Chris
What do you mean by "Boundary conditions are not applied"? What do you do at the boundaries? Are you using one-sided differencing there?
Given your description, I assume that all derivatives would be identically zero, that there would not even be a floating-point round-off error. Is that so? If not, why not?
Are you really setting the lapse to zero initially? It should be one for Minkowski, same as e.g. g_xx.
What happens if you choose a very small grid size (e.g. just 10 points) and compare the initial conditions between the Einstein Toolkit and the Python code? Are they identical? You would need to compare ASCII output and compare all the digits.
What happens if you take a single Euler time step? Are the values still all identical? Try to find out and which iteration the first grid point differs between the two codes.
-erik
On Fri, May 13, 2022 at 9:59 PM Chris Stevens < chris.stevens@canterbury.ac.nz> wrote:
Hi all,
I've been having trouble implementing the BSSN equations, as given by CTGamma, in Python using the package COFFEE, which I am very familiar with:
https://gitlab.com/thebarista/coffee https://doi.org/10.1016/j.softx.2019.100283
The main comments are:
-- For simplicity, I am starting with a 1+1 code (set 2/3 spatial derivatives to be exactly zero). -- The 1+log slicing for the lapse and a zero shift. -- The "phi" conformal factor type -- The evolution equations are written to Python by a very slight modification of the .m file in CTGEvolution. I've confirmed this .m file outputs the same file that is in the src directory. -- The vanishing of the trace of A is enforced after each full timestep in the same way as CTGamma. -- Boundary conditions are not applied. -- Fourth order SBP operators that lean over toward the boundaries are used. -- Minkowski initial data with all fields zero except \gamma11/22/33. -- Euler step is used for time integration to easier compare between Python and the Einstein toolkit.
This leads to a stable evolution in the Einstein toolkit with the attached .par file using a simple one patch Cartesian grid. However, in Python, I get that fields diverge, starting at the boundaries and then propagating inward (including with RK4). This behaviour does not go away with higher resolution or reduced CFL. It is however, fixed completely by setting the second spatial derivative of \alpha to be exactly zero. Trying many different SBP FD operators that have proved successful for other projects has not changed the behaviour.
The COFFEE code has been used and tested in a variety of projects and so there should not be a problem here. Thus, the only thing I can think of, is that the Einstein toolkit is doing something that I am not realizing, that is helping it remain stable. Hopefully, somebody on this mailing list might have an idea?
Thanks in advance,
Chris
*Dr Chris Stevens*
*Lecturer in Applied Mathematics*
Rm 602, Jack Erskine building
School of Mathematics and Statistics
T: +64 3 369 0396 (Internal 90396)
University of Canterbury | Te Whare Wānanga o Waitaha
Private Bag 4800, Christchurch 8140, New Zealand
*Director* SCRI Ltd http://www.scri.co.nz
Users mailing list Users@einsteintoolkit.org http://lists.einsteintoolkit.org/mailman/listinfo/users
