Essentials of Computational Chemistry

156 5 SEMIEMPIRICAL IMPLEMENTATIONS OF MO THEORY
Parameter AM1 AM1-SRP
C
U s
U p
b s
bp O
Us U p
b s
b p
−52.03
−39.61
−15.72
−7.72
−97.83
−78.26
−29.27
−29.27
−49.85
−40.34
−16.91
−9.19
−99.18
−80.76
−29.00
−29.25
H
C O
H
H
H
+ H •
H 2O
C •
O
H H
H
Source ∆E rxn D e (C−H) D e (H−H)
AM1
AM1-SRP
Expt.
−28.0
−4.9
−5.1
81.4
104.4
104.4
109.4
109.4
109.5
+ H 2
Figure 5.2 AM1 and AM1-SRP parameters (eV) optimized to reproduce the C–H bond dissociation
energy of methanol, the H–H bond dissociation energy of hydrogen, and the experimental energy for
the illustrated hydrogen-atom transfer (kcal mol −1 ). Note that in all cases but one, the magnitude of
the parameter change on going from AM1 to AM1-SRP is less than 10 percent
The concept is best illustrated with an example. Chuang et al. (1999) used an AM1-
SRP model to study the hydrogen-atom-mediated destruction of organic alcohols in water.
As illustrated in Figure 5.2, the AM1 model itself makes a very poor prediction for the
dissociation energy of the C–H bond in methanol, and hence for the reaction exothermicity.
By minor adjustment of a few of the AM1 parameters, however, the SRP model gives good
agreement with experiment. The resulting SRP model in this case was used as a very efficient
QM method for generation of a PES from which tunneling contributions to the reaction rate
constant could be estimated (see Section 15.3). The very large number of QM calculations
required to generate the PES made use of an SRP model preferable to more complete levels
of electronic structure theory like those discussed in Chapter 7. Ridder et al. (2002) followed
a similar protocol in determining SRP AM1 sulfur parameters so as to study the dynamics of
the conjugation of glutathione to phenanthrene-9,10-oxide as catalyzed by a rat glutathione
S-transferase enzyme (the enzyme was treated using molecular mechanics). Again, the very
large number of quantum calculations required during the dynamics made a semiempirical
model like AM1 an attractive choice.
A more global SRP reparameterization has been described by Sherer, York, and Cramer
(2003). In this case, select PM3 parameters for H, C, N, and O were modified to improve
the performance of the resulting SRP model, named PM3BP, for the computation of basepairing
energies between hydrogen-bonded nucleic-acid bases. The PM3BP model has a
root-mean-square error of about 1.5 kcal mol −1 for 31 such base pairing energies compared
to either experiment or well benchmarked higher-level theoretical calculations. This compares
to RMS errors of about 11, 6, and 6 kcal mol −1 for MNDO, AM1, and PM3 over the same