#2920: Loss of convergence with subcycling in time
Reporter: Miguel Zilhão Status: new Milestone: Version: Type: bug Priority: major Component: Carpet
in evolutions with electromagnetic fields (with the ProcaEvolve thorn) the violation of the Gauss constraint is not converging (to any order).
this can be observed in a very simple setting of a single, unperturbed, charged black hole and a Carpet grid without any moving components. everything is converging just fine at t=0, but once the evolution starts, convergence is lost very quickly. without subcycling in time, everything converges as expected to 4th order accuracy.
i've attached a thornlist and some parameter files that reproduce the issue. i'm also attaching plots (explained below) and the corresponding gnuplot scripts to reproduce them. to obtain the files in the "ygraph" format, use the usual Carpet2ygraphCat.pl script: Carpet2ygraphCat.pl procaevolve-gauss.x.asc Carpet2ygraphCat.pl leanbssnmol-ham.x.asc
these are evolutions for a single charged BH, with Q=0.2. for the parfiles 202602_RN01_nosubcycl_hf20.par and 202602_RN01_nosubcycl_hf40.par, all timelevels update at the same rate with
Carpet::time_refinement_factors = "[1, 1, 1, 1, 1, 1, 1]"
for the other parfiles (202602_RN01_hf20.par and 202602_RN01_hf40.par) all timelevels update at the same rate *except* the two inner ones.
the plots gc_x_RN_nosubcycl.pdf and gc_x_RN_subcycl.pdf show the violation of the Gauss constraint along the x-axis for different time steps:
- in the gc_x_RN_nosubcycl.pdf plot, everything looks fine; - in the gc_x_RN_subcycl.pdf case, notice how a lot of noise propagates from the two inner refinements levels -- the ones that are updated more frequently than the rest.
this noise is *not* contained in the buffer region (if it was, we'd be fine), but propagates out, contaminates the rest of the grid, and totally ruins any convergence study. this can be seen in the convergence plots gc_x_RN_nosubcycl_conv.pdf and gc_x_RN_subcycl_conv.pdf, which are convergence plots of the violation of the Gauss constraint along the x-axis at t=10:
- in the gc_x_RN_nosubcycl_conv.pdf plot we can see very nice 4th order convergence; - in the gc_x_RN_subcycl_conv.pdf case, we can only see convergence (to 4th order) for large x-values (which have not yet been contaminated with the propagating noise). until x~10, there is no convergence to any order.
i'm also sending the equivalent plots for the Hamiltonian constraint violation, where similar features are seen, but the overall effect seems smaller and does not completely spoil convergence studies. attachment: 2026-02_bug_report.zip (https://api.bitbucket.org/2.0/repositories/einsteintoolkit/tickets/issues/29...)
-- Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2920/loss-of-convergenc...