Semiempirical methods accuracy

A highly readable account of early efforts to apply the independent-particle approximation to problems of organic chemistry. Although more accurate computational methods have since been developed for treating all of the problems discussed in the text, its discussion of approximate Hartree-Fock (semiempirical) methods and their accuracy is still useful. Moreover, the view supplied about what was understood and what was not understood in physical organic chemistry three decades ago is... 

Jones et al. [144,214] used direct dynamics with semiempirical electronic wave functions to study electron transfer in cyclic polyene radical cations. Semiempirical methods have the advantage that they are cheap, and so a number of trajectories can be run for up to 50 atoms. Accuracy is of course sacrificed in comparison to CASSCF techniques, but for many organic molecules semiempirical methods are known to perform adequately. 

In order to obtain the best accuracy results as quickly as possible, it is often advantageous to do two geometry optimizations. The first geometry optimization should be done with a faster level of theory, such as molecular mechanics or a semiempirical method. Once a geometry close to the correct geometry has been obtained with this lower level of theory, it is used as the starting geometry for a second optimization at the final, more accurate level of theory. 

A basis set is a set of functions used to describe the shape of the orbitals in an atom. Molecular orbitals and entire wave functions are created by taking linear combinations of basis functions and angular functions. Most semiempirical methods use a predehned basis set. When ah initio or density functional theory calculations are done, a basis set must be specihed. Although it is possible to create a basis set from scratch, most calculations are done using existing basis sets. The type of calculation performed and basis set chosen are the two biggest factors in determining the accuracy of results. This chapter discusses these standard basis sets and how to choose an appropriate one. 

Another related issue is the computation of the intensities of the peaks in the spectrum. Peak intensities depend on the probability that a particular wavelength photon will be absorbed or Raman-scattered. These probabilities can be computed from the wave function by computing the transition dipole moments. This gives relative peak intensities since the calculation does not include the density of the substance. Some types of transitions turn out to have a zero probability due to the molecules symmetry or the spin of the electrons. This is where spectroscopic selection rules come from. Ah initio methods are the preferred way of computing intensities. Although intensities can be computed using semiempirical methods, they tend to give rather poor accuracy results for many chemical systems. 

In most cases the chemist only needs differences of values and/or relative estimations in comparison with a standard. The inaccuracies introduced by semiempirical methods with its relatively drastic approximations can be limited by applying the differences causing the calculated values to possess suitable accuracy. 

The usefulness of quantum-chemical methods varies considerably depending on what sort of force field parameter is to be calculated (for a detailed discussion, see [46]). There are relatively few molecular properties which quantum chemistry can provide in such a way that they can be used directly and profitably in the construction of a force field. Quantum chemistry does very well for molecular bond lengths and bond angles. Even semiempirical methods can do a good job for standard organic molecules. However, in many cases, these are known with sufficient accuracy a C-C single bond is 1.53 A except under exotic circumstances. Similarly, vibrational force constants can often be transferred from similar molecules and need not be recalculated. 

Equation (4-5) can be directly utilized in statistical mechanical Monte Carlo and molecular dynamics simulations by choosing an appropriate QM model, balancing computational efficiency and accuracy, and MM force fields for biomacromolecules and the solvent water. Our group has extensively explored various QM/MM methods using different quantum models, ranging from semiempirical methods to ab initio molecular orbital and valence bond theories to density functional theory, applied to a wide range of applications in chemistry and biology. Some of these studies have been discussed before and they are not emphasized in this article. We focus on developments that have not been often discussed. 

A useful compromise between speed and accuracy is provided by semiempirical methods. In this context, semiempirical methods can be used as fitting tools rather than predictive methods [76], Optimization of the semiempirical parameters to reproduce experimental or high-level ah initio results, for a specific reaction, can yield an accurate potential for the problem of interest at relatively low cost [77]. This approach has been shown to be successful in a QM/MM framework also [78] and is a powerful tool in the accurate study of enzyme reactions [57, 79] (see also Section 6.5.2). 

In an overall assessment, the established semiempirical methods perform reasonably for the molecules in the G2 neutral test set. With an almost negligible computational effort, they provide heats of formation with typical errors around 7 kcal/mol. The semiempirical OM1 and OM2 approaches that go beyond the MNDO model and are still under development promise an improved accuracy (see Table 8.1). 

In our own validation sets, experimental heats of formation are preferentially taken from recognized standard compilations [38-40]. If there are enough experimental data for a given element, we normally only use reference values that are accurate to 2 kcal/mol. If there is a lack of reliable data, we may accept experimental heats of formation with a quoted experimental error of up to 5 kcal/mol. This choice is motivated by the target accuracy of the established semiempirical methods. If experimental data are missing for a small molecule of interest, we consider it legitimate [18] to employ computed heats of formation from high-level ab initio methods as substitutes. 

For a complete picture, however, some critical remarks have to be added. For medium-sized molecules, the obviously successful combination of theoretical prediction and experimental verification rather should be considered as an individually tailored interplay, which is based on numerous assumptions i.e. so-called chemical intuition. Tobeginwith, any hypersurface design confined to a few selected coordinates will only cover the aspects introduced thereby. Even several independent cuts through the (3n-6) dimensional hyperspace cannot unravel the complexity of a molecule in a specific molecular state. And although semiempirical methods - especially MNDO (10)- are not only fast but to quite an extent also reliable, their numerical accuracy should not be overstressed. Thus semiempirical hypersurfaces might better be considered as supplying trends in essential features, which can be compared to and correlated with measurement data to add the numerical accuracy, and, in return, to test and to substantiate all underlying assumptions. 

Semiempirical methods, on the other hand, utilize minimum basis sets to speed up computations, and the loss in rigor is compensated by the use of experimental data to reproduce important chemical properties, such as the heats of formation, molecular geometries, dipole moments, and ionization potentials (Dewar, 1976 Stewart, 1989a). As a result of their computational simplicity and their chemically useful accuracy, semiempirical methods are widely used, especially when large molecules are involved (see, for example, Stewart, 1989b Dewar et al., 1985 Dewar, 1975). 

This is probably a natural consequence of the high accuracy and high speed of MM calculations. However, in recent years the ease of use, accuracy, and generality of semiempirical methods have improved considerably. While MM techniques should continue to be used for the study of ground state systems, carbohydrate chemists should be aware of the potential of semiempirical methods as a research tool, particularly for the study of reactions. 

Extensive tests have been carried out to establish the reliability of quantum-chemical schemes for metal oxide investigations. This includes schemes at a variety of levels of sophistication suitable for calculations of very large systems. In particular density functional methods offer good possibilities to treat sufficiently large systems to be applicable to many central problems in the field of photoelectrochemistry with reasonable accuracy and at very competitive computational costs. Semiempirical methods still offer a last possibility to perform reasonably accurate calculations on nanostructured systems containing several hundred atoms where first principles methods still cannot be applied routinely. 

These RM1 barriers do not represent any improvement over those from AMI and PM3, but the reaction energies are all within 10 kJ mol-1 of the experimental, while some of the AM1/PM3 errors are as much as —41 (HNC, PM3). Despite the lack of quantitative accuracy, semiempirical methods have been used fairly frequently in recent years to study transition states in biochemical reactions, because of the large molecules involved [97]. 

---