Preconditioned Conjugate Gradient Method

25 July 2006, by Nadir Soualem

All the versions of this article: [English] [français]

Solving linear systems resulting from the finite differences method or of the finite elements shows the limits of the conjugate gradient. Indeed, Spectral condition number of such matrices is too high. The technique of Preconditioned Conjugate Gradient Method consists in introducing a matrix C subsidiary.

Problem

We want to solve the following system:

where

is a $n\times n$ symmetric definite and positive matrix( $A^\top =A$ and $x^\top Ax>0$ , for all $x\in \mathbf{R}^n$ non zero).
Let $x_\star$ be the exact solution of this system.

Spectral condition number

It happens sometimes that the spectral condition number $\kappa(A)$ is too high (eigenvalues are not well distributed). Preconditionnement
consists in introducing regular matrix $C\in\mathcal{M}_n(\mathbb{R})$ and solving the system:

$C^{-1}(Ax)=C^{-1}b\Leftrightarrow Ax=b$

such that the new spectral condition number is smaller for a judicious choice of the matrix

Preconditioned Conjugate Gradient Method

Let $x_0\in \mathbb{R}^{n}$ be an intial vector, Preconditioned Gradient Method algorithm
is the following one:

$r_{0}=b-Ax_{0}$

$z_{0}={C}^{-1}r_{0}$

$d_{0}=z_{0}$

For $k=0,1,2,\ldots$

$\alpha_k={{z_{k}^{\top}r_{k}}\over{{ d}_k^{\top}Ad_k}}$

$x_{k+1}=x_{k}+\alpha_kd_k$

$r_{k+1}=r_{k}-\alpha_kAd_k$

$z_{k+1}={C}^{-1}r_{k+1}$

$\beta_{k+1}={{z_{k+1}^{\top}r_{k+1} } \over { z_{k}^{\top}r_{k} } }$

$d_{k+1}=z_{k+1}+\beta_{k+1}d_{k}$

EndFor

Jacobi Preconditionner

Jacobi Preconditioner consists in taking the diagonal of for the matrix , i.e.

$C_{ij}= \left\{ \begin{array}{cc} A_{ii} & \textrm{si }i=j ,\\ 0 &\textrm{sinon}. \end{array} \right.$

Advantages of such preconditioner are the facility of its implementation and the low memory it needs.
But we can find other preconditioners such that resolution of the linear system is fastest, it is the case of the
SSOR Preconditioner.

SSOR Preconditioner(Symmetric Successive Over Relaxation)

We decompose the symmetric matrix like follows:

$A=L+D+L^{\top}$

where

is the strictly lower part of

and

is the diagonal of

. SSOR Preconditioner consists
in taking

$C=(\frac{D}{\omega}+L)\frac{\omega}{2-\omega}D^{-1}(\frac{D}{\omega}+L^{\top})$

where $\omega$ is a relaxation parameter. A necessary and sufficient condition of the Preconditioned Gradient Method
algorithm is to fix the parameter $\omega$ in the interval $]0,2[$ .

Math-Linux.com

Preconditioned Conjugate Gradient Method

Problem

Spectral condition number

Preconditioned Conjugate Gradient Method

Jacobi Preconditionner

SSOR Preconditioner(Symmetric Successive Over Relaxation)

Any message or comments?

Help Math-Linux !

Thank you !!!

Mathematics

Also in this section