We attempt to simultaneously diagonalize two noisy (random
noise) versions
of diag(1:10)
and diag(1:10)^2
. Our initial
point is a small random perturbation of the identity.
>> !cp examples/simdiag/*.m . >> randn('state',0); >> [A,B] = noisy; >> parameters(A,B); >> Y0 = guess; >> [fn,Yn] = sg_min(Y0,'frcg','euclidean'); iter grad F(Y) flops 0 2.887592e+03 9.671138e+02 30940 1 1.419011e+03 2.411349e+02 343392 2 6.199732e+02 7.601495e+01 718792 3 3.051636e+02 3.300591e+01 1106535 4 2.538995e+02 2.078422e+01 1518380 5 1.440247e+02 1.258174e+01 1923273 6 1.189063e+02 9.500450e+00 2332153 7 9.279907e+01 7.309071e+00 2743897 8 8.483878e+01 6.093323e+00 3161294 9 7.763100e+01 4.117233e+00 3609412 10 7.448803e+01 3.220241e+00 4026633 11 6.404668e+01 2.328540e+00 4439092 .......(many iterations later)....... 169 2.196881e-06 4.010660e-07 67843218 170 2.309773e-06 4.010660e-07 68212219 171 2.435095e-06 4.010660e-07 68558476 172 1.763263e-06 4.010660e-07 68946596 173 2.081518e-06 4.010660e-07 69314166 174 1.251419e-06 4.010660e-07 69639992 175 1.474565e-06 4.010660e-07 70028565 176 8.949991e-07 4.010660e-07 70353929 177 8.506853e-07 4.010660e-07 70742988 178 7.154996e-07 4.010660e-07 71068090 179 8.003997e-07 4.010660e-07 71395085 180 3.164977e-07 4.010660e-07 71721113 181 3.701357e-07 4.010660e-07 72030590
Figure 9.6 shows the convergence curve for this run.