To obtain accurate results in molecular dynamic (MD) simulations of glasses, it is essential to chose proper force fields (FF), proper length of the simulation cell, and proper cooling cycle to generate 300 K structures from liquid. Herein we establish guidelines for these choices. We find that the MS-Q force field (FF) and the Molecular Simulations Incorporated (MSI) glass FF both lead to agreement with the radial distribution function (RDF) from small angle neutron scattering (SANS) experiments. We find that the simulation cell should contain about 3000 atoms to obtain run to run density variation less than 1% and consistency in the first two RDF peaks of 0.001 nm. A cell of 648 atoms gives run to run density variation of up to 5% and consistency in the first three RDF peaks of 0.001 nm. We find that a good compromise between accuracy and reproducibility of results and simulation time is to start with NVT dynamics at 8000 K followed by cooling to room temperature at a rate of 100 K/2 ps.
