Synopsis of the workshop

A.M. Uhrmacher

The workshop ONE SIMULATION MODEL IS NOT ENOUGH brought together scientists working on modeling and simulation studies in diverse application fields, such as robotics, cell biology, demography, design of electrical and electro-mechanical components, electrical stimulation of cell biological systems. The talks covered different aspects to pursue the question why one simulation model might not be enough from the view point of their application domain. In this context methodological requirements to support developing simulation models in pairs have been identified. Existing methods have been probed what kind of support they provide for developing modeling in pairs, and possible directions for future research discussed. The last part of the workshop was dedicated to a World Cafe Discussion. Four questions were identified to be discussed in small working group. Those were a) what does it mean for a simulation model to be valid?, b) how do methods find users and vice versa?, c) how to synchronize evolving model pairs? and d) what methods and tools are desirable independently of any application domain?

a) What does it mean for a simulation model to be valid?

The quality of any simulation results depends on the validity of the simulation model, and thus this is one of the main questions to answer. Thereby the validity of a simulation model relates simulation model, system of interest, and the research question or objective of a simulation study. Therefore, all three need to be described in an accessible, unambiguous manner. Data are needed for calibration and fitting, as well as for validating the model. Both data sets should not overlap and adhere to some quality criteria and whose provenance (how have they been generated) should be made explicit. Not only the diverse data used in simulation studies, other simulation models the model has been derived from, or the model has been cross-validated with, but also the simulation experiments executed belong to the provenance of a simulation model. The more data are used (from various sources) and the more simulation experiments are done, the more credible a simulation product will be, if this information is made suitably accessible. Constraints, hypotheses and requirements will add to the context of a simulation study and by making the context explicit will facilitate assessing, replicating, and, reusing results and processes. E.g., the later would allow to automatically reuse simulation experiments done with one simulation model with the other, and to put those simulation models, also referring to constraints, hypotheses and requirements into relation.

b) How do methods find users and vice versa?

Sometimes useful methods are out there and users are not aware about them. Platforms such as research gate and the definition of joint projects facilitate the exchange between software developers and domain experts in need of methods and tools. Methodological advances are typically realized in software prototypes, only few research groups have the capability to maintain a software and to support users over a longer period of time. Some of the methods have the potential to be integrated into commercial products of modeling and simulation. For the other tools, being published as open software, well documented, with online tutorials will enlarge the number of potential users.
Also simulation scientists should see public relations as part of their work, rather than as the responsibility of someone else.

Still to embrace new methods and technologies in the everyday work of scientists or within the industry will still require a lot of "leg work" and additional efforts.

c) How to synchronize evolving model pairs?

To evolve models in pairs the explicit link (or morphism between both models) needs to evolve automatically as well, and thus needs to be adapted automatically. Pre-requisites for this are to distinguish between relevant and not relevant changes and to make the mappings explicit. This would also form a first step to automatically adapt parameter values from one model to the other model.  Also, automatic transformations between the different modeling approaches and simulation models would help to reveal different perspectives on the system of interest, and at the same time to keep the relationships between two simulation models consistent.

d) What methods and tools are desirable independently of any application domain?

Independently of any application domain, not only for simulation models we need implementation-independent specifications but also for simulation experiments. Some areas such as cell biology have made more advances towards the standardization of simulation models and simulation experiments. Repositories where all information referring to the main products of a simulation study can be accessed is highly desirable. Open questions remain e.g. how much information at which level of detail will allow to instill trust in a particular simulation product without overpowering user and developer likewise. Here user-specific solutions are required.

In all discussions, one re-occuring theme was to make relation-ships between data and simulation model, between simulation models, and between studies explicit to allow an automatic notification of users and adaptation and thus, to contribute to the consistency and quality of industrial and scientific modeling and simulation results.