Synthese (2009) 169:557–574 DOI 10.1007/s11229-008-9430-7 Computer simulations as experiments Anouk Barberousse · Sara Franceschelli · Cyrille Imbert Received: 1 November 2006 / Accepted: 4 August 2007 / Published online: 18 November 2008 © Springer Science+Business Media B.V. 2008 Abstract Whereas computer simulations involve no direct physical interaction between the machine they are run on and the physical systems they are used to investigate, they are often used as experiments and yield data about these systems. It is commonly argued that they do so because they are implemented on physical machines. We claim that physicality is not necessary for their representational and predictive capacities and that the explanation of why computer simulations generate desired information about their target system is only to be found in the detailed analysis of their semantic levels. We provide such an analysis and we determine the actual consequences of physical implementation for simulations. Keywords Experiments · Data · Machine implementation · Programs 1 Introduction: the practice of computer simulation Some computer simulations undoubtedly share some epistemic functions with field experiments: they are run to provide new data about systems that are difficult or impossible to investigate with ordinary instruments. For instance, neutron–matter interaction A. Barberousse, S. Franceschelli and C. Imbert have contributed equally to this work. A. Barberousse (B) · C. Imbert IHPST, CNRS, Université Paris 1, ENS, Paris, France e-mail: Anouk.Barberousse@ens.fr S. Franceschelli ENS-LSH, Lyon, France S. Franceschelli REHSEIS, Paris, France 123 558 Synthese (2009) 169:557–574 has been studied intensively by means of molecular dynamics simulations. Owing to these simulations, it is possible to investigate, among other phenomena, the diffusion of neutrons by a phonon, diffraction, anomalous dispersion. Comparing computer simulations with field experiments is a good way to shed light on many distinctive features of simulations. In both cases, vast amounts of data are first produced and then analyzed with the help of similar techniques. How are these data produced and validated? This question is even more difficult to answer in the case of simulations than in the case of field experiments, for computer simulations do not involve any direct physical interactions with the systems they are used to investigate. We use “direct physical interaction” in the sense of measurement interaction between us and the target system. Even if sometimes simulations are initialized with data obtained by measuring the target system, they never involve any direct physical interaction with it. This paper aims to give a qualified answer to the above question for the case of computer simulations. We argue that, after its earlier, mostly descriptive phase, the epistemology of computer simulation can now take an explanatory turn (cf. Winsberg 1999, 2001, 2003). Putting simulations and models on the same philosophical footing, we claim in this paper that the best candidates for such an explanation is to be found in a detailed semantic analysis of computer simulations, showing how the physical states of the computer can step after step be interpreted as computational states, as values of variables and finally as representations of the target system’s successive states. When performing this semantic analysis, we do not ignore the fact that a decisive component for a good simulation for a simulation to generate desired information about its target system is that it relies on a good model of the target system. This model has to be transformed (sometimes drastically) into algorithms. We carefully analyze some of these transformations, which lead to well-calibrated simulations, obtained by trial-and-error procedures and comparisons with already available results. A frequently given explanation as to why (some) computer simulations are used as experiments appeals to their being run on physical machines. We call this claim the “physicality claim”. In Sect. 2, we present different arguments in the literature that may be used to support the physicality claim. In Sect. 3, we try to answer what we believe to be the best argument in favor of a strong version of the physicality claim. In Sect. 4, we present our own semantic analysis of computer simulations as well as its implications. In Sect. 5, we ground our analysis in a renewed analysis of the actual meaning of machine implementation. 2 Simulations can be used as experiments because they are run on machines Setting up an epistemology for computer simulations begins with considering the following explanandum E: E: Computer simulations, although they are basically computations and do not involve any measurement interactions, nevertheless do generate new data about empirical systems, just as field experiments do. 123