The options available to describe the behavior of matter in silico resemble a ladder (
Figure 5.8), where each of the steps from the bottom up can be thought as representing different scales of ever-decreasing physical detail.
At the bottom of the ladder is the full electronic quantum mechanical description of matter: essentially exact numerical solution of the Schrodinger equation for the many-body wavefunction, using, for example, configuration interaction or renormalization group methods. This is only feasible for a handful of atoms at most because the complexity of the many-body wavefunction increases exponentially with the number of particles. If one is willing to accept a few simplifications, density functional theory (DFT) offers a very successful combination of accuracy and relatively low computational cost, and makes it possible to run simulations on hundreds of atoms as a matter of routine, and thousands of atoms on large supercomputers (Burke, 2012). Though DFT has become a widely accepted tool it is burdened with two important consequences of the inherent simplifications: the use of more complicated wave functions do not necessarily provide more accuracy and
there is no rigorous way of accounting for dispersion forces within a DFT treatment, which for an accurate description of fluid-phase equilibria is a crucial requirement. Larger scales must be sought. Up a rung of the ladder is a description of atoms or tightly bound chemically distinct groups, such as the CH
3 and CH
2 groups of
n-alkanes, using the simpler force fields of classical mechanics, which are commonly represented with spherical symmetrical pair potentials with Coulombic charged sites. At this level of abstraction one therefore considers the interaction sites to be made up of a collection of atoms and atom groups or even small molecules, e.g., carbon dioxide or benzene. Such a methodology is commonly referred to a treatment at the coarse grained (CG) level. The intermolecular interactions at this CG level are typically described with semiempirical analytical functions of the interseparation taken to represent the pairwise potential energy,
u(
r). It is common practice to estimate the parameters of the force fields by attempting to reproduce target structural (radial distribution function, structure factor, etc.), thermodynamic (second virial coefficient, liquid density,
vapor pressure, internal energy, enthalpy, etc.), or transport (viscosity, diffusivity, etc.) properties of the fluid or a combination of these properties. In an effort to resolve the dilemma of the degeneracy of macroscopic parameters for the intermolecular potential, it is then tempting to employ a more detailed, higher-resolution atomistic model (a step down the ladder) as the basis for determination of the parameters for CG models by means of an appropriate integration of the degrees of freedom which are to be removed. A review of current practice including a discussion of the key approaches to the simultaneous modeling of several time and/or length scales (
multiscale modeling) is the subject of a book (Voth, 2009) and themed journal issues (
Faller, 2009; Visscher et al., 2011). Also reviews on the topic have been published by Klein and Shinoda (2008), McCullagh et al. (2008), and Brini et al. (2013) to name just a salient few. It is at this level of abstraction that one may start envisaging the modeling of crude oils as the molecular level of representation must be coarsened with the complexity in order to allow for a tractable description of the molecular behavior. Models such as those employed to describe microphase separation, e.g., dissipative particle dynamics techniques (Groot and Warren, 1997), are representative examples of this next step up the ladder. These high-level CG models exhibit some of the consequences of the selective removal of degrees of freedom, echoed in issues of robustness, transferability, and representability. Further abstraction into more coarse grained models commonly removes all type of molecular descriptors and moves one into the realm of continuum models. The formal connection between these steps higher up our ladder is ever more sparse and very few models effectively bridge these gaps. This is an area of very dynamic and topical research.