The dissertation examines two putative explanations from statistical mechanics with the aim of understanding the nature and role of idealizations in those accounts, namely, the Yang-Lee account of phase transitions and the Boltzmannian account of irreversible behavior. Like most explanations in physics, these accounts involve idealizations to some extent. Many idealized explanations hold out the hope that the idealizations can be removed or eliminated with further work. However, the idealizations that occur in the accounts of phase transitions and irreversibility are ineliminable. The only way (in principle) to obtain a description – let alone an explanation – of these phenomena is to invoke various idealizing assumptions.
Ineliminably idealized explanations are not well-understood from a philosophical point of view. Indeed, most philosophers of science would probably hold that no idealizations are ineliminable. The dissertation argues that this view is mistaken, showing where and why extant accounts of idealization miss this fact by distinguishing the widely-accepted understanding of idealizations as falsehoods from a novel understanding of idealizations as abstractions. As abstractions, idealizations are devices for ignoring certain details about the real world. The dissertation argues that ineliminable idealizations cannot be falsehoods, and that they should be understood as abstractions.
The dissertation also examines the confirmation of idealized hypotheses and their role as guides to what the world is like. At least some idealized hypotheses have some degree of confirmation; and less idealized hypotheses tend to be better confirmed than their more idealized counterparts. If idealizations are falsehoods, Bayesian confirmation theory seems unable to obtain these results, because it lacks a way of defining the prior probabilities of idealized hypotheses. If idealizations are abstractions, however, idealized hypotheses about a system are incomplete claims that omit certain details about the system. Since prior probabilities are assigned to such hypotheses in the same way they are assigned to incomplete descriptions, understanding idealizations as abstractions allows Bayesianism to secure the above-mentioned results. This understanding of idealizations also allows idealized hypotheses to be guides to what the world is like, because the incompleteness of such hypotheses is compatible with the cogency of inference to the best explanation.