When can a Computer Simulation act as Substitute for an Experiment? A Case-Study from Chemisty
|Table of Contents|
|2 Similarities and Differences between Simulations and Experiments|
|3 Case Study: Simulation of H-2-Formation in Outer Space|
|4 Summary and Conclusions|
As our case-study shows, computer simulations can in many respects be compared to experiments. Yet, as we argued earlier, being comparable to experiments does not mean that computer simulations are experiments. But it seems promising to pursue the question under what conditions a computer simulation can act as substitute for an experiment and what kind of research design a simulation of phenomena that are not directly accessible by experiment must follow.
From our case study we can learn that there is a good chance for employing simulations as an experiment-surrogate if the investigated phenomena fall in the realm of a comprehensive theory, i.e. a tried and tested theory that fully covers the phenomena that are simulated. If this is the case then the question of validating the simulation is reduced to justifying the employed modeling techniques and approximations which does not necessarily require empirical validation. The underlying research design of such simulations could roughly be described as “comprehensive theory plus well-approved modeling techniques”. This research design appears to be a valid research design for experiment-surrogate simulations.
Here, we consider a research design as valid when, if executed properly, it has the chance of generating reliable scientific knowledge. In contrast, a research design is invalid or “broken” or “unsound” if, even when properly followed, it cannot generate reliable scientific knowledge about the investigated subject matter. It is important to understand that even if the research design is valid a particular research project can still fail, namely, if one or more of the required steps have not been executed properly. On the other hand, if the research design itself is unsound, any research project following it will inevitably fail.
We conjecture that the same research design and therefore the same epistemic justification also underlies many other examples of simulation research in the natural sciences. A salient candidate for future investigation would be QM/MM simulations in chemistry. The situation in the case of QM/MM simulations is slightly more complicated than in our case, because QM/MM simulations make use of two background theories, molecular mechanics and quantum mechanics and introduce coupling terms to bridge the molecular mechanics and quantum mechanics part (Senn/Thiel 2009). This raises the question if our idea of a “comprehensive theory” can still be applied to describe the research logic of such simulations or if it needs to be adjusted.
Simulations in astronomy that, like simulations of the collision of galaxies, cannot directly be validated by observation might also be an example for a similar kind of research logic and epistemic justification. But they probably introduce a further problem that did not play a prominent role in our example. Our example was a simulation of a reaction of benzene and hydrogen. The structure and the properties of the involved molecules and chemical elements are very well known. But can the same be said of the initial data from which simulations in astronomy start? And, if not, what would be the consequences for their epistemological assessment?
A question that has not been touched in this paper is how simulations that do not or cannot rely on comprehensive theories at all get their credentials. One could say that a comprehensive theory and well-approved modeling techniques jointly form a sufficient condition for a proper experiment-surrogate. But does that also mean that being able to rely on a “comprehensive theory” is a necessary condition? While we have not investigated and therefore cannot exclude the possibility that there are other valid research designs for experiment-surrogate simulations, it appears to us that it would probably be much harder to establish the credibility of an experiment-surrogate simulation of phenomena that are not covered by a comprehensive theory. At least it is difficult to imagine how in this situation an experiment-surrogate simulation could claim credibility without direct empirical validation. For, how were we to know without direct empirical validation that our simulation did not omit one or more relevant causal factors?
However, this is just a conjecture and we do not want to dogmatically exclude the possibility of valid research designs for experiment-surrogate simulations that do not rely on a comprehensive theory. It would indeed be unfortunate, if there weren't any, because, except for the exact natural sciences and their technological application fields, there are only few areas of science where we can rely on comprehensive theories.
But for the same reason a healthy amount of skepticism is also advisable when the use of numerical methods in the humanities is justified by reference to their success in the exact natural sciences - as it is sometimes done by those schools in the social sciences that try to repeat the success of the natural sciences by imitating their methods (Shapiro 2005). Successful research designs can only be transferred from one science to another if the conditions for their applicability have been properly understood. By presenting a case-study from theoretical chemistry we hope to have made a contribution to the better understanding of a particular kind of simulation research design for simulations that act as substitute for experiments.
Acknowledgement The authors thank the German Research Foundation (DFG) for financial support of the project within the Cluster of Excellence in Simulation Technology (EXC 310/1) at the University of Stuttgart .