHS ector. For 33 sstem b b 3 3 b3 3 33 3 b3 33 3 D D b b 3 3 3 D 3 b b b 3 For snglr sstem D= Solton cn not be obtned. For lrge sstems Crmer s rle s not prctcl becse clcltng determnnts s costl. To sole n nn sstem of eqtons, Crmer s rle needs n+ determnnt eltons. Usng recrse lgorthm, determnnt of n nn mtr reqres n!+n- rthmetc opertons (+,-,, ). Therefore solng n nn sstem wth the Crmer s Rle needs more thn (n+)(n!+n-). For sstem ths mens bot opertons (Check these nmbers). 8

C nd Mtrces: A mtr s nothng bt n rr of rrs. In C rr ndces strt from. C does not do check rr bonds. Complers mght he swtches for ths. A D rr ( mtr) s declred s doble A[4][5] nd ts elements re reched s A[][j] If o do not know the sze of the mtrces before strtng the progrm, o need to se dnmc memor llocton wth ponters. Ths cn be qte confsng. Eercse : () Wrte two fnctons to dd nd mltpl two mtrces. These fnctons shold tke two mtrces nd ther szes s npt rgments nd clclte thrd mtr. The shold check f the npt mtrces re stble for ddton or mlplcton. () Wrte mn progrm. Generte two smll mtrces. Add nd mltpl them b cllng proper fnctons. Eercse : () Wrte fncton tht clcltes the determnnt of mtr. Cll ths fncton recrsel to sole sstems of eqtons sng the Crmer s Rle. () Use ths progrm to sole set of eqtons. Generte the coeffcent mtr of the sstem nsde loop. Mke sre tht or sstem s not snglr. How long t tkes to sole ths sstem? Tr solng lrger sstems. 9

Elmnton of Unknowns Method Emple : Gen set of eqtons:.5 + 6. = 3. 4.8-8.6 = 5.5 Mltpl the st eqn b 8.6 nd the nd eqn b 6.:.5 + 53.3 = 5.8 9.76 53.3 = 34. Add these eqtons: 5.6 + = 59.9 Sole for : = 59.9/5.6 =.6855478 Usng the st eqn sole for : = (3..5*.6855478)/6. =.68454 Check f these stsf the nd eqn: 4.8*.6855478 8.6*.68454 = 5.54 (Dfference s de to the rond-off errors).

Ne Gss Elmnton Method It s formlzed w of the preos elmnton technqe. Consder the followng sstem of n eqtons. + +... + n n = b () + +... + n n = b ()... n + n +... + nn n = b n (n) Step (optonl): Form the gmented mtr of [A B]. Step Forwrd Elmnton: Redce the sstem to n pper trnglr sstem. (.) Frst elmnte from nd to n th eqtons. - Mltpl the st eqn. b / & sbtrct t from the nd eqton. Ths s the new nd eqn. - Mltpl the st eqn. b 3 / & sbtrct t from the 3 rd eqton. Ths s the new 3 rd eqn.... - Mltpl the st eqn. b n / & sbtrct t from the n th eqton. Ths s the new n th eqn. Importnt: In these steps the st eqn s the pot eqton nd s the pot element. Note tht dson b zero m occr f the pot element s zero. Ne-Gss Elmnton does not check for ths.

Ne Gss Elmnton Method (cont d) (.) Now elmnte from 3 rd to n th eqtons. ndctes tht the sstem s modfed once. The modfed sstem s n 3 n 3 nn n3 n 3n 33 3 n 3 n 3 b b b b ndctes tht the sstem s modfed once. The modfed sstem s n 3 n 3 nn n3 3n 33 n 3 n 3 b b b b Repet (.) nd (.) pto (.n-). b nn 3n 33 n 3 n 3 Prmes re remoed for clrt. At the end of Step, we wll get ths pper trnglr sstem

Ne Gss Elmnton Method (cont d) Step Bck sbsttton: Fnd the nknowns strtng from the lst eqton. (.) Lst eqton noles onl n. Sole for t. (.) Use ths n n the (n-) th eqton nd sole for n-.... (.n) Use ll preosl clclted les n the st eqn nd sole for. Emple : Sole the followng sstem sng Ne Gss Elmnton. 6 + 3 + 4 4 = 6 8 + 6 3 + 4 = 6 3 3 + 9 3 + 3 4 = -9-6 + 4 + 3-8 4 = -34 Step : Form the gmented mtr 6 4 6 8 6 6 3 3 9 3-9 -6 4-8 -34 3

Ne Gss Elmnton Method (cont d) Step : Forwrd elmnton (.) Elmnte 6 4 6 (Does not chnge. Pot s 6) 4-6 8-7 3-4 -8 (.) Elmnte 6 4 6 (Does not chnge.) 4-6 (Does not chnge. Pot s -4) -5-9 4-3 - (.3) Elmnte 3 6 4 6 (Does not chnge.) 4-6 (Does not chnge.) -5-9 (Does not chnge. Pot s ) -3-3 Step : Bck sbsttton (.) Fnd 4 4 = (-3)/(-3) = (.) Fnd 3 3 = (-9+5*)/ = - (.3) Fnd = (-6-*(-)-*)/(-4) = (.4) Fnd = (6+*-*(-)-4*)/6 = 3 4

Psedocode for the Ne Gss Elmnton Method For generl nn sstem [A] {} = {B} Forwrd Elmnton LOOP k from to n- LOOP from k+ to n FACTOR = A k / A kk LOOP j from k+ to n A j = A j FACTOR * A kj END LOOP B = B FACTOR * B k ENDLOOP ENDLOOP Bck Sbsttton X n = B n / A nn LOOP from n- to SUM =. LOOP j from + to n SUM = SUM + A j * X j END LOOP X = (B SUM) / A ENDLOOP Eercse 3: Implement ths n C. Wrte mn progrm nd two fnctons. Yo need to know how to pss mtrces to the fnctons. 5

Operton Cont for the Ne Gss Elmnton Method Let s cont the FLOPS (flotng pont opertons). Consder onl mltplctons nd dsons, snce the re more epense to perform. Forwrd Elmnton FACTOR = A k / A kk s elted (n k) tmes. A j = A j FACTOR * A kj s elted tmes. B = B FACTOR * B k s elted tmes. Totl * nd / Bck Sbsttton SUM = SUM + A j * X j s elted (n ) tmes. X = (B SUM) / A s elted tmes. Totl * nd / Totl n 3 /3 + n / + O(n ) + O(n) s n ncreses n 3 /3 n k n (n k)(n k ) (n ) n 3 n 3 O(n) O(n n k n k n k ) n n (n k)(n k) (n k) Emple 3: n * & / n FE * & / n BS Totl * & / n 3 /3 % de to FE 375 55 43 333 87. % 3385 55 3433 333333 98.5 % 3.34 E8 55 3.34 E8 3.33 E8 99.8 % 6

Potng In Ne Gss Elmnton, dson b zero occrs f the pot element s zero. Note tht zero pot elements m be creted drng the elmnton step een f the re not present n the orgnl mtr. Potng s sed to od ths problem. We nterchnge rows nd colmns t ech step to pt the coeffcent wth the lrgest mgntde on the dgonl. In ddton to odng the dson b zero problem, potng redces the rond-off errors. It mkes the solton of ll-condtoned sstems eser. Complete potng ses both row nd colmn nterchnges. It s not sed freqentl. Prtl potng ses onl row nterchnges. We wll se ths. When there re lrge dfferences n mgntde of coeffcents n one eqton compred to the other eqtons we m lso need sclng. Detls wll come lter. 7

Potng Emple Emple 4: Sole the followng sstem sng Gss Elmnton wth potng. + + 4 = + + 3 3 + 4 = - 4 3 + 4 = -7 6 + - 6 3-5 4 = 6 Step : Form the gmented mtr 3-4 3-7 6-6 -5 6 Step : Forwrd Elmnton (.) Elmnte. Bt the pot element s. We he to nterchnge the st row wth one of the rows below t. Interchnge t wth the 4 th row becse 6 s the lrgest possble pot. 6-6 -5 6 6-6 -5 6 3 - Now elmnte.6667 5 3.6667-4 4 3-7 -3.6667 4 4.3333-8

Potng Emple (cont d) (.) Elmnte.from the 3 rd nd 4 th eqns. Pot element s.6667. There s no dson b zero problem. Stll we wll perform potng to redce rond-off errors. Interchnge the nd nd 3 rd rows. Note tht complete potng wold nterchnge nd nd 3 rd colmns. 6-6 -5 6 6-6 -5 6-3.6667 4 4.3333 - Elmnte -3.6667 4 4.3333 -.6667 5 3.6667-4 6.88 5.6364-9..88 3.3636-5.9999 (.3) Elmnte 3. 6.88 >.88, therefore no potng s necessr. 6-6 -5 6-3.6667 4 4.3333-6.88 5.6364-9..56-3.99 Step : Bck sbsttton 4 = -3.99 /.56 = -.9999 3 = [-9. 5.6364*(-.9999)] / 6.88 =.3335 = [- 4.3333*(-.9999) 4*.3335] / -3.6667 =. = [6 (-5)*(-.9999) (-6)*.3335 *.] / 6 = -.5 Ect solton s = [- /3 -.5] T. Use more thn 5 sg. fgs. to redce rond-off errors. 9

Sclng Normlze the eqtons so tht the mmm coeffcent n eer row s eql to.. Tht s, dde ech row b the coeffcent n tht row wth the mmm mgntde. It s dsed to scle sstem before clcltng ts determnnt. Ths s especll mportnt f we re clcltng the determnnt to see f the sstem s ll-condtoned or not. Consder the followng sstems 3 = 5-3 = 5 3.98 6 = 7 39.8-6 = 7 The re ctll the sme sstem. In the second one the eqtons re mltpled b. Determnnt of the st sstem s (-6) (-3)(3.98) = -.6, whch s close to zero. Determnnt of the nd sstem s (-6) (-3)(39.8) = -6, whch s not tht close to zero. So s ths sstem ll-condtoned or not? Scle ths sstem. The re the sme, se the st one..6667 + = -.6667 -.6633 + = -.667 Now clclte the determnnt s.6667* *(-.6633) =.33. Use ths le n jdgng the sstems condton. Note tht How close to zero? s stll n open qeston. There re other ws to determne sstems condton (See pge 77).

Scled Prtl Potng Sclng s lso sefl when some rows he coeffcents tht re lrge compred to those n other rows. Consder the followng sstem + = + = Ect solton s (.,.99998) () If we sole ths sstem wth Gss Elmnton, no potng s necessr ( s lrger thn ). Use onl 3 sg. fgs to emphsze the rond-off errors..* +.* 5 =.* 5 =.* WRONG - 5.* 4 = -5.* 4 =.* OK (b) Frst scle the sstem nd thn sole wth Gss Elmnton..* -5 + = + = Ths sstem now needs potng. Interchnge the rows nd sole..* -5 + =.* + = =. OK + =.* = =. OK Conclson: Sclng showed tht potng s necessr. Bt sclng tself s not necessr (pot the orgnl sstem nd sole). Sclng lso ntrodces ddtonl rond-off errors. Therefore se sclng to decde whether potng s necessr or not bt thn se the orgnl coeffcents.

Emple for Gss Elmnton wth Scled Prtl Potng Emple 5: Sole the followng sstem sng Gss Elmnton wth scled prtl potng. Keep nmbers s frctons of ntegers to elmnte rond-off errors. 3-3 9 3-9 -6 4-8 -34 6-4 6-8 6 6 Strt b formng the scle ector. It hs the lrgest coeffcent (n mgntde) of ech row. SV = {3 8 6 } Ths scle ector wll be pdted f we nterchnge rows drng potng. Step : Forwrd Elmnton

Emple for Gss Elmnton wth Scled Potng (cont d) (.) Compre scled coeffcents 3/3, 6/8, 6/6, /. Thrd one s the lrgest (ctll forth one s the sme bt we se the frst occrnce). Interchnge rows nd 3. 6-4 6-6 4-8 -34 3-3 9 3-9 -8 6 6 Updte the scle ector SV = {6 8 3 } Elmnte. Sbtrct (-6/6) tmes row from row. Sbtrct (3/6) tmes row from row 3. Sbtrct (/6) tmes row from row 4. Resltng sstem s 6-4 6 3-4 -8-8 -7-4 -6 3

Emple for Gss Elmnton wth Scled Potng (cont d) (.) Compre scled coeffcents /8, /3, 4/. Second one s the lrgest. Interchnge rows nd 3. 6-4 6-8 -7 3-4 -8-4 -6 Updte the scle ector SV = {6 3 8 } Elmnte. Sbtrct (/(-)) tmes row from row 3. Sbtrct ((-4)/(-)) tmes row from row 4. Resltng sstem s 6-4 6-8 -7 3/3-83/6-45/ -/3 5/3 3 4

Emple for Gss Elmnton wth Scled Potng (cont d) (.3) Compre scled coeffcents (3/3)/8, (/3)/. Frst one s lrger. No need for potng. Scle ector remns the sme SV = {6 3 8 } Elmnte 3. Sbtrct ((-/3)/(3/3)) tmes row 3 from row 4. Resltng sstem s 6-4 6-8 -7 3/3-83/6-45/ -6/3-6/3 Step : Bck sbsttton Eqton 4 4 = Eqton 3 3 = - Eqton = Eqton = 3 These re ect reslts becse rond-off errors re elmnted b not sng flotng pont nmbers. 5

Ill-condtoned Sstems The he lmost snglr coeffcent mtrces. The he nqe solton. Consder the followng 33 sstem 3. -.5.53 -.6 4.33.56 -.87 7.3 -.83 -.54.47-3.38 The solton of ths sstem s [ ] T Chnge b from.6 to.6. The soltons chnges to [.566 4.5.35] T Or chnge from 3. to 3.. The soltons chnges to [.77 6.5.7333] T Ths s tpcl ll-condtoned sstem. Its solton s er senste to the chnges n the coeffcent mtr or the rght-hnd-sde ector. Ths mens t s er senste to rond-off errors. Doble precson mst be sed to redce rond-off errors s mch s possble. Scled potng shold lso be sed to decrese rond-off errors. Fortntel not mn engneerng problems wll reslt n n ll-condtoned sstem. 6

Other Uses of Gss Elmnton Ges the LU decomposton of A sch tht [L][U]=[A]. LU decomposton s sefl f we re solng mn sstems wth the sme coeffcent mtr A bt dfferent rght-hnd-sde ectors (See pge 66). It cn be sed to clclte the determnnt of mtr. At the end of the Forwrd Elmnton step we get n pper trnglr mtr. For ths mtr the determnnt s jst the mltplcton of dgonl elements. If we nterchnged rows m tmes, thn the detemnnt cn be clclted s A = (-) m * * * nn Emple 6: Remember the emple we sed to descrbe potng. The A mtr ws 3 4 3 6-6 -5 After the Forwrd Elmnton step we fond the followng pper trnglr mtr. 6-6 -5-3.6667 4 4.3333 6.88 5.6364.56 To get ths, we sed potng nd nterchnged rows twce. Therefore the determnnt of A s A = (-) * 6 * (-3.6667) * 6.88 *.56 = -34.8 7

Gss-Jordn Method Ths s nother elmnton technqe. It s rton of Gss Elmnton. The dfference s, when n nknown s elmnted, t s elmnted from ll other eqtons, not jst the sbseqent ones. At the sme tme ll rows re normlzed b ddng them to ther pot element. At the end of the forwrd elmnton step Gss-Jordn method elds n dentt mtr, not n pper trnglr one. There s no need for bck sbsttton. Rght-hnd-sde ector hs the reslts. Emple 7: Sole the followng 44 sstem sng G-J method wth potng. Note tht ths s the sme sstem tht we sed to demonstrte Gss Elmnton wth potng. 3-4 3-7 6-6 -5 6 Interchnge rows nd 4, dde the new frst row b 6 nd elmnte from the nd, 3 rd nd 4 th eqtons..667 - -.8333.6667 5 3.6667-4 -3.6667 4 4.3333-8

Gss-Jordn Method (cont d) Interchnge rows nd 3, dde the new second row b -3.6667 nd elmnte from the st, 3 rd nd 4 th eqtons. -.88 -.6364.5 -.99 -.88 3 6.88 5.6364-9.88 3.3636-6 No potng s reqred. Dde the thrd row b 6.88 nd elmnte 3 from the st, nd nd 4 th eqtons..4.58 -.8.56 -.3.5599-3. Dde the lst row b.5599 nd elmnte 4 from the st, nd nd 3 rd eqtons. -.5..3333 - Rght-hnd-sde ector s the solton. No bck sbsttton s reqred. Gss-Jordn reqres lmost 5% more opertons thn Gss Elmnton (n 3 / nsted of n 3 /3). 9

Clcltng the Inerse of Mtr wth Gss-Jordn Clcltng the nerse of mtr s mportnt f we wnt to sole mn sstems wth the sme coeffcent mtr, bt dfferent rght-hnd-sde ectors. To tke the nerse of A, gment t wth n dentt mtr nd ppl the Gss-Jordn method. Emple 8: Fnd the nerse of the followng mtr. - A = 3 Agment A wth 33 dentt mtr. - 3 Appl the Gss-Jordn method to ths sstem to get.4. -..6 33 mtr t the rght s the nerse of A. Ths nerse cn now be sed to sole sstem wth the coeffcent mtr A nd dfferent rght-hnd-sde ectors s {} = [A] - {B} LU decomposton cn be sed for the sme prpose nd t s more effcent (See pge 73). 3

Iterte Methods - Gss-Sedel Method Gss Elmnton nd ts rnts re clled drect methods. The re not preferred for lrge sstems. Iterte methods strt wth n ntl solton ector nd tertel conerge to the tre solton. The stop when the pre-specfed tolernce s reched nd rond-off errors re not n sse. Gen the sstem [A]{}={B} nd strtng les {}, Gss-Sedel ses the frst eqton to sole for, second for, etc. = (b 3 3..... n n ) / = (b 3 3..... n n ) /...... n = (b n n n..... (n-)(n-) n- ) / nn After the frst terton o get {}. Use these les to strt new terton. Repet ntl the tolernce s stsfed s k k, % k s for ll the nknowns (=,,n), where k nd k- represent the present nd preos tertons. 3

Gss-Sedel Emple Sole the followng sstem sng the Gss-Sedel Method 6 + 3 = - + 7 + 3 = 5 strtng wth = = 3 =. + 5 3 = - Rerrnge the eqtons = ( + - 3 ) / 6 = (5 + - 3 ) / 7 3 = ( + + ) / 5 Frst terton = ( + - 3 ) / 6 = ( + - )/6 =.833 = (5 + - 3 ) / 7 = (5 + *.8333 - )/7 =.38 3 = ( + + ) / 5 = ( +.8333 + *.38)/5 =.6 Second terton = ( + - 3 ) / 6 = ( + *.38.6)/6 =.69 = (5 + - 3 ) / 7 = (5 + *.69 *.6)/7 =. 3 = ( + + ) / 5 = ( +.69 + *.)/5 =.5 Contne lke ths to get (Do not forget to check conergence sng the specfed tolernce) Zeroth Frst Second Thrd Forth Ffth..833.69.998.999...38..995.. 3..6.5.998.. 3

Jcob Method Gss- Sedel lws ses the newest lble les. Jcob Method ses les from the preos terton. Emple 9: Repet the preos emple sng the Jcob method. Rerrnge the eqtons = ( + - 3 ) / 6 = (5 + - 3 ) / 7 3 = ( + + ) / 5 Frst terton (se les) = ( + - 3 ) / 6 = ( + - )/6 =.833 = (5 + - 3 ) / 7 = ( 5 + - )/7 =.74 3 = ( + + ) / 5 = ( + + )/5 =. Second terton (se les) = ( + - 3 ) / 6 = ( + *.74.)/6 =.38 = (5 + - 3 ) / 7 = (5 + *.833 *.)/7 =.8 3 = ( + + ) / 5 = ( +.833 + *.74)/5 =.85 Contne to get Zeroth Frst Second Thrd Forth Ffth... Eghth..833.38.85.4.994......74.8.53..99.... 3...85.8.38..... 33

Conergence of the Iterte Methods Note tht these methods cn be seen s the mltple pplcton of Smple One-Pont Iterton. The m conerge or derge. A conergence crter cn be dered strtng from the conergence crter of the Smple One-Pont Iterton method. When the sstem of eqtons cn be ordered so tht ech dgonl entr of the coeffcent mtr s lrger n mgntde thn the sm of the mgntdes of the other coeffcents n tht row (dgonll domnnt sstem) the tertons conerge for n strtng les. Ths s sffcent bt not necessr crter. For ll,...,n Mn engneerng problems stsf ths reqrement. n j j j The emple we sed hs dgonll domnnt coeffcent mtr. Gss-Sedel conerges sll fster thn the Jcob method. Eercse 4: Interchnge the nd nd 3 rd eqtons of the preos emple. Is the sstem dgonll domnnt now? Sole ths sstem sng both terte methods. Compre ther conergence rtes wth tht of the soled emple. Do o obsere n dergence? 34

Improng Gss-Sedel wth Relton After ech new le of s compted, tht le s modfed b weghted erge of the reslts of the preos nd present tertons, new = l new + ( - l) old < l < : weghtng fctor If l = no relton (Orgnl Gss-Sedel) If <l< If <l< nder relton (Used to mke dergng sstem conerge) oer relton (Used to speed p the conergence of n lred conergng sstem. Clled SOR, Sccesse (or Smltneos) Oer Relton) Choce of l s problem specfc. Eercse 5: Sole the sstem we sed n the Gss-Sedel emple wth dfferent weghtng fctors. Do o notce n dfference n the conergence rte? 35

Psedocode for the Gss-Sedel Method LOOP k from to miter LOOP from to n = B LOOP j from to n IF ( j) = - A j j ENDLOOP = / A ENDLOOP CONVERGED = TRUE LOOP from to n Eercse 6: Improe ths psedocode wth relton. Eercse 7: Modf ths psedocode for the Jcob method. OUTPUT = ( old ) / * IF ( > tolernce) CONVERGED = FALSE ENDLOOP IF (CONVERGED = TRUE) STOP ENDLOOP 36

Solng Sstem of Nonlner Eqtons Emple : Solng Sstem Usng Smple One-Pont Iterton Sole the followng sstem of eqtons e 4 Pt the fnctons nto the form =g (,), =g (,) g g () ln( ) () 4 Select strtng les for nd, sch s =. nd =.. The don t need to stsf the eqtons. Use these les n g fnctons to clclte new les. = g ( ) = = g ( ) = - = g ( ) =.98689 = g ( ) = -.67436 3 = g ( ) =.98543669 3 = g ( 3 ) = -.746 4 = g ( 3 ) =.8698 4 = g ( 4 ) = -.773 The solton s conergng to the ect solton of =.469, =-.79637 Eercse 8: Sole the sme sstem bt rerrnge the eqtons s =ln(-) nd strt from = =-.7. Remember tht ths method m derge. = (4- )/ 37

38 Solng Sstem of Nonlner Eqtons (cont d) Solng Sstem Usng Newton-Rphson Method Consder the followng generl form of two eqton sstem Wrte st order TSE for these eqtons ) (, ) (, ) ( ) ( ) ( ) ( To fnd the solton set + = nd + =. Rerrnge The frst mtr s clled the Jcobn mtr. Sole for +, + nd terte. Cn be generlzed for n smltneos eqtons

39 Solng Sstem of Eqns Usng NR Method (cont d) Sole for + nd + sng Crmer s rle (ME ), The denomntor s the determnnt of the Jcobn mtr J. Emple : Sole the sme sstem of eqtons sttng wth =, = e 4 =, = = =- =.788 J =-3.436564 =, =-.63953 =3.335387 =8.479595 =.647636 J =-4.485 =, =-3.4568 =.337769 =8.3469 =.3767 J =-6.59555 =3, 3 =-.3643 3 =.959 3 =.48497 3 =.7787 J 3 =-4.499343 Looks lke t s conergng to the root n the nd qdrnt -.8,.8. Eercse 9: Cn o strt wth = nd =? Eercse : Tr to fnd strtng ponts tht wll conerge to the solton n the 4 th qdrnt. J e, J e - J, e,,

4 Solng Sstem of Nonlner Eqtons (cont d) Preos N-R solton of sstem of nonlner eqtons cn be generlzed to the solton of sstem of n nonlner eqtons (See pge 57 for detls). Gen the followng n nonlner eqtons wth n nknowns f (,,..., n ) = f (,,..., n ) =..... f n (,,..., n ) = Form the Jcobn mtr. k s the terton conter. Clclte new [Z] t ech terton. Sole the followng sstem sng n of the technqes tht we lerned n ths chpter. [Z] k {} k+ = -{F} k + [Z] k {} k where F k = f ( k, k,..., n k ) k n n n n n n k f f f f f f f f f [Z]

Emple (Fll Em Qeston): 3 q 3 Lnk lengths: 4 = 4, =, 3 = 4, 4 = 5 q q 4 Crnk ngle: q = 45 = p/4 Use the Newton-Rphson method to sole the followng set of eqtons for q 3 nd q 4. Strt wth the followng ntl gesses: q 3 = 55, q 4 = 8 Perform two tertons. f (q 3, q 4 ) = cos(q ) + 3 cos(q 3 ) 4 cos(q 4 ) = f (q 3, q 4 ) = sn(q ) + 3 sn(q 3 ) 4 sn(q 4 ) = 4

4 N-R cn be ppled to nonlner eqtons s follows [Z] k {X} k+ = -{F} k + [Z] k {X} k where [Z] k s the Jcobn mtr of the k th terton Iterton : q 3 = 55, q 4 = 8 k k f f f f [Z] ) cos( f f ) cos( f f ) sn( f f ) sn( f f 4 4 4 3 3 3 4 4 4 3 3 3 q q q q q q q q Emple (cont d)

Emple (cont d) Iterton : f f f f 3.766 4.94.943.868 q q q q 3 4 3 4 [Z] 3.766.943 4.94.868 {F}.597.33 Therefore the lner sstem s 3.766.943 4.94 q.868q 3 4.597.33 3.766.943 4.94 55.8688 4.8667 56.9637 Sole ths sstem sng Gss Elmnton (or n other technqe tht we lerned) to get q 3 = 55.55 q 4 = 8.45 43

Emple (cont d) Iterton : f f f f 3.868 4.93.797.8334 q q q q 3 4 3 4 [Z] 3.868.797 4.93.8334 {F}.395.9 Therefore the lner sstem s 3.868.797 4.93 q.8334q 3 4.395.9 3.868.797 4.93 55.55.83348.45 5.933 59.84 Sole ths sstem sng Gss Elmnton (or n other technqe tht we lerned) to get q 3 = 55.4996 q 4 = 8.7994 44