KrylovSolver relative tolerance

Question

I figured that if I set up a KrylovSolver like:

solver = KrylovSolver("gmres", "ml_amg")
solver.parameters["relative_tolerance"] = 0.75
solver.parameters["maximum_iterations"] = 200
solver.parameters["monitor_convergence"] = True

however, the relative tolerance is below 0.75 but the solver didn't stop ... (yes, I know that 0.75 is probably way too high).

194 KSP preconditioned resid norm 5.467875874924e+01 true resid norm 4.616074188122e-03 ||r(i)||/||b|| 6.662113600126e-01
195 KSP preconditioned resid norm 5.467875739536e+01 true resid norm 4.616078004382e-03 ||r(i)||/||b|| 6.662119107914e-01
196 KSP preconditioned resid norm 5.467875604046e+01 true resid norm 4.616077635943e-03 ||r(i)||/||b|| 6.662118576167e-01
197 KSP preconditioned resid norm 5.467875541936e+01 true resid norm 4.616092554551e-03 ||r(i)||/||b|| 6.662140107334e-01
198 KSP preconditioned resid norm 5.467875537716e+01 true resid norm 4.616093612623e-03 ||r(i)||/||b|| 6.662141634388e-01
199 KSP preconditioned resid norm 5.467875439581e+01 true resid norm 4.616148455643e-03 ||r(i)||/||b|| 6.662220786155e-01
200 KSP preconditioned resid norm 5.467875232867e+01 true resid norm 4.616110502291e-03 ||r(i)||/||b|| 6.662166010272e-01

Answer 1

It's PETSc I take it? I haven't used the KrylovSolver object, but here are some suggestions:
Add the line:

PETScOptions.set('ksp_rtol', '.75')

To the beginning of the file. Sometimes what's in the parameters doesn't seem to get transferred over to the solver.
You can also try the "PETScKrylovSolver", which I use and with inputted tolerances...
BTW: Are you sure you're checking the correct 'norm'? i.e.: the true norm vs the preconditioned norm? From your output, they are quite different.

Comments

Yes, it is PETSc ... I should probably read the documentation more carefully. I think that I just copied someone's code and liked how the option looked. Thanks!