Scilab Website | Contribute with GitLab | Mailing list archives | ATOMS toolboxes
Scilab Online Help
6.0.2 - Français

Change language to:
English - 日本語 - Português - Русский

Please note that the recommended version of Scilab is 2025.0.0. This page might be outdated.
See the recommended documentation of this function

Aide de Scilab >> Matrices creuses > Linear Equations (Iterative Solvers) > qmr

qmr

quasi minimal residual method with preconditioning

Syntax

[x,flag,err,iter,res] = qmr(A,Ap,b,x0,M1,M1p,M2,M2p,maxi,tol)
[x,flag,err,iter,res] = qmr(A,b,x0,M1,M2,maxi,tol)

Arguments

A

matrix of size n-by-n or function.

  • matrix.If A is a matrix, it can be dense or sparse

  • function.If A is a function which returns A*x, it must have the following header :

    function y=A(x)

    If A is a function which returns A*x or A'*x depending t. If t = "notransp", the function returns A*x. If t = "transp", the function returns A'*x. It must have the following header :

    function y=A(x, t)
Ap

function returning A'*x. It must have the following header :

function y=Ap(x)
b

right hand side vector

x0

initial guess vector (default: zeros(n,1))

M1

left preconditioner : matrix or function (In the first case, default: eye(n,n)). If M1 is a function, she returns either,

  • only M1*x

  • or

  • M1*x or M1'*x depending t.

M1p

must only be provided when M1 is a function returning M1*x. In this case M1p is the function which returns M1'*x.

M2

right preconditioner : matrix or function (In the first case, default: eye(n,n)). If M2 is a function, she returns either

  • only M2*x

  • or

  • M2*x or M2'*x depending t.

M2p

must only be provided when M2 is a function returning M2*x. In this case M2p is the function which returns M2'*x

maxi

maximum number of iterations (default: n)

tol

error tolerance (default: 1000*%eps)

x

solution vector

flag
  • flag=0: qmr converged to the desired tolerance within maxi iterations,

  • flag=1: no convergence given maxi,

  • -7 < flag < 0: A breakdown occurred because one of the scalar quantities calculated during qmr was equal to zero.

res

residual vector

err

final residual norm

iter

number of iterations performed

Description

Solves the linear system Ax=b using the Quasi Minimal Residual Method with preconditioning.

Examples

// If A is a matrix
A=[ 94  0   0   0    0   28  0   0   32  0
    0   59  13  5    0   0   0   10  0   0
    0   13  72  34   2   0   0   0   0   65
    0   5   34  114  0   0   0   0   0   55
    0   0   2   0    70  0   28  32  12  0
    28  0   0   0    0   87  20  0   33  0
    0   0   0   0    28  20  71  39  0   0
    0   10  0   0    32  0   39  46  8   0
    32  0   0   0    12  33  0   8   82  11
    0   0   65  55   0   0   0   0   11  100];
b=ones(10,1);
[x,flag,err,iter,res] = qmr(A, b)

[x,flag,err,iter,res] = qmr(A, b, zeros(10,1), eye(10,10), eye(10,10), 10, 1d-12)

// If A is a function
function y=Atimesx(x, t)
A=[ 94  0   0   0    0   28  0   0   32  0
    0   59  13  5    0   0   0   10  0   0
    0   13  72  34   2   0   0   0   0   65
    0   5   34  114  0   0   0   0   0   55
    0   0   2   0    70  0   28  32  12  0
    28  0   0   0    0   87  20  0   33  0
    0   0   0   0    28  20  71  39  0   0
    0   10  0   0    32  0   39  46  8   0
    32  0   0   0    12  33  0   8   82  11
    0   0   65  55   0   0   0   0   11  100];
 if (t == 'notransp') then
       y = A*x;
   elseif (t ==  'transp') then
       y = A'*x;
   end
endfunction

 [x,flag,err,iter,res] = qmr(Atimesx, b)

 [x,flag,err,iter,res] = qmr(Atimesx, b, zeros(10,1), eye(10,10), eye(10,10), 10, 1d-12)

 // OR

 function y=funA(x)
A = [ 94  0   0   0    0   28  0   0   32  0
    0   59  13  5    0   0   0   10  0   0
    0   13  72  34   2   0   0   0   0   65
    0   5   34  114  0   0   0   0   0   55
    0   0   2   0    70  0   28  32  12  0
    28  0   0   0    0   87  20  0   33  0
    0   0   0   0    28  20  71  39  0   0
    0   10  0   0    32  0   39  46  8   0
    32  0   0   0    12  33  0   8   82  11
    0   0   65  55   0   0   0   0   11  100];
 y = A*x
endfunction

 function y=funAp(x)
A = [ 94  0   0   0    0   28  0   0   32  0
    0   59  13  5    0   0   0   10  0   0
    0   13  72  34   2   0   0   0   0   65
    0   5   34  114  0   0   0   0   0   55
    0   0   2   0    70  0   28  32  12  0
    28  0   0   0    0   87  20  0   33  0
    0   0   0   0    28  20  71  39  0   0
    0   10  0   0    32  0   39  46  8   0
    32  0   0   0    12  33  0   8   82  11
    0   0   65  55   0   0   0   0   11  100];
 y = A'*x
endfunction

 [x,flag,err,iter,res] = qmr(funA, funAp, b)

 [x,flag,err,iter,res] = qmr(funA, funAp, b, zeros(10,1), eye(10,10), eye(10,10), 10, 1d-12)

 // If A is a matrix, M1 and M2 are functions
 function y=M1timesx(x, t)
 M1 = eye(10,10);
   if(t=="notransp") then
       y = M1*x;
   elseif (t=="transp") then
       y = M1'*x;
   end
endfunction

function y=M2timesx(x, t)
 M2 = eye(10,10);
   if(t=="notransp") then
       y = M2*x;
   elseif (t=="transp") then
       y = M2'*x;
   end
endfunction

[x,flag,err,iter,res] = qmr(A, b, zeros(10,1), M1timesx, M2timesx, 10, 1d-12)

// OR

function y=funM1(x)
M1 = eye(10,10);
y = M1*x;
endfunction

function y=funM1p(x)
M1 = eye(10,10);
y = M1'*x;
endfunction

function y=funM2(x)
M2 = eye(10,10);
y = M2*x;
endfunction

function y=funM2p(x)
M2 = eye(10,10);
y = M2'*x;
endfunction

[x,flag,err,iter,res] = qmr(A, b, zeros(10,1), funM1, funM1p, funM2, funM2p, 10, 1d-12)

// If A, M1, M2 are functions
[x,flag,err,iter,res] = qmr(funA, funAp, b, zeros(10,1), funM1, funM1p, funM2, funM2p, 10, 1d-12)
[x,flag,err,iter,res] = qmr(Atimesx, b, zeros(10,1), M1timesx, M2timesx, 10, 1d-12)

See also

  • gmres — Generalized Minimum RESidual method
  • conjgrad — conjugate gradient solvers

History

VersionDescription
5.4.0 Calling qmr(A, Ap) is deprecated. qmr(A) should be used instead. The following function is an example :
function y=A(x, t)
Amat = eye(2,2);
if ( t== "notransp") then
y = Amat*x;
elseif (t == "transp") then
y = Amat'*x;
end
endfunction
Report an issue
<< gmres Linear Equations (Iterative Solvers) Sparse Matrix Conversion >>

Copyright (c) 2022-2024 (Dassault Systèmes)
Copyright (c) 2017-2022 (ESI Group)
Copyright (c) 2011-2017 (Scilab Enterprises)
Copyright (c) 1989-2012 (INRIA)
Copyright (c) 1989-2007 (ENPC)
with contributors
Last updated:
Thu Feb 14 14:59:57 CET 2019