Please translate to English, so that it can be discussed here! I am primarily interested in how the virtual ensemble differs from an ordinary statistical ensemble, i.e., a large collection of actually identically prepared systems. The latter is the usual ensemble on which one can make statistics by counting, averaging, etc.. The former (the virtual ensemble) is (at least according to Gibbs) one of which you have only a single copy (a single metal bar, say) on which you make the predictions and verify them, while the others copies are just imagined to exist, but never realized. It was Gibbs' trick that enabled him to use (or misuse) the statistical machinery to get results that apply to single macroscopic systems.