Molecular dynamics is a tool for simulating the interactions between atoms or molecules. It utilizes basic Newtonian physics, by treating bonds and angles as springs, to approximate the motions of molecular structures. Because it is computational, it allows researchers to test hypotheses with unprecedented resolution, whereas standard experimental techniques can only look at more macroscopic properties.
Molecular dynamics depends on the ergodic hypothesis, where it is assumed that the averages of statistical ensembles are the same as time averages. This justification is rooted deeply in statistical mechanics and allows researchers to run long simulations and make conclusions about the properties of a system based on this simulation. As such, “sampling” is a critical issue in MD. Researchers are constantly debating whether a system has been sufficiently sampled to provide both statistically significant results and to see all structural conformations.
Types of MD
In traditional molecular dynamics, Newtonian physics and molecular mechanics is used to create the motion of atoms, using empirically derived potentials to describe attributes such as bonds, angles, torsions, dihedrals, van der Waals radii, and electrostatics. These potentials are called “force fields”. The main computational expense with traditional MM comes with the non-bonded interactions, specifically the van der Waals and electrostatics. These interactions involve every pairwise combination of particles, as opposed to interactions such as bonds which only involve neighboring atoms.
Molecular mechanics represents all bonded interactions as simple harmonic springs. While this makes the calculation cheap and simple, it removes the ability to break bonds. This limits the applications of molecular mechanics to general system dynamics and movement, rather than simulating chemical reactions.
Semi-empirical methods take advantage of the accuracy of quantum mechanics to make molecular dynamics more powerful. Matrix representations from quantum mechanics are used to determine energy contributions from electron orbitals and then utilize these energies to move the system. In polarizable methods, induced dipoles are introduced through various means including fluctuating charges, and allow particles to change as the environment around the particles change.
One method for speeding up molecular dynamics involves the use of coarse-graining methods. While many of the types listed above try to include more information to make molecular dynamics more realistic, coarse-graining removes some of the accuracy to speed up the calculations. In coarse-graining, some sort of vast approximation is made to greatly increase simulation speed. For instance, some united atom methods represent groups of atoms with one large pseudo-atom that attempts to represent the overall properties of the represented atoms.
The future of field appears to be in polarizable MD simulations, as they have been shown, in general, to more accurately represent true physical systems. However, they are fairly expensive compared to traditional MD. Because of this, many groups are pioneering the use of coarse-graining to speed up calculations, while allowing for polarizability to drastically increase accuracy.
Uses of MD
Molecular dynamics has found a great deal of use in the simulation of biological macromolecules. The motions of molecules can be used to predict protein folding, secondary structure, and large scale structural changes of molecules such as channels in biological membranes.
The most well-known example of molecular dynamics is through the work of Vija Pande at Stanford as part of the Folding@Home project. This project took advantage of idle computer time on hundreds of thousands of personal computers to perform a folding simulation of the villin headpiece. This simulation involved 20,000 atoms for a simulation time of 500 microseconds, a huge computational effort.
Since the 1970s, molecular dynamics has been used heavily in material science. These researchers use MD to examine the physical properties of specific environments or nano-technologies. Research scientists in crystallography and NMR utilize MD to refine their solved protein structures.
