1 ## Copyright (C) 2004-2012 Piotr Krzyzanowski
3 ## This file is part of Octave.
5 ## Octave is free software; you can redistribute it and/or modify it
6 ## under the terms of the GNU General Public License as published by
7 ## the Free Software Foundation; either version 3 of the License, or (at
8 ## your option) any later version.
10 ## Octave is distributed in the hope that it will be useful, but
11 ## WITHOUT ANY WARRANTY; without even the implied warranty of
12 ## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 ## General Public License for more details.
15 ## You should have received a copy of the GNU General Public License
16 ## along with Octave; see the file COPYING. If not, see
17 ## <http://www.gnu.org/licenses/>.
20 ## @deftypefn {Function File} {@var{x} =} pcr (@var{A}, @var{b}, @var{tol}, @var{maxit}, @var{m}, @var{x0}, @dots{})
21 ## @deftypefnx {Function File} {[@var{x}, @var{flag}, @var{relres}, @var{iter}, @var{resvec}] =} pcr (@dots{})
23 ## Solve the linear system of equations @code{@var{A} * @var{x} = @var{b}}
24 ## by means of the Preconditioned Conjugate Residuals iterative
25 ## method. The input arguments are
29 ## @var{A} can be either a square (preferably sparse) matrix or a
30 ## function handle, inline function or string containing the name
31 ## of a function which computes @code{@var{A} * @var{x}}. In principle
32 ## @var{A} should be symmetric and non-singular; if @code{pcr}
33 ## finds @var{A} to be numerically singular, you will get a warning
34 ## message and the @var{flag} output parameter will be set.
37 ## @var{b} is the right hand side vector.
40 ## @var{tol} is the required relative tolerance for the residual error,
41 ## @code{@var{b} - @var{A} * @var{x}}. The iteration stops if
42 ## @code{norm (@var{b} - @var{A} * @var{x}) <=
43 ## @var{tol} * norm (@var{b} - @var{A} * @var{x0})}.
44 ## If @var{tol} is empty or is omitted, the function sets
45 ## @code{@var{tol} = 1e-6} by default.
48 ## @var{maxit} is the maximum allowable number of iterations; if
49 ## @code{[]} is supplied for @code{maxit}, or @code{pcr} has less
50 ## arguments, a default value equal to 20 is used.
53 ## @var{m} is the (left) preconditioning matrix, so that the iteration is
54 ## (theoretically) equivalent to solving by @code{pcr} @code{@var{P} *
55 ## @var{x} = @var{m} \ @var{b}}, with @code{@var{P} = @var{m} \ @var{A}}.
56 ## Note that a proper choice of the preconditioner may dramatically
57 ## improve the overall performance of the method. Instead of matrix
58 ## @var{m}, the user may pass a function which returns the results of
59 ## applying the inverse of @var{m} to a vector (usually this is the
60 ## preferred way of using the preconditioner). If @code{[]} is supplied
61 ## for @var{m}, or @var{m} is omitted, no preconditioning is applied.
64 ## @var{x0} is the initial guess. If @var{x0} is empty or omitted, the
65 ## function sets @var{x0} to a zero vector by default.
68 ## The arguments which follow @var{x0} are treated as parameters, and
69 ## passed in a proper way to any of the functions (@var{A} or @var{m})
70 ## which are passed to @code{pcr}. See the examples below for further
71 ## details. The output arguments are
75 ## @var{x} is the computed approximation to the solution of
76 ## @code{@var{A} * @var{x} = @var{b}}.
79 ## @var{flag} reports on the convergence. @code{@var{flag} = 0} means
80 ## the solution converged and the tolerance criterion given by @var{tol}
81 ## is satisfied. @code{@var{flag} = 1} means that the @var{maxit} limit
82 ## for the iteration count was reached. @code{@var{flag} = 3} reports t
83 ## @code{pcr} breakdown, see [1] for details.
86 ## @var{relres} is the ratio of the final residual to its initial value,
87 ## measured in the Euclidean norm.
90 ## @var{iter} is the actual number of iterations performed.
93 ## @var{resvec} describes the convergence history of the method,
94 ## so that @code{@var{resvec} (i)} contains the Euclidean norms of the
95 ## residual after the (@var{i}-1)-th iteration, @code{@var{i} =
96 ## 1,2, @dots{}, @var{iter}+1}.
99 ## Let us consider a trivial problem with a diagonal matrix (we exploit the
105 ## A = sparse (diag (1:n));
110 ## @sc{Example 1:} Simplest use of @code{pcr}
116 ## @sc{Example 2:} @code{pcr} with a function which computes
117 ## @code{@var{A} * @var{x}}.
121 ## function y = apply_a (x)
125 ## x = pcr ("apply_a", b)
129 ## @sc{Example 3:} Preconditioned iteration, with full diagnostics. The
130 ## preconditioner (quite strange, because even the original matrix
131 ## @var{A} is trivial) is defined as a function
135 ## function y = apply_m (x)
136 ## k = floor (length (x) - 2);
138 ## y(1:k) = x(1:k) ./ [1:k]';
141 ## [x, flag, relres, iter, resvec] = ...
142 ## pcr (A, b, [], [], "apply_m")
143 ## semilogy ([1:iter+1], resvec);
147 ## @sc{Example 4:} Finally, a preconditioner which depends on a
148 ## parameter @var{k}.
152 ## function y = apply_m (x, varargin)
153 ## k = varargin@{1@};
155 ## y(1:k) = x(1:k) ./ [1:k]';
158 ## [x, flag, relres, iter, resvec] = ...
159 ## pcr (A, b, [], [], "apply_m"', [], 3)
165 ## [1] W. Hackbusch, @cite{Iterative Solution of Large Sparse Systems of
166 ## Equations}, section 9.5.4; Springer, 1994
168 ## @seealso{sparse, pcg}
171 ## Author: Piotr Krzyzanowski <piotr.krzyzanowski@mimuw.edu.pl>
173 function [x, flag, relres, iter, resvec] = pcr (A, b, tol, maxit, m, x0, varargin)
177 if (nargin < 6 || isempty (x0))
178 x = zeros (size (b));
187 if (nargin < 4 || isempty (maxit))
193 if (nargin < 3 || isempty (tol))
202 if (isnumeric (A)) # is A a matrix?
204 else # then A should be a function!
205 r = b - feval (A, x, varargin{:});
208 if (isnumeric (m)) # is M a matrix?
209 if (isempty (m)) # if M is empty, use no precond
211 else # otherwise, apply the precond
214 else # then M should be a function!
215 p = feval (m, r, varargin{:});
221 q_old = p_old = s_old = zeros (size (x));
223 if (isnumeric (A)) # is A a matrix?
225 else # then A should be a function!
226 q = feval (A, p, varargin{:});
229 resvec(1) = abs (norm (r));
232 while (resvec(iter-1) > tol*resvec(1) && iter < maxit)
234 if (isnumeric (m)) # is M a matrix?
235 if (isempty (m)) # if M is empty, use no precond
237 else # otherwise, apply the precond
240 else # then M should be a function!
241 s = feval (m, q, varargin{:});
250 lambda = b_top / b_bot;
255 if (isnumeric(A)) # is A a matrix?
257 else # then A should be a function!
258 t = feval (A, s, varargin{:});
261 alpha0 = (t'*s) / b_bot;
262 alpha1 = (t'*s_old) / b_bot_old;
267 p = s - alpha0*p - alpha1*p_old;
268 q = t - alpha0*q - alpha1*q_old;
275 resvec(iter) = abs (norm (r));
280 relres = resvec(iter-1) ./ resvec(1);
285 warning ("pcr: maximum number of iterations (%d) reached\n", iter);
286 warning ("the initial residual norm was reduced %g times.\n", 1.0/relres);
288 elseif (nargout < 2 && ! breakdown)
289 fprintf (stderr, "pcr: converged in %d iterations. \n", iter);
290 fprintf (stderr, "the initial residual norm was reduced %g times.\n",
297 warning ("pcr: breakdown occurred:\n");
298 warning ("system matrix singular or preconditioner indefinite?\n");
306 %! # Simplest usage of PCR (see also 'help pcr')
309 %! A = diag(linspace(-3.1,3,N)); b = rand(N,1); y = A\b; #y is the true solution
311 %! printf('The solution relative error is %g\n', norm(x-y)/norm(y));
313 %! # You shouldn't be afraid if PCR issues some warning messages in this
314 %! # example: watch out in the second example, why it takes N iterations
315 %! # of PCR to converge to (a very accurate, by the way) solution
318 %! # Full output from PCR
319 %! # We use this output to plot the convergence history
322 %! A = diag(linspace(-3.1,30,N)); b = rand(N,1); X = A\b; #X is the true solution
323 %! [x, flag, relres, iter, resvec] = pcr(A,b);
324 %! printf('The solution relative error is %g\n', norm(x-X)/norm(X));
325 %! title('Convergence history'); xlabel('Iteration'); ylabel('log(||b-Ax||/||b||)');
326 %! semilogy([0:iter],resvec/resvec(1),'o-g;relative residual;');
329 %! # Full output from PCR
330 %! # We use indefinite matrix based on the Hilbert matrix, with one
331 %! # strongly negative eigenvalue
332 %! # Hilbert matrix is extremely ill conditioned, so is ours,
333 %! # and that's why PCR WILL have problems
336 %! A = hilb(N); A(1,1)=-A(1,1); b = rand(N,1); X = A\b; #X is the true solution
337 %! printf('Condition number of A is %g\n', cond(A));
338 %! [x, flag, relres, iter, resvec] = pcr(A,b,[],200);
340 %! printf('PCR breakdown. System matrix is [close to] singular\n');
342 %! title('Convergence history'); xlabel('Iteration'); ylabel('log(||b-Ax||)');
343 %! semilogy([0:iter],resvec,'o-g;absolute residual;');
346 %! # Full output from PCR
347 %! # We use an indefinite matrix based on the 1-D Laplacian matrix for A,
348 %! # and here we have cond(A) = O(N^2)
349 %! # That's the reason we need some preconditioner; here we take
350 %! # a very simple and not powerful Jacobi preconditioner,
351 %! # which is the diagonal of A
353 %! # Note that we use here indefinite preconditioners!
357 %! for i=1:N-1 # form 1-D Laplacian matrix
358 %! A(i:i+1,i:i+1) = [2 -1; -1 2];
360 %! A = [A, zeros(size(A)); zeros(size(A)), -A];
361 %! b = rand(2*N,1); X = A\b; #X is the true solution
363 %! printf('System condition number is %g\n',cond(A));
364 %! # No preconditioner: the convergence is very slow!
366 %! [x, flag, relres, iter, resvec] = pcr(A,b,[],maxit);
367 %! title('Convergence history'); xlabel('Iteration'); ylabel('log(||b-Ax||)');
368 %! semilogy([0:iter],resvec,'o-g;NO preconditioning: absolute residual;');
371 %! # Test Jacobi preconditioner: it will not help much!!!
373 %! M = diag(diag(A)); # Jacobi preconditioner
374 %! [x, flag, relres, iter, resvec] = pcr(A,b,[],maxit,M);
376 %! semilogy([0:iter],resvec,'o-r;JACOBI preconditioner: absolute residual;');
379 %! # Test nonoverlapping block Jacobi preconditioner: this one should give
380 %! # some convergence speedup!
382 %! M = zeros(N,N);k=4;
383 %! for i=1:k:N # get k x k diagonal blocks of A
384 %! M(i:i+k-1,i:i+k-1) = A(i:i+k-1,i:i+k-1);
386 %! M = [M, zeros(size(M)); zeros(size(M)), -M];
387 %! [x, flag, relres, iter, resvec] = pcr(A,b,[],maxit,M);
388 %! semilogy([0:iter],resvec,'o-b;BLOCK JACOBI preconditioner: absolute residual;');
392 %! #solve small indefinite diagonal system
395 %! A = diag(linspace(-10.1,10,N)); b = ones(N,1); X = A\b; #X is the true solution
396 %! [x, flag] = pcr(A,b,[],N+1);
397 %! assert(norm(x-X)/norm(X)<1e-10);
402 %! #solve tridiagonal system, do not converge in default 20 iterations
403 %! #should perform max allowable default number of iterations
407 %! for i=1:N-1 # form 1-D Laplacian matrix
408 %! A(i:i+1,i:i+1) = [2 -1; -1 2];
410 %! b = ones(N,1); X = A\b; #X is the true solution
411 %! [x, flag, relres, iter, resvec] = pcr(A,b,1e-12);
413 %! assert(relres>0.6);
418 %! #solve tridiagonal system with 'prefect' preconditioner
419 %! #converges in one iteration
423 %! for i=1:N-1 # form 1-D Laplacian matrix
424 %! A(i:i+1,i:i+1) = [2 -1; -1 2];
426 %! b = ones(N,1); X = A\b; #X is the true solution
427 %! [x, flag, relres, iter] = pcr(A,b,[],[],A,b);
428 %! assert(norm(x-X)/norm(X)<1e-6);
429 %! assert(relres<1e-6);
431 %! assert(iter,1); #should converge in one iteration