Tools for Evaluating the Consequences of Prior Knowledge, but no Experiments. On the Role of Computer Simulations in Science

Eckhart Arnold

1 Introduction
2 Common features of simulations and experiments
3 Distinguishing features of experiments and simulations
4 Borderline cases
5 Summary and conclusion: Computer simulations as a tools for drawing conclusions from prior knowledge

1 Introduction

In the literature on computer simulations there is an ongoing debate about whether simulations are experiments (Winsberg 2009, Morrison 2009). In this article I defend the view that computer simulations and experiments are two separate and clearly distinguishable categories and that notwithstanding the fact that the usage of the words “simulation” and “experiment” may be somewhat vague, a sharp line can be drawn between simulations and experiments from an epistemological point of view. In order to do so, I discuss the various similarities and differences between simulations and experiments that have been pointed out in the literature. Only those differences are considered as crucial that are epistemically relevant in the sense that they make a difference regarding the justification of the simulation or experiment in question.[1] The discussion of similarities and differences between simulations and experiments shows that there are at least three fundamental and epistemically relevant differences between simulations and experiments which justify a sharp distinction between the category of simulations and the category of experiments: 1) New empirical data can only be gathered by doing experiments or real measurements, but not by doing simulations. 2) Only experiments can operate directly on the target system, simulations never do. 3) Experiments, empirical measurements and observations alone can be used for testing fundamental hypotheses. Simulations can only be used for testing the consistency of a hypothesis with a background theory, but not for testing the fundamental theories themselves.

As the formulation of these differences suggests, they do not become acute in every single instance of a simulation or experiment. Rather, the categories of simulations and experiments are both large and inhomogeneous and the concepts of simulations and experiments must therefore remain somewhat vague. Also, there is an overlap region between simulations and experiments, because some experiments essentially fulfil the function of analog simulations. In order to clarify the situation the border cases of the experimentum crucis that can for principle reasons never be substituted by a simulation as well as the cases of simulation-like experiments and experiment-like simulations will be discussed.

The most critical case, however, is that of hybrid methods which combine empirical measurements with sophisticated simulation-like computations. I argue that a hybrid method remains essentially an empirical method (i.e. experiment or measurement) - in contradistinction to a computer simulation based on empirical input data - as long as the result the hybrid method yields is a result about exactly the same system from which the input data was taken.

In the conclusion I argue that the differences between simulations and experiments follow from the fact that a computer simulation cannot yield any other results than those that are logically implied by its premises. This fact places computer simulations firmly on the theoretical side of science in contradistinction to its empirical side (experiments, observations, experiences), which contradicts the view that computer simulations are a “third way of doing science” (Axelrod 2003, p. 90) between induction (empirical research) and deduction (theory) and relativises the view that “materiality” is not a proper distinguishing criteria (Parker 2009).

[1] By justification I mean answers to the questions if and what they allow us to learn about nature and why we should believe it to be true.

t g+ f @