SUMMIT

Published: Nov 29, 2024 by Liam Pattinson

SUMMIT is a new software package developed at the University of York which can predict the behaviour of magnetically confined plasmas under the influence of microwave-induced current drive during start-up. A central solenoid is the typical method for driving the initial currents, but these devices take up valuable space in the central column and are vulnerable to neutron irradiation. It has been proposed that future reactors may instead make use of microwave launchers around the outer wall to drive currents, as these are more resistant to radiation and will free up space for the breeding blanket or other components.

SUMMIT was developed as a PhD code, and is comprised of a mix of Fortran and Python. The Fortran code is split into three parts, each compiling to a separate executable which must be run in sequence. Each executable requires its own input file, and although there are many similarities between these input files, they are not mutually compatible. The Python portion of the code coordinates the running of this pipeline and simplifies the user interface.

The majority of our time on SUMMIT was spent upgrading the Fortan portion of the code to make use of modern best practices and to merge the three Fortran sub-programs into a single consistent program. Each sub-program previously shared some key data structures, but code was duplicated between each of them, often with minor discrepancies. The aim of this merge was to increase code reuse, improve maintainability, and to unify the input file format. This proved to be a challenging task, as sometimes small edits would necessitate cascading upstream changes. The widespread use of module-scope variables also made it difficult to track the overall control flow, as many key calculations occured as function/subroutine side-effects. Ultimately, all three sub-programs were unified, but only two of the three sub-programs were fully refactored to an object-oriented design free from module-scope data, while the third was only partially refactored.

The build-system was updated from a collection of Makefiles to instead use CMake, as this is portable over a wider range of platforms and permits easier integration between Python and Fortran via scikit-build-core. The Python portion of the code previously required the user to enter the top-level project directory in order for Python to find all of the required modules and the compiled Fortran executables, so we upgraded this so that the project could instead be installed with pip and run from anywhere on the user’s system. Most of the existing Python workflows have not yet been updated to use the upgraded Fortran, so further work is required here.

Finally, the Fortran testing framework was migrated from naturalFRUIT to pfUnit, as the former appears to be unmaintained while the latter is under active development. pfUnit also offers CMake support and has powerful pre-processing capabilities to reduce the amount of boilerplate needed to set up tests. Additional Python tests were added using pytest, and these primarily compare the results of full simulation runs using the upgraded code to results gathered using the previous code.