Scribd
Upload
49 views0 pages

Solution of Discrete Algebraic Systems

This document discusses methods for solving discrete algebraic systems that arise from computational fluid dynamics simulations. It covers sequential or Picard iteration, which generates successive approximations to reach the solution. Newton-Raphson method is also covered, which linearizes the nonlinear system and solves a sequence of linear systems to converge more rapidly if initialized close to the solution. Both methods require solving linear systems at each iteration, so efficient linear solvers are important. Choice of solution method depends on factors like sparsity and symmetry of the resulting system of equations.

Uploaded by

Saumya Sinha
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
49 views0 pages

Solution of Discrete Algebraic Systems

This document discusses methods for solving discrete algebraic systems that arise from computational fluid dynamics simulations. It covers sequential or Picard iteration, which generates successive approximations to reach the solution. Newton-Raphson method is also covered, which linearizes the nonlinear system and solves a sequence of linear systems to converge more rapidly if initialized close to the solution. Both methods require solving linear systems at each iteration, so efficient linear solvers are important. Choice of solution method depends on factors like sparsity and symmetry of the resulting system of equations.

Uploaded by

Saumya Sinha
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 0

COMPUTATIONAL FLUID DYNAMICS: ALGEBRAIC SYSTEMS: General Techniques

Dr K M Singh, Indian Institute of Technology Roorkee NPTEL 24.1


Lecture 24
SOLUTION OF DISCRETE ALGEBRAIC SYSTEMS

24.1 INTRODUCTION
Application of FDM, FVM or FEM leads to a system of algebraic equations which may be
linear or non-linear depending on the problem. For non-linear systems, we have to use an
iterative procedure which can be
Newton-Raphson type method: rate of convergence is very fast provided initial guess
is selected properly, OR
A global method (e.g. Picard iteration): guarantees converged solution starting from
an arbitrary initial guess, but rate of convergence is often very slow.
A globally convergent method which employs a mix of global and Newtons method
(Press et al., 2002).
Most of these methods require solution of a linear system at each iteration. Thus, we require
an efficient linear equation solver for linear as well as non-linear problems, and hence, we
concentrate on solution of linear systems in this and next two lectures.

Choice of a particular solution scheme must account for
Sparse nature of the algebraic system obtained from FD/FV/FE discretization
Nature of the linear system (whether its symmetric or not)

Before proceeding further, let us have a brief look at methods for non-linear equations. There
are numerous methods available in the literature for solution of nonlinear algebraic equations
(see Press et al., 2002). We discuss two simple, but widely used, approaches in this lecture.
24.2 SEQUENTIAL (PICARD) ITERATION
The simplest approach to solve a nonlinear equation ( ) 0 f x is to recast it in the form
( ) x g x (24.1)
Then, starting from an initial guess x
0
, successive iterates can be computed as

1
( ), 0,1, 2,....
k k
x g x k

(24.2)
Iterations are continued till a converged solution is obtained. Similar procedure can also be
extended to a system of coupled non-linear equations
F(x) = 0 (24.3)
which can be re-cast as
x = G(x) (24.4)
Permitting computation of improved iterates starting from an initial guess x
0
as

1
( ), 0,1, 2,....
k
k
i i
x G k

x (24.5)
Another iterative approach can be formulated by re-casting the system of equations (24.3) as
a linearized system given by
A(x)x = b(x) (24.6)
in which system matrix A and load vector b are functions of unknown vector x. In this case,
starting from the initial guess x
0
, improved iterates are computed by solving a linear system
of equation at each iteration, i.e.




COMPUTATIONAL FLUID DYNAMICS: ALGEBRAIC SYSTEMS: General Techniques
Dr K M Singh, Indian Institute of Technology Roorkee NPTEL 24.2

1
, 0,1, 2,.....
k k k
k

A(x )x = b(x ) (24.7)


In the preceding equation, system matrix A load vector b are evaluated using the current
known iterate of vector x. This

24.3 NEWTON-RAPHSON METHOD
Newton-Raphson method is based on Taylor series expansion of function f(x) around a
guessed value, i.e.

2
( )
( ) ( ) ( ) ( ) ( ) ...
2!
k
k k k k
x x
f x f x x x f x f x

(24.8)
In Eq. (24.8), superscript primes indicate derivatives. If we retain only first two terms on
RHS, then an improved guess can be estimated as

1
( )
, 0,1, 2,....
( )
k
k k
k
f x
x x k
f x

(24.9)
The preceding approach can be easily generalized for a coupled nonlinear system (24.3). For
a small increment x , function F
i
can be expanded in Taylor series as

2
1
( ) ( ) ( )
n
i
i i j
j j
F
F F x O
x

x x x x (24.10)
By neglecting higher order terms and setting ( ) 0
i
F x x , we get

1
( ) or ( ),
n
i i
j i ij j i ij
j j j
F F
x F J x F J
x x

x x (24.11)
In the preceding equation, J is called the Jacobian matrix. For a given iterate x
k
, we first
compute (i) the function values and (ii) the Jacobian, and then solve Eq. (24.11) to get the
correction vector
k
x . Thus, new iterate is given by

1 1
, ( ) ( )
k k k k k k


x = x x x J x F x (24.12)
Newton iteration converges very rapidly if the initial guess is close to the solution. Its
convergence not guaranteed starting from an arbitrary guess. To get over this problem, a
globally convergent algorithm based on line searches and back-tracking in conjunction with
Newton-Raphson iteration is usually preferred (see Press et al., 2002 for further details of this
algorithm).



REFERENCES/FURTHER READING
Ferziger, J. H. And Peri, M. (2003). Computational Methods for Fluid Dynamics. Springer.
Press, W. H., Teukolsky, S. A., Vetterling, W. T. and Flannery, B. P. (2002). Numerical
Recipes in C++. Cambridge University Press, Cambridge.

You might also like