The models, above all, require a force field. With the exception of the all-atom model, each model uses a force field that is custom made for the particular molecular model. Here we restrict our discussion to the all-atom force fields, which are the most commonly used and generally considered to be the most accurate. Thus from here on, when we specify a force field, we mean an all-atom force field. As mentioned above there are a number of mature and well established force fields for modeling amino acids and nucleic acids.

The introduction of usable carbohydrate force fields to the AMBER (34, 35), CHARMM (36, 37), OPLS (38, 39), and GROMOS (40) force fields occurred after the force fields were already established and verified for proteins, nucleic acids, and lipids. Carbohydrate research is driven mainly by research in food science, and has largely concentrated on starches and glycosylated amino acids with little attention to cellulose. The small amount of unambiguous experimental data, especially structural data, contributes to the reluctance to model cellulose since the force fields cannot be verified as being appropriate for cellulose modeling in their current forms, and a more complex reparameterization may be necessary to reliably simulate cellulose. The molecular structures associated with cellulose are quite large in terms of current molecular modeling capabilities and require large computational resources. Although the current carbohydrate force fields have been carefully constructed for small molecules and carbohydrates, the force fields have not been tested extensively for such large structures in the same way that the minute details of force fields for proteins have been adjusted to reproduce known structures and known probabilities ofalpha helices and beta sheets. Few researchers are willing to apply serious intellectual or computational effort toward such a speculative endeavor. On the other hand, the same lack of unambiguous experimental data empowers modelers to simulate cellulose and its interactions to suggest possible structures and behaviors as well as eliminate highly unlikely ones and to attribute structural features to their underlying physics, even if the modeling is crude. The stage is set for a significant contribution of MD to the understanding of cellulose structure, function, and behavior as new experimental techniques and data are available, and larger and faster computers are accessible to MD modelers.