Add error checking for the input GB radii in the GBn methods. The radii must be

between 1 and 2 Angstroms, or they fall outside the bounds of the neck lookup
tables.

At least on my Mac, this results in a C++ exception that does not get translated
into a Python exception (i.e., it is not catchable and crashes the application)

wrappers/python/simtk/openmm/app/internal/amber_file_parser.py

@@ -564,6 +564,12 @@ class PrmtopLoader(object):
                     screen[i] = 0.602256336067
                 else:
                     screen[i] = 0.5
+                # radii is currently in Angstroms right now. GBn lookup tables
+                # only support radii between 1.0 and 2.0
+                if radii[r] < 1.0 or radii[r] > 2.0:
+                    raise ValueError('GBn requires intrinsic radii between 1 and '
+                                     '2 Angstroms (%.3f found for atom %d)' %
+                                     (radii[r], i))
         if gbmodel == 'GBn2':
             if elements is None:
                 raise Exception('GBn2 model requires element information')
@@ -598,6 +604,12 @@ class PrmtopLoader(object):
                     alpha[i] = 1.0
                     beta[i] = 0.8
                     gamma[i] = 4.85
+                # radii is currently in Angstroms right now. GBn lookup tables
+                # only support radii between 1.0 and 2.0
+                if radii[r] < 1.0 or radii[r] > 2.0:
+                    raise ValueError('GBn2 requires intrinsic radii between 1 and '
+                                     '2 Angstroms (%.3f found for atom %d)' %
+                                     (radii[r], i))
         lengthConversionFactor = units.angstrom.conversion_factor_to(units.nanometer)
         if gbmodel == 'GBn2':
             for rad, scr, alp, bet, gam in zip(radii, screen, alpha, beta, gamma):