When can a Computer Simulation act as Substitute for an Experiment? A Case-Study from Chemisty

Johannes Kästner and Eckhart Arnold

1 Introduction
2 Similarities and Differences between Simulations and Experiments
    2.1 Similarities of Simulations and Experiments
    2.2 Differences between Simulations and Experiments
3 Case Study: Simulation of H-2-Formation in Outer Space
4 Summary and Conclusions
Bibliography

2.1 Similarities of Simulations and Experiments

One obvious reason why computer simulations are often labeled as “computer experiments” is that the process of designing, setting up, running and evaluating a simulation is by its very appearance quite similar to that of designing, setting up, running and evaluating an experiment. Both simulations and experiment share the same structure: They operate on an object to learn something about a target system. With object we mean the entity on which a computer simulation or an experiment operates. With target system we mean that entity in nature that we want to learn something about with a simulation or an experiment. The object must in some way or the other be representative of the target system. In the case of an experiment, however, the object can also be an instance of the target system itself, while in a simulation the object is always a representation of the target system.[2]

Both simulations and experiments run in a controlled environment and both allow interventions on the object (Parker 2009, p. 487). In simulations the same tools are applied that were formerly thought of as typical for experimental data analysis like visualisation, statistics, data mining (Winsberg 2010, p. 33). This includes similar techniques of error management for simulations and experiments. Among these are: Validation of the set-up (or the apparatus) against cases with known results, testing for the responsiveness on interventions, replicating the results under slightly different conditions, testing for the conformance of the results with undisputed theoretical and phenomenological background knowledge. Also, simulations just as experiments allow us to learn something new and potentially surprising about their object and, if the object is truely representative of the target system, also about the target system itself.

Further shared characteristics between simulation and experimental practice are mentioned in the literature. It suffices to summarize them here: 1) the “constant concern for uncertainty and error” (Winsberg 2010, p. 34), 2) that simulations -- just like experiments according to Hacking (Hacking 1983) - “have a life of their own” and are in part “self-vindicating” (Winsberg 2010, p. 45), 3) that simulations and experiments share the same challenge of bridging the gap between their object and the target system, or, as one might also say, between the experimental set-up and the real world outside the experiment (Arnold 2008, p. 174/175).

[2] Mary S. (Morgan 2003) introduces a distinction between being a representative of and being a representation of to describe this difference. The term representative in contrast to representation is reserved by her for those cases where the object under study is either identical or an instance of the target system. A very similar distinction is suggested by (Guala 2002) and picked up, though not endorsed by Winsberg (2009). Emphasizing the relation to the target rather than the role of the object, Guala speaks of material similarity and formal similarity.

t g+ f @