Bergische Universität Wuppertal
Fachbereich Mathematik und Naturwissenschaften
Angewandte Mathematik - Numerische Analysis (AMNA)

People
Research
Publications
Teaching


Bergische Universität Wuppertal
Fachbereich C: Mathematik und Naturwissenschaften
Lehrstuhl für Angewandte Mathematik und Numerische Analysis
Prof. Dr. Matthias Ehrhardt
Dipl.-Math Alexander Schranner

Course in the Winterterm 2009/2010:

Introduction to Numerical Methods for Computer Simulation



Outline
 

§ 0. Introduction

§ 1. Error Analysis

1.1 Number Representation: Floating-Point and Roundoff
  • Definition 1: Floating Point
  • Example 1: Normalised Floats - decimal
  • Example 2: Normalised Floats - dual
  • Definition 2: Machine Numbers
  • IEEE-Standard 754
  • Example 3: Short real
  • Rounding
  • Definition 3: Round to nearest
  • Definition 4: Machine precision
  • Example 0.1 revisited
  • Floating point operations
1.2 Roundoff Error Analysis
  • Forward vs. Backward Analysis
  • Definition 5: Backward Analysis
  • Example 4: Backward Analysis for ..
  • Horner-Scheme
  • Algorithm 1: Horner-Scheme
1.3 Error propagation and condition
  • Definition 6: Condition numbers
  • Example 5:
  • Condition plus roundoff error
  • Error in Algorithm
  • Definition 7: Acceptable result
  • Definition 8: Well behaved / Numerically stable algorithm

§ 2. Vectors and Matrices

2.1 Notations
2.2 Determinant, Inverse, Eigenvalues
2.3 Special matrices
  • Definition 1: symmetric/hermitian matrix
  • Definition 2: orthogonal/unitary matrix
  • Definition 3: positive (semi)definite matrix
2.4 Norms
  • Definition 4: (vector) norm
  • Definition 5: matrix norm
  • Definition 6: subordinate matrix norm lub
  • Definition 7: spectral radius
  • Definition 8: condition of a matrix
2.5 Elementary matrices
  • Scaling
  • Transposition
  • Permutation
  • Row/column operations
2.6 BLAS Package

§ 3. Direct Methods for Linear Systems

3.1 Gaussian Elimination and LU-Decomposition
  • Algorithm 1: Forward Substitution
  • Algorithm 2: Backward Substitution
  • Algorithm 3: Gaussian Elimination without pivoting
  • Example 1:
  • Theorem 1: LU-Decomposition
  • Algorithm 4: LU-Decomposition
  • Complexity
  • Pivoting
  • Example 2:
  • Theorem 2:
  • Cholesky-Decomposition
  • Theorem 3:
  • Algorithm 5: Cholesky-Decomposition
3.2 Condition and Roundoff Error
  • Residual
  • Roundoff errors in decompositions
3.3 Linear Least Squares
  • Problem description
  • Example 3: Radio-active decay
  • Normal Equations
  • Theorem 4:
  • Condition of the least squares problem
  • Computation
  • Example 4:
  • Householder Transformation and QR-Decomposition
  • Theorem 5: Householder transformation

§ 4. Iterative Methods for Linear Systems

Motivation forr iterative methods
4.1 Illustrative Example
  • Example: Poisson problem on unit square
4.2 Classical iterative methods
  • Properties of iteration matrix
  • Theorem 1: Convergence of classical methods
  • Theorem 2: Speed of convergence
  • Jacobi method
  • Definition 1: strong row/column sum criterion
  • Gauss-Seidel method
  • Relaxation methods
  • under/over-relaxation
  • successive overrelaxation (SOR) method
  • Application to illustrative example
  • Iterative refinement
4.3 Conjugate gradient (CG) method
  • Krylov space
  • Krylov-subspace methods
  • energy norm
  • A-orthogonal basis
  • Theorem 3: basis of CG method
  • method of conjugate directions
  • Algorithm 1: Conjugate gradient method
  • Interpretation via optimization
  • preconditioning
  • incomplete Cholesky decomposition
4.4 GMRES method
  • method of generalised minimal residuals

§ 5. Methods for Nonlinear Systems

5.1 Newton's method
  • The Univariate Case
    • Algorithm 1: Bisection
    • Algorithm 2: Newton's method
    • Definition 1: local and global convergence
    • Definition 2: convergence of order p
  • The Multivariate Case
    • Algorithm 3: Newton-Raphson method
    • Theorem 1: Convergence of multivariate Newton method
    • Theorem 2 (newton-Kantorovich):
5.2 Simplified Newton method
  • Theorem 3: Convergence of simplified multivariate Newton method
  • Broyden's rank one method
5.3 Modified Newton method
  • Algorithm 4: modified Newton method
  • Theorem 4: Convergence of modified multivariate Newton method
  • natural scaling
5.4 Gauss-Newton scheme
  • Algorithm 5: Gauss-Newton method
  • Theorem 5: Convergence of multivariate Gauss-Newton method


University of Wuppertal
Faculty of Mathematics and Natural Sciences
Department of Mathematics
Applied Mathematics & Numerical Analysis Group

Last modified:   Disclaimer   ehrhardt@math.uni-wuppertal.de