In quantum chemistry, the first step of most simulations will be to "relax" a molecular structure; this means to search the potential energy surface to find the minimal energy structure (i.e., the most stable structure). The relaxed structure is assumed to be an accurate model of the physical compound. Higher quality relaxations, typically those using a higher level of theory such as perturbation or coupled-cluster, make for a more realistic geometry, in principle. However, higher quality always translates to more "expensive" which, in scientific computation, refers to the amount of resources (time and computational hardware, usually) one must throw at a particular simulation.
Molecular dynamics is regarded as a very "cheap" method because it does not use an algorithm to search the energy landscape for a path to the minimal energy structure. Instead, the straight-forward approach of molecular dynamics is to integrate the forces applied to each atom directly using something like a Verlet or damped-Newtonian model. This is mathematically inefficient since you might not converge to the minimal energy very quickly but the method is computationally "cheap" so it is a good starting point, especially if you know your initial structure is very far from the correct geometry as I assumed for my calculation: Algorithms like conjugate gradient or BFGS are only most efficient when the structure is already close to a minimum in the energy landscape.
There is a lot more to be said about structure relaxation, the theory and the best way to do it in practice, and vibrational modes but I will leave that for another time.