Computational vs Statistical Physics

Statistical physics was created by Boltzmann (1844-1906) in an attempt to explain observed irreversibility of thermodynamic processes as a necessary evolution from more ordered/less probable to less ordered/more probable states in a microscopic particle-collision model of a gas. This was captured in Boltzmann's macroscopic equations derived from an assumption of molecular chaos (StossAnzahlAnsatz) stating that particle velocities prior to collision are uncorrelated. Boltzmann's H-theorem states that a gas left alone will approach a uniform rest state with a Maxwellian velocity distribution. 

Statistical physics is based on some assumption of statistical nature, such as molecular chaos, to be compared with computational physics where the evolution of a gas as a collection of colliding particles is simulated simply by computing the trajectories of all particles subject to collision with chaos/unordered motion as an emergent phenomenon without any assumption. 

One can argue that computational physics is real physics because particle trajectories subject to Newton's laws of motion = real physics,  are computed. On the other hand, statistical physics is not real physics in the sense that real physics cannot do statistics and decide to evolve according to an assumption of molecular chaos.  

On the other hand it is possible for human beings to do statistics by computing mean values and standard deviations in particle-collision models. 

Statistical physics was developed before the computer when computational physics could not deliver. Today with the computer computational physics can answer the questions posed in statistical physics, see Euler Right! showing physics emerging in a discrete finite element model by computation. 

PS The classical approach is to derive a continuum model in the form of a partial differential equation from a particle model. A computational model can then be derived by discretising the differential equation using the finite element method, which can be viewed as a form of particle method, in a way closing the circle with the particle model as the real model and the continuum model as a fictional model.