fair-flexi - A trustworthy CFD code for simulation and training

点击翻译按钮获取中文摘要

In this project, we aim at increasing the usability, user-friendliness and impact of the CFD framework FLEXI. FLEXI is a well-established open source ecosystem and is positioned optimally to grow its community and ensure long-lived usefulness. The high quality of FLEXI is witnessed by a number of publications. However, we have identified the need to grow it in terms of the documentation and training aspects and to develop strategies to foster the community. We plan on addressing these aspects through a novel framework which we call fair-flexi. At its core, it is based on the recognition that extending, using and also learning a complex scientific software stack goes far beyond making sure that the simulation results or the source code follow the FAIR principles. While that is of course a first step, it does not solve the following issues: a) Learners are confronted not just with learning how to run the code, but how to operate the full simulation stack, including pre- and postprocessing steps. These tend to include a lot of implicit knowledge that is seldom expressed in an explicit way. b) Experienced users struggle with the same issue at a higher level: The many options, models and parameters of the code and full framework together with often iterative nature of computational research make it very hard to track what worked and what didn’t. Even for very experienced researchers it is surprisingly hard to reproduce their own results. c) Code developers often find themselves in conflict with other code maintainers: Their new feature might slow down the execution, have unwanted interactions with other parts of the code or even break it. Often, a good piece of code might be incompatible with features the developer might not even be aware of. The problems described at each of these levels are an expression of the fact that simulation environments are systems with many parameters and non-linear interactions. This induces a residual ambiguity in the trustworthiness, the usefulness and reproducability of the simulation results. This hampers not just scientific advancement, but also makes learning and using the code painful. Our proposed novel framework fair-flexi is based on the idea of making the full simulation environment, from the first step in a preprocessing tool to a color figure showing the simulation results ’FAIR’. For this, we will generate an automated framework that tracks, provides a DOI, stores and publishes all the steps taken in a specific simulation campaign in a Dataverse. The dataset can augmented by interactive feedback from others and will serve as an invaluable tool to researchers and developers as well as learners. This makes the simulation results truly trustworthy and reproducible down to the most granular level. We plan on field-testing this approach with graduate students and research collaborators. To the best of our knowledge, this is a novel approach and has not been followed in the community of open source CFD codes.