MOPAC6.0 ESP, Again

From: mxm |-at-| biosym.com (Max Muir )
Subject: MOPAC6.0 ESP, Again
Date: Fri, 10 Sep 93 21:02:50 -0700
   Dear CyberChemists,
             Calcn. of ESP fitted charges with MOPAC6.0
             _________________________________________
   a collegue has been using MOPAC6.0's ESP option to get a set atom
   charges fitted (least squares) to a bunch of electrostatic potential
   points on Connolly surfaces at 1.4, 1.6,1.8 and 2.0 times the vdWs
   atom radius.  Old hat, yes, but he is unable to reproduce the values
   that appear in Kolman's paper (J. Comp. Chem., Vol. 11, No. 4, 431-
   439 (1990)) and neither am I.  Consider the case of formaldehyde
   (does anyone call it methanal?).  I'm using the default values for
   SCALE (1.4) NSURF(4) and SCINCR (0.2).  Here's my input deck . .
 GRAPH NODIIS MMOK ESP SLOPE=1.422 POTWRT NOINTER PRECISE
 I shot an elephant in my pyjamas the other day
  C    0.000000 0    0.000000 0    0.000000 0  0  0  0
  O    1.230000 1    0.000000 0    0.000000 0  1  0  0
  H    1.090000 1  120.000092 1    0.000000 0  1  2  0
  H    1.070000 1  120.000114 1  180.000000 1  1  2  3
   The fitted charges are . . (from .out)
             ELECTROSTATIC POTENTIAL CHARGES
        ATOM NO.    TYPE    CHARGE   SCALED CHARGE
           1          C     0.3897      0.5541
           2          O    -0.3613     -0.5137
           3          H    -0.0142     -0.0202
           4          H    -0.0142     -0.0202
             THE NUMBER OF POINTS IS:       375
             THE RMS DEVIATION IS:       0.7655
             THE RRMS DEVIATION IS:      0.0794
   While Kollman et al report (in figure 1) the following MNDO charges.
   Formaldehyde
       H          +0.006
       C          +0.339
       O          -0.352
   The unscaled fitted charges aren't too bad, but the H's have changed sign.
   The scaled fitted charges seem incredible and your humble correspondent
   does not believe them.  Have I missed something obvious?
   Yours faithfully,
   Max Muir