Help wanted with MOPAC

 Can anyone help me with a MOPAC user's problem.  The following system
 optimizes with great difficulty:
   C    0.00000000  0    0.0000000  0    0.0000000  0    0    0    0     -0.0166
   C    2.81240936  1    0.0000000  0    0.0000000  0    1    0    0     -0.1114
   C    1.40080247  1   59.9807428  1    0.0000000  0    1    2    0     -0.0758
   C    1.39578680  1   60.0724519  1 -175.6464885  1    1    2    3     -0.0750
   C    1.40657165  1   59.8002589  1   -4.8265486  1    2    1    3      0.0074
   C    1.40766029  1   59.8397867  1  178.8447129  1    2    1    3      0.0041
   H    1.09158307  1  178.4643089  1   75.5785133  1    1    2    3      0.0680
   H    1.09420894  1  119.8471439  1  173.0219276  1    3    1    2      0.0707
   H    1.09090199  1  120.3429052  1 -179.8159897  1    4    1    2      0.0713
   H    1.09021134  1  121.5310642  1 -177.3488763  1    5    2    1      0.0783
   H    1.09124131  1  121.0364375  1  178.8654572  1    6    2    1      0.0817
   C    1.50817055  1  176.8162974  1 -176.5943938  1    2    1    3      0.3420
   O    1.19541323  1  131.3383140  1 -157.6596572  1   12    2    5     -0.2820
   O    1.42321352  1  114.6569269  1   18.4221294  1   12    2    5     -0.1627
 The problem is this:  The odd electron can go into one of two M.O.'s.  If it
 goes into the lower energy M.O., then after Dewar's half-electron correction,
 the resulting state energy is higher than if it had gone into the higher energy
 M.O.  All the math seems correct, and the theory looks O.K.
   During the optimization, the odd electron goes into the lower energy M.O,
 and the geometry optimizes to the point where the other M.O. becomes the lower
 energy M.O.  The state energy then discontinuously rises, and the geometry
 optimizes so that the first M.O. becomes lower in energy once again.
 This is a frustrating system, as it appears to be un-optimizable.  After
 spending
 several days working on it, I now appeal for suggestions.  The constraints
 on suggestions are (1) the method used must be RHF, (2) the geometry must
 not be restrained (all parameters must be optimized).  I would prefer the
 EF method to be used for optimizing the geometry, but BFGS would be acceptable.
 Please send suggestions to me directly, and use as subject "C6H5O2".
 Jimmy Stewart