Generalized statistical thermodynamics. Thermodynamics of probability distributions and stochastic processes (Q1626961)

Thermodynamics is typically associated with equilibrium states of matter. In its statistical encoding, thermodynamics is about probability distribution -- the distribution of microstates. The main thesis of the present monograph is that the web of mathematical relationships that is recognised as thermodynamics, can be interpreted as a probabilistic calculus that exists independently of the underlying physics, and even completely independently of the particular problem to which it may be applied. Such model-independent features of thermodynamics were identified by Jaynes in his development of the maximum entropy principle as a tool for statistical inference. Ultimately, the formalism of statistical thermodynamics has been thought to operate in any problem that involves unknown probability distributions. However, when addressing cluster-type systems where the giant component may emerge in finite-size populations (polymer gelation, forest fires, epidemics, network connectivity, percolated states), one encounters hallmarks of the phase transition, which are not literally addressed in the fully-fledged thermodynamic limit. Moreover, the maximum entropy method offers no organised methodology to construct constraints representing our knowledge about the system under study, which are amenable to a consistent variational analysis. The main idea of the author is that the specific choice of the physical (or whatever) model is encoded not in constraints, but in the rule that prescribes the assignment of probabilities. That is accomplished by means of the selection functional (first introduced in Chapter 2, in connection with the mircocanonical cluster ensemble). One ultimately arrives at a variational theory that provides a ``thermodynamical description'' for any probability distribution whose domain is the positive real axis. What is named ``thermodynamics'' is a set of canonical relationships associated with this probability distribution. The major object of studies in the present text is the cluster ensemble where all necessary ingredients of what is named generalised statistical thermodynamics are identified. The most probable distribution gets fixed once a selection functional is specified. The composite effect of entropy and of the selection functional is encoded in the partition function. Its variational properties define the ``equilibrium'' state of the population (of clusters). The partition function and its partial derivatives produce a network of ``thermodynamic relationships'' and as well provides criteria for the coexistence of distinct populations (with a phase transition concept being involved). Standard concepts of entropy and its combinatorial background are reviewed. The core of the theory is formulated in the discrete phase space of the cluster ensemble. The notion of the thermodynamic limit and thermodynamic relationships for the ensemble are derived, with a focus on how much information about the sought for selection functional may be recovered from the knowledge of the distribution itself. Next, stability criteria are discussed and the notion of the giant cluster is introduced. In the theory of bicomponent populations, the selection functional for random mixing is introduced. In the crucial Chapter 7, the whole theory is reformulated as the calculus of variations that makes no reference to the explicit notion of the cluster ensemble. The core of the theory is a probability functional that generates the entire network of thermodynamic relationships for any probability distribution. The remaining three chapters address applications of the theory for stochastic processes, with reference to binary aggregation and fragmentation, both irreversible in real time and both exhibiting a phase transition.