X-Git-Url: http://gitweb.michael.orlitzky.com/?p=octave.git;a=blobdiff_plain;f=tests%2Fpreconditioned_conjugate_gradient_method_tests.m;h=6a11a63efd743df7fa7a73349de80fff3f488827;hp=d44ee421570ead5908c49a14a959650c8cb88495;hb=36afdc59f58ac6a8a1fe366e1cde070db24905ac;hpb=84b8fb9002d091f84d0205e923c3989d0138ec9e

diff --git a/tests/preconditioned_conjugate_gradient_method_tests.m b/tests/preconditioned_conjugate_gradient_method_tests.m
index d44ee42..6a11a63 100644
--- a/tests/preconditioned_conjugate_gradient_method_tests.m
+++ b/tests/preconditioned_conjugate_gradient_method_tests.m
@@ -1,3 +1,9 @@
+## Used throughout. The PCGM uses the infinity norm as the stopping
+## condition, so we had better also.
+max_iterations = 100000;
+tolerance = 1e-11;
+
+## First a simple example.
 A = [5,1,2; ...
      1,6,3; ...
      2,3,7];
@@ -6,37 +12,70 @@ M = eye(3);
 b = [1;2;3];
 x0 = [1;1;1];
 
-cgm  = conjugate_gradient_method(A, b, x0, 1e-6, 1000);
-pcgm = preconditioned_conjugate_gradient_method(A, M, b, x0, 1e-6, 1000);
-diff = norm(cgm - pcgm);
+cgm  = conjugate_gradient_method(A, b, x0, tolerance, max_iterations);
+pcgm = preconditioned_conjugate_gradient_method(A, ...
+						M, ...
+						b, ...
+						x0, ...
+						tolerance, ...
+						max_iterations);
+diff = norm(cgm - pcgm, 'inf');
 
 unit_test_equals("PCGM agrees with CGM when M == I", ...
 		 true, ...
-		 norm(diff) < 1e-6);
+		 diff < 2*tolerance);
 
-pcgm_simple = simple_preconditioned_cgm(A, M, b, x0, 1e-6, 1000);
-diff = norm(pcgm_simple - pcgm);
+pcgm_simple = simple_preconditioned_cgm(A, M, b, x0, tolerance, max_iterations);
+diff = norm(pcgm_simple - pcgm, 'inf');
 
 unit_test_equals("PCGM agrees with SimplePCGM when M == I", ...
 		 true, ...
-		 norm(diff) < 1e-6);
+		 diff < 2*tolerance);
 
 ## Needs to be symmetric!
 M = [0.97466, 0.24345, 0.54850; ...
      0.24345, 0.73251, 0.76639; ...
      0.54850, 0.76639, 1.47581];
 
-pcgm = preconditioned_conjugate_gradient_method(A, M, b, x0, 1e-6, 1000);
-diff = norm(cgm - pcgm);
+pcgm = preconditioned_conjugate_gradient_method(A, ...
+						M, ...
+						b, ...
+						x0, ...
+						tolerance, ...
+						max_iterations);
+diff = norm(cgm - pcgm, 'inf');
 
 unit_test_equals("PCGM agrees with CGM when M != I", ...
 		 true, ...
-		 norm(diff) < 1e-6);
+		 diff < 2*tolerance);
 
 
-pcgm_simple = simple_preconditioned_cgm(A, M, b, x0, 1e-6, 1000);
-diff = norm(pcgm_simple - pcgm);
+pcgm_simple = simple_preconditioned_cgm(A, M, b, x0, tolerance, max_iterations);
+diff = norm(pcgm_simple - pcgm, 'inf');
 
 unit_test_equals("PCGM agrees with Simple PCGM when M != I", ...
 		 true, ...
-		 norm(diff) < 1e-6);
+		 diff < 2*tolerance);
+
+
+# Test again Octave's pcg() function.
+for n = [ 5, 10, 25, 50, 100 ]
+  A = random_positive_definite_matrix(5, 1000);
+  C = random_positive_definite_matrix(5, 1000);
+  M = C*C';
+
+  # Assumed by Octave's implementation when you don't supply a
+  # preconditioner.
+  x0 = zeros(5, 1);
+  b  = unifrnd(-1000, 1000, 5, 1);
+  [o_x, o_flag, o_relres, o_iter] = pcg(A, b, tolerance, max_iterations, C, C');
+  [x, k] = preconditioned_conjugate_gradient_method(A,
+						    M,
+						    b,
+						    x0,
+						    tolerance,
+						    max_iterations);
+  diff = norm(o_x - x, 'inf');
+  msg = sprintf("Our PCGM agrees with Octave's, n=%d.", n);
+  unit_test_equals(msg, true, diff < 2*tolerance);
+end