print diagnostic messages if anything goes wrong during minimization

nifty/minimization/conjugate_gradient.py

@@ -18,7 +18,7 @@
 
 from __future__ import division
 from .minimizer import Minimizer
-from .. import Field
+from .. import Field, dobj
 from ..utilities import general_axpy
 
 
@@ -80,10 +80,12 @@ class ConjugateGradient(Minimizer):
             q = energy.curvature(d)
             ddotq = d.vdot(q).real
             if ddotq == 0.:
+                dobj.mprint("Error: ConjugateGradient: ddotq==0.")
                 return energy, controller.ERROR
             alpha = previous_gamma/ddotq
 
             if alpha < 0:
+                dobj.mprint("Error: ConjugateGradient: alpha<0.")
                 return energy, controller.ERROR
 
             general_axpy(-alpha, q, r, out=r)
@@ -98,8 +100,9 @@ class ConjugateGradient(Minimizer):
 
             gamma = r.vdot(s).real
             if gamma < 0:
-                raise RuntimeError(
+                dobj.mprint(
                     "Positive definiteness of preconditioner violated!")
+                return energy, controller.ERROR
             if gamma == 0:
                 return energy, controller.CONVERGED
 

nifty/minimization/descent_minimizer.py

@@ -20,6 +20,7 @@ from __future__ import division
 import abc
 from .minimizer import Minimizer
 from .line_search_strong_wolfe import LineSearchStrongWolfe
+from .. import dobj
 
 
 class DescentMinimizer(Minimizer):
@@ -89,9 +90,12 @@ class DescentMinimizer(Minimizer):
             f_k_minus_1 = energy.value
 
             if new_energy.value > energy.value:
+                dobj.mprint("Error: Energy has increased")
                 return energy, controller.ERROR
 
             if new_energy.value == energy.value:
+                dobj.mprint(
+                    "Warning: Energy has not changed. Assuming convergence...")
                 return new_energy, controller.CONVERGED
 
             energy = new_energy

nifty/minimization/gradient_norm_controller.py

@@ -70,6 +70,8 @@ class GradientNormController(IterationController):
         # Are we done?
         if self._iteration_limit is not None:
             if self._itcount >= self._iteration_limit:
+                dobj.mprint(
+                    "Warning:Iteration limit reached. Assuming convergence")
                 return self.CONVERGED
         if self._ccount >= self._convergence_level:
             return self.CONVERGED

nifty/minimization/line_search_strong_wolfe.py

@@ -105,6 +105,7 @@ class LineSearchStrongWolfe(LineSearch):
         phi_0 = le_0.value
         phiprime_0 = le_0.directional_derivative
         if phiprime_0 >= 0:
+            dobj.mprint("Error: search direction is not a descent direction")
             raise RuntimeError("search direction must be a descent direction")
 
         # set alphas

nifty/operators/inversion_enabler.py

@@ -17,7 +17,8 @@
 # and financially supported by the Studienstiftung des deutschen Volkes.
 
 from ..minimization.quadratic_energy import QuadraticEnergy
-from ..field import Field
+from ..minimization.iteration_controller import IterationController
+from ..field import Field, dobj
 
 
 class InversionEnabler(object):
@@ -30,7 +31,9 @@ class InversionEnabler(object):
     def _operation(self, x, op, tdom):
         x0 = Field.zeros(tdom, dtype=x.dtype)
         energy = QuadraticEnergy(A=op, b=x, position=x0)
-        r = self._inverter(energy, preconditioner=self._preconditioner)[0]
+        r, stat = self._inverter(energy, preconditioner=self._preconditioner)
+        if stat != IterationController.CONVERGED:
+            dobj.mprint("Error detected during operator inversion")
         return r.position
 
     def _times(self, x):