Truly predictive modeling is a rarity in chemistry and biology. This is true for both explicit calculational techniques (quantum chemistry, force fields) and for implicit methods such as quantitative structure-property relationship (QSPR) models and other interpolation techniques. It is true that, for sufficiently small molecules for which a single-determinant Hartree-Fock wavefunction provides a good description, CCSD(T) [1, 2] calculations using an adequate basis set produce gas-phase results that usually rival or better experimental accuracy. This is very positive but of limited use in the impure, finite-temperature condensed-phase world in which many important processes take place. The automobile industry is used to a situation in which the results of crash simulations agree very closely with experimental crash testsso much so that the latter no longer form part of the design process. The fields of chemistry and biology, however, are a very long way from this idyllic situation. This is perhaps surprising, because the physics behind the processes and the systems considered is believed to be well known, such that their accurate modeling should be possible. This aim of this chapter is to analyze the current situation, to identify weaknesses and opportunities in chemical and life-science modeling, and maybe even to suggest some improvements in what can be done. The first stage, however, is to consider the nature of both models and modeling.
Models and ModelingModels are exactly that. Unlike CCSDT, they are not an accurate and defined truncation of a physically exact theory, but are rather a simplified representation of a complex object that is intended to behave like the object itself. Behave is a key concept here. It is well known to the multiscale community [3] that a coarse-grain version of a more complex force field can only reproduce a limited set of the Modeling of Molecular Properties, First Edition. Edited by Peter Comba.