Cian Wilson, Carnegie Science, DTM
Modern, advanced, open-source computational libraries are giving an increasing amount of power to researchers to develop customized numerical models, tailor made for individual problems. While incredibly powerful and flexible, this approach raises questions about the reliability and maintainability of scientific models when each new problem is solved using a new script or piece of code. During this seminar we will discuss TerraFERMA, a model building framework built on the FEniCS, PETSc and SPuD libraries, that aims to increase the maintainability of models by providing a common core to all models designed using TerraFERMA's interface. This core library allows updates to the underlying libraries to be implemented largely unseen by the user and for new functionality to be made available to all existing models without having to edit each one individually. Along with an extensive test suite, it also ensures that verifying one model provides some trust in another. We will discuss our goals in designing TerraFERMA, give a brief overview of its structure, demonstrate some examples of its use and discuss our future development goals.