Add NRPyEllipticET to the Einstein Toolkit

Create issue
Issue #2616 new
Former user created an issue

NRPyElliptic is a new, hyperbolic relaxation elliptic solver (Assumpcao et al, https://doi.org/10.1103/PhysRevD.105.104037). NRPyEllipticET is the associated Einstein Toolkit thorn, which we propose for inclusion.

In its first application, NRPyElliptic sets up conformally flat, binary black hole (BBH) puncture initial data (ID) on a single numerical domain, similar to TwoPunctures. While it is about 6x slower than TwoPunctures, NRPyElliptic requires only about 0.3% of the runtime for a full BBH simulation in the Einstein Toolkit.

As consumers of numerical relativity ID generally already possess expertise in solving hyperbolic PDEs, they will generally find NRPyElliptic easier to tweak and extend than other elliptic solvers. We are actively working to extend NRPyElliptic to solve other types of ID, and these new ID types will be added to the ET thorn.

Comments (10)

  1. Leonardo Werneck

    Hi @Roland Haas ! Yes, I’ve already completed most of the major changes we were planning for the thorn. I’m working with Thiago Assumpção (lead developer of NRPyElliptic) to add an extra functionality to the thorn where the damping parameter of the hyperbolic system is computed automatically, yielding optimized relaxations for all grid resolutions (unless the user chooses to override it). This should be complete soon, at which point we will be in touch with the reviewers.

  2. Roland Haas

    “Soon” is a vague statement. Note that the initial review of harmlessness must finish before 2022-09-01 for NrPyElliptic to be included. The reviewers will also require some time to review which the proposers should provide them with.

    Also note that major changes to the code may pose an issue wrt to stating that the code that is being reviewed is actually the one used in a prior publication (ie code quality may actually go down due to newly introduced bugs and loss of correctness compared to the code used in publication).

    Usually the ET does not include “bleeding edge” components but components that have proven to themselves already in real world applications (ie used in a scientirfic publication) and it may be desirable to postpone inclusion if that is still to be seen (without passing judgement on the component of course).

  3. Zach Etienne

    I don’t think this “new” functionality qualifies as “bleeding edge”, as it was hashed out in the publication. It’s basically an empirically derived means to set a more optimal value for the damping parameter.

    Also I believe “soon” means “within a few days”.

  4. Roland Haas

    I would be quite worried about having any sort extra functionality (in particular automatisms) that has not yet been used in publication in the new contribution. There is the worry that one may end up with a new contribution to the ET that claims to have been used in publication X only for a ET users trying to reproduce X and finding out that this is not possible due to the change. Obviously this would be quite embarrassing.

    While I trust the NRPyElliptic group to carefully run tests, I wonder why take the risk? Submit the version that was actually used for publication, then add the new functionality in the next release cycle.

    Also note that potentially giving the reviewers only a single week for the “does no harm” review is really pushing it. They may be out on a conference, vacation or unavailable for other reasons.

    Note that we had called for new contributions in June already (https://docs.einsteintoolkit.org/et-docs/Meeting_agenda#2022-06-02) so starting to adapt the contribution now is somewhat late in my opinion. Before the June call should have been the time for adaptation.

  5. Log in to comment