ME 503

Advanced Numerical Analysis

Dr. Engr. Muhammad Ishfaq Khan

Department of Mechanical Engineering,

Faculty of Engineering & Technology,
International Islamic University, Islamabad.
 Course Objectives
1. To integrate a discussion of the properties of engineering and
physical problems with the discussion of methods by which such
problems may be solved numerically
2. To provide understanding of main sources of numerical errors and
the power of numerical methods that minimize these errors
 Course Outline
• Introduction: • Preview:
• Basic Ideas, • Choice of numerical mesh,
• Cartesian,
• Concepts, • Polar-Cylindrical,
• Terminology, • General Orthogonal,
• Regular Non-Orthogonal,
• Elements of a Numerical Method:
• Arbitrary Triangular Meshes,
• Differential Formulation,
• Discretization,
• Solution Domain and Mesh,
• Truncation and Discretization Errors,
• Discretization,
• Newton’s Method of Solving Algebraic
• Set of Algebraic Equations, Equation of Single Variable,
• Solution Algorithm. • Crammer’s Rule for Solving Set of
Equations,
• Round Off Errors and their Estimation.
 Course Outline
• Polynomials and Finite Differences: • Solution of Equation Sets:
• Collocation-Type Polynomials, • Ill-Conditioning,
• Finite Difference- Operator Algebra, • Iterative Solution Methods,
• Forms of Polynomials, • Decomposition,
• Relationship to Taylor Series. • Eigen-Value Problem,
• Finite Differences: • System Stability,
• Differences and Differential Operators, • Characteristic Polynomial,
• Basic Operator Relations, • Roots,
• Relations of First, Second and Higher • Eigen-Values,
Order Derivatives to Difference Series,
• Convergence of Solution Scheme.
• Solutions Errors.
 Course Outline
• Ordinary Differential Equations: • Partial Differential Equations:
• Order, • Variants of Partial Differential Equation,
• Methods of Solving First Order • Choice of Finite-Difference Formulation
Ordinary Differential Equations, and Solution Algorithm,
• Higher Order Differential • Elliptic, Parabolic and Hyperbolic
Equations.
Equations and their Conversion
into Set of First Order Ordinary • Discretization and Solution of Second
Equations. Order 2-D Steady Diffusion Equation,
• First Order 1-D Transient Diffusion
Equation,
• First Order 1-D Convection Equation,
• Finite Volume Approach
 Text Books
Numerical Analysis
Richard L. Burden, John Douglas Faires, 9th Edition,
Cengage Learning, 2010,
ISBN: 0538733519, 9780538733519.
Applied Numerical Analysis
Curtis F. Gerald, Patrick O. Wheatley,
7th Edition (August 10, 2003),
Pearson,
ISBN-10: 0321133048, ISBN-13: 978- 0321133045
 Introduction
•Basic Idea
• Most Mathematical problems from science and engineering are very complex, so, either
• Cannot be solved analytically or
• the analytical method is intractable i.e. too many equations need to be solved
simultaneously
• The solution to such problems are “approximated”/ “guessed” i.e. they are solved
numerically
• Approximation(s) are done through established mathematical techniques and
corresponding concepts discussed later in the course
• In the process of “approximating solutions”, many empirical equations arise that are
handled by:
• Matrix algebra on paper, or
• Developing algorithms in computers, or
• both
 Introduction
• In this course we will get theoretical background and build concepts
pertaining to:
• Techniques to develop “approximate solutions” for complex problems
• Utilization of techniques to simultaneously solve a large of number of
equations on paper, and
• Hands on experience on utilizing concerned modern computer
tools/established algorithms (if time permits!)
• Challenges and terms in implementation of the Numerical Analysis to any
situation
 Introduction
•Most widely used N.A. Techniques
• Interpolation • Runge-Kutta Method • Weightage Methods
• Extrapolation • Newton’s Methods • Finite Element Methods
• Regression • Trapezoidal law • Finite Difference Methods
• Euler method • Simpson’s Law • Finite Difference Methods
• Mathematical Systems Solved by N.A.
• ODEs • Algebraic Equations • Integrals
• PDEs • Data Set Manipulations
• Statistical & Mathematical Concepts Employed by N.A
• Mean • Kurtosis • Taylor Series
• Median • Skewness • Fourier Series
• Mode • Standard Deviation • Mc Lauren Series
 Introduction
• Areas of Application of N.A.
• Structural Mechanics ( Structural Analysis)
• Celestial Mechanics (Movement of Stars & Galaxies)
• Heat Transfer ( Convection, Conduction, Radiation)
• Mechanical Vibrations ( Mode Shapes, Modal Frequencies)
• Fluid Mechanics ( Fluid Flows, FSI)
• Optimization ( Mechanical Engineering Designing)
• Electromagnetism ( Designing of Electronic Appliances)
• Famous Software Tools Employing N.A.
• MATLAB
• ANSYS
• COMSOL
History of Numerical Analysis
 Introduction
• Concepts & Terminology Related to N.A.
• Iterative Method
• To repeat
• Keep adding extra terms from mathematical series or
• Keep repeating the operation(s) to achieve required accuracy or conditions required by an
algorithm
• Algorithm
• A complete and unambiguous set of procedures that are used to find the solution to a
Mathematical problem
• A computer program
• A software employs vast number of algorithms
• Research methodology also requires an algorithm
• Accuracy
• Degree of closeness of computed value with the true value
• Precision
• Degree of closeness of computed values with each other
 Introduction
• Concepts & Terminology Related to N.A.
• Errors
• Difference between a true value and computed value through algorithm
• It may be
• Truncation Errors (ignoring all after selecting a finite number of terms from an infinite series)
• Round Off Errors (linked with significant figures, limited ability to represent numbers in computers)
• Discretization Errors ( due to division/solution of a continuous problem at discrete points/intervals)
• Stability
• Error of a numerical scheme or algorithm remains in reasonable/defined limits as it progresses
• Consistency
• Error of a numerical scheme vanishes or approaches zero with decrease in size of time and
space
• Convergence
• Solution of a numerical scheme or algorithm converges towards/systematically approaches real
value
• Discretization
• Splitting of a continuous function into a discrete function at each point in space and/or time
 Introduction
Elements of a Numerical Method ∆
x
• Discretization i,j+
2
• Equation Discretization (ODEs & PDEs) y
i,j+
• FDM (Based on Taylor Series) 1
i-2, i-1, i,
• FEM (Weak form) j j j

• Spatial Discretion ∆
i+1,
j
i+2,
j
• Division of domain into small parts y i,j-
1
• Numerical schemes are applied to it on its nodes i,j- x
• Discretization for a two-dimensional case in 2

space is elaborated in Figure

• Mesh size or step size
• Temporal discretization
• can also be carried out for time dependent
problems
• Marching Problems
 Introduction
•
 Introduction
•

∆
x
i,j+
2
y

i,j+
1
i-2, i-1, i,
j j j
i+1, i+2,
∆ j j
y i,j-
1

i,j-
x
2
 Introduction
Elements of a Numerical Method
• Solution Domain
• Representation of region of which analysis is needed to be carried out
• Region of space of a particular geometric shape with in which Numerical Schemes
are to be applied
• It may be
• Model of a physical body of which Numerical Analysis is to be carried out
• A control volume in which analysis is to be carried out
• Or combination
•
 Elements of a Numerical Method
• Mesh Introduction
• Subdivision of a continuous geometric space into discrete
cells or elements
• Each element or cell has:
i. cell centroid
ii. Faces in case of 3-D element or cell
iii. nodes or vertices
iv. Edges
a. In 2-D faces and edges are the
same
b. In 3-D the face is a surface
surrounded by edges
• Equations are solved for each element to get values of
unknown variables
a) At nodes/vertices of the element/cell
• node-based or vertex-based schemes
b) At the center of the element/cell
• cell-based schemes
• Calculated values of variables of interest at nodes of a
cell/element are termed as Local Solution
• Local solutions of each element/cell are combined to get
Global Solution for the entire solution domain
 Introduction
Elements of a Numerical Method
• Mesh
• Mesh elements may be:
• 2-D
a) Triangle
b) Quadrilateral
• 3-D
a) Tetrahedron
b) Hexahedron
c) Prism
d) Pyramid
•
 Introduction
Elements of a Numerical Method
• Mesh
• Mesh Types
a) Structured Mesh
• every interior vertex in the domain is connected to the same number of
neighbor vertices
• Often structured meshes have orthogonal quadrilateral (2D) or hexahedral (3D)
elements
• coefficient matrices that are banded
• Require more number of elements than unstructured mesh for the same problem
• Easy for simple geometry and difficult to adopt for complicated geometry
• Memory efficient
• Fast to Solve
• Angled and curved geometries are approximated (leads to stair stepping)
 Introduction
Elements of a Numerical
Method
• Mesh
• Mesh Types
a) Structured Mesh
• It has 3 sub-types:
➢ Uniform Mesh
➢ Rectilinear Mesh
➢ Curvilinear Mesh
•
 Introduction
Elements of a Numerical Method
➢ Mesh
■ Mesh Types
● Unstructured Mesh
◆ every interior vertex in the domain is
mostly connected to different number of
neighbor vertices
◆ non-orthogonal, such as triangles (2D) and
tetrahedra (3D)
◆ Complex geometries easier to mesh
◆ local adaptive mesh refinement
◆ Greater memory requirement
◆ slower to solve
◆ It has two subtypes
➢ Single Cell Type Topology
➢ Mixed Cell Type Topology
 Introduction
Elements of a Numerical Method
• Mesh
• Mesh Types
c) Block Structured Mesh
• Geometry is divided into blocks
• Mesh inside each block is structured
• arrangement of the blocks
themselves is not necessarily
structured
• Each vertex is connected to an
arbitrary number of neighbor
vertices
• Hanging nodes (algorithms to
handle these nodes)
• It may be
• Matching
• Non-matching
• Chimera (composite)
 Introduction
Elements of a Numerical
Method
• Mesh
• Mesh Types
d) Conforming Mesh
• a matching of nodes at the
interface
• no (additional)
interpolation required at a
conformal interface
• computation faster and
more accurate
 Introduction
Elements of a Numerical Method
• Mesh
• Mesh Types
e) Non-Conforming Mesh
• a partial or zero matching of nodes at the
interface
• (additional) interpolation required at a
non-conformal interface
f) Adaptive Mesh
• difficult to estimate solution in some regions
due to discontinuities, steep gradients, shocks,
etc.
• refining the mesh of a simulation based on the
solution
 Introduction
Elements of a Numerical
Method
• Mesh
• Mesh Quality
a) Aspect Ratio
• ratio of a cell’s longest length to the
shortest length
• ideal aspect ratio is 1.
• Solution accuracy and convergence
depends on it
b) Non-orthogonality
• angle between the vector
connecting two adjacent cell
centers and the normal of the face
shared by these cells
• The range of non-orthogonality is
between 0 (ideal) and 90 (worst)
• numerical instability depends on
this factor
 Introduction
Elements of a Numerical Method
• Mesh
• Mesh Quality
c) Skewness
• deviation between the optimal
cell size to the existing cell
size
• range of skewness is between
0 (ideal) to 1 (worst)
• accuracy depends on it
because of interpolation with
in regions
 Introduction
Elements of a Numerical Method:
• Set of Algebraic Equations
• When equations are solved for each element of a mesh or node of a grid, a large set of
algebraic equations are formed
• The equations may be linear or nonlinear
• For linear equations, only one solution exists
• For nonlinear equations, more than one solution exist which depend on different factors
like initial guess
• There are two methods to solve the equations
1. direct
• If discretization methods result in A = φ B or φ = inv( A ) B
• If A is a small matrix, direct solution methods do good
• If A is a large matrix as in case of Fluid Dynamics (large number of cells + nonlinear
equations + large number of variables), direct solution methods are embedded with in
iterative processes
2. Iterative
• Initial guess is systematically improved based on convergence criteria
•
 Introduction
Elements of a Numerical Method:
• Set of Algebraic Equations
2. Iterative
• They are
• Stationary Methods
• Also called relaxation or smoothing methods
• performs in each iteration the same operations on the current iteration vectors
• It has:
• Jacobi Method
• Gauss-Seidel Method
• Successive Over-Relaxation Method
• Symmetric Successive Over-Relaxation Method
• Multigrid Methods
• very useful in problems exhibiting multiple scales
• It may be
• V-cycle
• W-cycle
• Full Multigrid
 Introduction
Elements of a Numerical Method:
• Set of Algebraic Equations
• Non-Stationary Methods
• iteration-dependent coefficients
• It may be:
• Conjugate Gradient Method (CG)
• Generalized Minimal Residual (GMRES)
• BiConjugate Gradient (BiCG)
• Quasi-Minimal Residual (QMR)