Many properties seem to need a special experiment. However, they come straight out of an ordinary equilibrium run, if you know which fluctuation or correlation to measure.
A Langevin thermostat adds friction, which removes energy, and random kicks, which add it. The two are not independent choices. The fluctuation-dissipation theorem fixes the size of the kicks from the friction and the temperature.
If the ratio is wrong, the system settles at the wrong temperature. If it is right, the friction only changes how quickly things move, not where they end up. The same balance sits inside the DPD thermostat, as explained on the DPD page.
Try: set s to 1.4, and the measured temperature doubles. Then, put s back to 1 and change the friction over its whole range. The distribution stays the same, and only the motion changes.
The pressure in a simulation has two parts. The momenta carry the ideal gas part, and the forces between particles carry the rest.
Here, d is the dimension. At low density, the virial part is small. In a dense liquid, both parts are large and nearly cancel, so the pressure of a liquid fluctuates by hundreds of bar in an atomistic run and needs long averages. Cutting the potential also changes the pressure, which is why codes add a tail correction.
Try: step through several densities at kT = 1, and then lower kT to 0.5 and repeat. The equation of state bends below the ideal gas line, and at low T, the pressure becomes negative in the region where gas and liquid coexist.
Transport coefficients are integrals of equilibrium time correlation functions. The diffusion coefficient is the integral of the velocity autocorrelation function, the viscosity is the integral of the stress autocorrelation, and the conductivity is the integral of the current autocorrelation.
This gives the same D as the slope of the mean squared displacement, because it is the same information read in another way. In a dilute gas, the memory decays as a simple exponential. In a dense liquid, particles bounce off their neighbors, and the function dips below zero.
In 2D, the correlation function has a slow tail, so the integral keeps creeping up with the cutoff time. In 3D, it converges, but the finite box still lowers D, and this correction matters when comparing with experiment.
Try: compare densities 0.1 and 0.7. At high density, the negative dip appears, and D drops sharply.