Experiments on theories: The construction of scientific computer simulation.

摘要

Scientific computer simulation involves interacting with a mathematical model, in a way that is analogous to performing a laboratory experiment. Based on interviews with scientists, using a framework of grounded theory and symbolic interactionist sociology, the thesis describes crucial features of this novel mode of scientific work.;A dualistic comparison of simulation with 'theory' and 'experiment' (in the second chapter) gives rise to two apparently independent discussions: of the 'experimental' practices associated with simulation (chapter three), and of the 'theoretical' concerns that shape the technique (chapter four). Those discussions are drawn together in chapter five, which describes the interdependence of those two essential aspects of simulation.;Scientists use comparisons with 'theory' and 'experiment' to emphasise simulation's role as a representation, to differentiate between types of mathematical model, to highlight the agency of the machine in simulation, and to present a computer program as functionally equivalent to a laboratory system.;In order that simulation be "the right tool for the job" in scientific research, a computer program (which encodes a mathematical model) must be conceptualised as an opaque entity, the surface behaviour of which can be tinkered with, observed and explored in an 'experimental' manner. Such semi-structured interaction with the interface of a simulation gives rise to strongly intuitive insight into the mathematical model.;Furthermore, the same program must also be open to analysis, in order that its similarity to whatever system it 'simulates' can be established to the satisfaction of the scientific community. The possibility of informative comparison between the computer and the simulated system is epistemologically essential to simulation-based science. Various 'theoretical' strategies for constructing the analogy between simulation and 'reality' are possible.;Simulation-based research depends upon both of these factors simultaneously. A trade-off between 'experimental' interaction and 'theoretical' legitimacy results from limited computer power, such that there is a delicate balance between the electronic technology, the epistemological requirements and the practice requirements of simulation. These mutually dependent factors are jointly essential to scientists' ability to interact 'experimentally' with mathematical models in a way that provides useful information about the simulated system.