1
$\begingroup$

From what I understand, in a molecular dynamics simulation the number of molecules included is limited by the computational power available.

In order to simulate macroscopic properties, one needs to ensure that the particles experience forces as if they were in the bulk, so the system is replicated in all directions through the application of a periodic boundary condition.

How is the creation of this larger, periodic system any more computationally efficient than just simulating a non-periodic system of similar size?

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ A non-periodic system will likely suffer from many particles being on the surface of the system. Consider 1000 particles in a cube: almost half of them will be on the surface. Their characteristics will be different from a particle in bulk. $\endgroup$
    – TAR86
    Apr 5 '19 at 4:03
2
$\begingroup$

The application of periodic boundary condition does not mean that the system is explicitly replicated. You only deal with the particles within the unit cell (the box that you replicate), but you let them interact with the replicated ones outside of this unit cell.

Consider a simple 1-dimensional case with three atoms in the unit cell: $\ce{A-B-C}$, and lets assume we only need to treat nearest-neighbor-interactions. So the "monomer" has only 2 interactions: AB and BC.

One option to simulate the macroscopic system would be to explicitly deal with $\ce{A-B-C-A-B-C-A-B-C-A-B-C}$ (or even longer), but this requires a lot of atoms with accordingly many interactions between them. And even then both ends of the chain would be missing one neighbor.

Using periodic boundary conditions, one explicitly treats $\ce{A-B-C}$ only. But this time we will allow $\ce{C}$ to interact with $\ce{A}$, since the periodic boundary conditions implicitly tell us, there is another $\ce{C}$ to the left. Similar, there is an implicit $\ce{A}$ to the right of $\ce{C}$. Compared to the monomer from above, we have again 3 atoms, but now there are 3 interactions to be considered: AB, BC and AC.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Just be aware that each entity will have multiple, in fact an infinity, of images and technically you should sum over the interaction with each and every one with the reference particle - but in practice for most force laws the potential decays sufficiently quickly that you can truncate that summation, most commonly just including the nearest image. The big exception is Coulomb interactions where special methods should be used - the magic word here is "Ewald" $\endgroup$
    – Ian Bush
    Apr 5 '19 at 13:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.