Present: Beyhan, Samuel Cupp, Helvi Witek, Peter Diener, Kenneth Sible, Roland Haas, Steven Brandt, Alois Peter Schaffarczyk, Miguel Gracia, Zachariah Etienne, Maria Hamilton, Yosef E Zlochower
Chair: Helvi Minutes: Samuel
Chairs and minute takers for next two weeks were decided. 10/8 Meeting Chair: Steve Minute taker: Miguel
10/15 Meeting Chair: Peter Minute taker: Roland
SPEC Benchmarks: The Einstein Toolkit submission to the SPEC benchmark was discussed again. The deadline for submission is February 21st, so the code should be submitted before this date. Since this submission date is soon after winter break, the code should be prepared for submission prior to the break if possible. Submission requires a standalone version of the code, which will likely consist of an ML_BSSN McLachlan code.
Roland offered to create an Overleaf document with information on SPEC benchmark submission process, and he will send the link to the document to those on the call, so others can assist with preparation for submission. Erik Schnetter was previously submitted to SPEC, so he will be contacted to provide input on the Toolkit's previous submissions to this program.
Moving off the Jenkins server: An intern at NCSA is working to port the testing features on Jenkins over to Github Actions. One goal is to move the continuous integration tests to an external system that does not require as much maintenance by Toolkit maintainers. Also, Github should not have the security issues that were present with the Jenkins server. The solutions to these issues prevented Jenkins from being easily accessible to anyone outside of NCSA, so moving to Github should allow the testing and testing results to be more publicly visible.
Currently, the Github port can compile and run the tests. It also caches compiled code so compilation time is minimized. The demo page is available at https://mojamil.github.io/einsteintoolkit/. It currently has limited UI and presentability, as the first goal is ensuring functionality. Roland requested input for suggestions and feedback, especially about any features that are missing from the prior testing system or improvements that could be made.
Helvi requested listing the unrunnable tests explicitly on the webpage. Roland explained the cause of most unrunnable tests and suggested adding a clickable UI to show details for those tests.
Another future goal is to provide data from previous tests. The information about old test results is kept; it simply needs to be accessible via this interface.
Upcoming release: The release still needs a release coordinator (7 weeks till release!). The individual release tasks have been itemized by Roland and Steve, but they haven't yet given estimates for the time commitment for each task. They aim to have this soon so that people will volunteer for these roles. Helvi offered some time and/or students for testing the new release.
NRPyPN ticket 2417: This needs to be reviewed for the next release. Peter has begun the process of review and doesn't expect any major issues.
Postdoc position announcement: Helvi announced a postdoc position with herself and Roland at University of Illinois at Urbana-Champaign. An official solicitation email is pending, and the solicitation is linked below. https://hyperspace.uni-frankfurt.de/2020/09/30/postdoctoral-postion-with-foc...
User questions: Beyhan asked about an error occurring specifically with non-equal mass simulations. The error does not occur with equal mass. For these simulations, he is using McLachlan + GRHydro. At some point, the lapse goes negative, and this ends the simulation. Zach suggested setting the minimum lapse to a small positive value, though the default setting for this parameter is 0. He also suggested being more conservative with the trigger for adding refinement levels, as they may need to be introduced slightly sooner for accurate data.
Another suggestion was to check if the horizon finder is the source of the error. If so, having the finder not search until further into the simulation may help. Interpolation early in the simulation may result in values which aren't a problem for the simulation but could cause errors in the horizon finder.