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

Johannes Kästner and Eckhart Arnold

3.1 Introductory Remarks on Simulations in Chemistry

One can safely say that by today computer simulations have a long standing tradition in chemistry. The interest in chemical simulations is, among other things like environmental considerations, motivated by the fact that simulations allow to study details of chemical reactions that cannot be obtained from experimental data or that are practically inaccessible by experiment at all. But also when experiments are possible, simulations can be used to double check the experimental results for their plausibility (Alexander/Etal 2002, Wang/Etal 2008).

In our example of the simulation of $H_2$-formation in outer space, the very slow reaction rate renders the details of the reaction mechanism practically inaccessible to experimental techniques (Goumans/Kaestner 2010, p. 7351). This simulation can thus be considered as a stand in for an otherwise impossible experiment or as a surrogate simulation. A direct validation of the simulation results is not possible in this case. Here, we speak of “direct validation” of a simulation when the same or almost the same process that has been simulated has also been tracked empirically either by a) an experiment of the same process under the same conditions or by b) observations in case the same process occurs under the same conditions in nature and is directly observable in nature.

However, consequences of the results, e.g. $H_2$-enrichment in outer space, can be compared to observational data (Goumans/Kaestner 2010, p. 7351f.). Also, indirect forms of experimental validation are possible for some aspects of the simulation and have indeed been applied (see section 3.4.3 on validation below). The simulations cannot rule out, though, that other mechanisms than those predicted by the simulation may explain the observations. The guiding question of our case study will be on what grounds we can consider the simulation a proper experiment-surrogate if no direct validation is possible.[5] In other words, where does this simulation get its credentials from or why is it trustworthy?

Simulations in chemistry are based on physical theories. Different approximations to those theories are used. The choice of the approximation depends on the particular reaction that is simulated and on the level of detail and accuracy that is desired as well as on the inevitable constraints in computing power. Two popular types of approximations can be distinguished. 1) Molecular dynamics simulations are cheap in terms of computing power but cannot describe changes in electronic structure, e.g. bond breaking and bond formation. 2) Quantum mechanical simulations treat electrons in much detail and, thus, allow to simulate breaking and formation of chemical bonds, charge transfer, and electronic excitation. They require significantly higher computational cost.

Our example belongs to the second category. Because the number of atoms involved in the simulated reaction ($H_2$-formation catalyzed by chemisorption of H on benzene) is small enough, a quantum mechanical simulation of the reaction is feasible.

[5] It should be noted that this question is different from that which is pursued by Barberousse et al. (2009) whose criticism of the the physicality-argument to bracket simulations and experiments is otherwise likeminded to our tenets. (Barberousse et al. 2009) examine the semantic relation between simulations and their target system, which we take for granted here, but do not ask the question of epistemic justification, which is our main concern.