Coarse-grain models are crude models compared to more detailed or fine-grain models. The figure illustrates this process in the case of a DNA molecule, where models at three levels of resolution are shown. One can coarse-grain at even more crude levels than the mesoscopic level shown. What is important to recognize is that each model is optimal for studying certain scientific questions, and hopeless for studying other questions.
A DNA model at the resolution of electron orbitals can give extremely detailed insights into chemical structure e.g. bond vibration spectra. One advantage of QM models is that interactions are automatically dictated by the electronic interactions. However, the main limitation is that since ab initio quantum chemical methods are extremely computationally expensive only a fairly small number of atoms/electrons can be studied.
An atomistic DNA model describes DNA at the level of atoms, where hydrogens can either be present or merged with the nearest carbon. Atomistic simulations allows considerably larger systems to be studied compared to QM models, and they also allow a solvent with water and counter ions to be modelled. Atomistic simulations require a force field to be defined, which defines the force between any type of atom with any other type of atom or more accurately atom in a chemical neighbourhood. The quality of the force field defines how meaningful the results are. Since the typical time step of atomistic simulations are on the femtosecond scale, it becomes rather expensive to reach e.g. microsecond time scales.
The mesoscopic DNA model is clearly very crude in terms of chemical detail. But this also translates into a model that is very fast to simulate, or alternatively that we can simulate very long molecules. Since coarse-graining also smoothen the energy landscape, the time step can be much larger in a coarse-grained simulation. How to develop coarse-grained force fields is routine for reproducing static equilibrium properties, but remains an open question for non-equilibrium or dynamical properties. Taken together, the coarse-grained model allows us to study very large systems for very long times, much longer than is possible with atomistic simulations. Hence this opens the possibility of studying not electronic, chemical or molecular properties, but instead very large scale properties such as static and dynamic properties in materials where DNA molecules plays a part.
In my polymer systems, I routinely use 1-10M particles, 20fs time steps, box sizes of 100x100x100nm and can simulate dynamics from femtosecond to microsecond time scales at a cost of about one CPU core year, or a week of simulation time on 4 standard ABACUS 2.0 nodes.