Here are some of the limitations and assumptions in classical molecular mechanics/dynamics.
Classical approximation
Molecular mechanics is based on classical methods to represent the system (using the force field) and propagate it forward in time (using Newton’s equations of motion). We usually don’t consider the dynamicity of electron densities and therefore can’t fully account for processes such as hydrogen bonding interactions, bond making and breaking, and excited states.*
Ergodic hypothesis
[details TBD]
Sampling
[details TBD]
Force field
When writing lab reports for lab classes in undergrad, we had to include a discussion of our results and explain why they might be as expected. The typical go-to answer was “human error.” (I’ve observed this as a TA, too, when grading post-lab assignments of gen chem students.) Then you give a laundry list of reasons of things you possibly did wrong — “I put in water instead of ethanol”, “my yield was low because I spilled a lot”, etc. The apparent go-to answer in a paper that uses MD simulations is “force field accuracy.” That’s not to say it’s not a valid reason! The accuracy of the simulation depends on the the accuracy of the force field comprising those simulations. And, force fields are inherently approximations of true reality.
There’s not one standard way to develop force fields, so then you also have variations in results if comparing different force fields. Which one is correct? Hard to say. We don’t even know how the uncertainty in the force field to adequately know the uncertainty in our results. Many in the field would say that we need better force fields to yield more reliable simulations.
Footnotes
* The force field represents a potential energy surface which is typically the ground state, and we’re usually not interested in excited states in MD simulations anyway. This means you wouldn’t use classical MD to study things like chemical reactions that involve bond breaking/making and multiple electronic states!
References