Closed tamiko closed 4 months ago
@nachosaurus Current convergence tables for this pull request:
dG Q0:
#dofs L1 error L2 error Linf error
------- ---------- ---- ---------- ---- ------------ ----
64 0.239303 0.445754 1.37159
256 0.0934643 1.36 0.189046 1.24 0.633356 1.11
1024 0.0235038 1.99 0.0494011 1.94 0.203721 1.64
4096 0.00579149 2.02 0.0124722 1.99 0.056288 1.86
16384 0.00146566 1.98 0.00326836 1.93 0.0293925 0.94
dG Q1:
#dofs L1 error L2 error Linf error
------- ----------- ---- ---------- ---- ------------ ----
256 0.0896792 0.173201 0.974556
1024 0.0305853 1.55 0.0691853 1.32 0.431899 1.17
4096 0.00886494 1.79 0.0213493 1.7 0.128639 1.75
16384 0.00232961 1.93 0.00575481 1.89 0.0321078 2.0
65536 0.000590042 1.98 0.00146795 1.97 0.00794425 2.01
dG Q2:
#dofs L1 error L2 error Linf error
------- ----------- ---- ----------- ---- ------------ ----
576 0.0412296 0.0670614 0.460117
2304 0.00579616 2.83 0.0138716 2.27 0.170333 1.43
9216 0.000593558 3.29 0.00193619 2.84 0.0342838 2.31
36864 7.25492e-05 3.03 0.000269699 2.84 0.00674325 2.35
147456 9.1628e-06 2.99 3.8928e-05 2.79 0.00137409 2.29
dG Q3:
#dofs L1 error L2 error Linf error
------- ----------- ---- ----------- ---- ------------ ----
1024 0.00735236 0.0147951 0.115376
4096 0.000625408 3.56 0.00194717 2.93 0.0303727 1.93
16384 4.28851e-05 3.87 0.000153276 3.67 0.00216174 3.81
65536 3.03506e-06 3.82 1.18803e-05 3.69 0.00019853 3.44
The slight reduction for dG Q0
comes from the fact that we use a rather strict relaxation window [1] and the relaxation is not expanded over the stencil. This works reasonably well for Q1
- Q3
, but not for dG Q0
. Hence the reduction in the L^\infty norm.
[1] https://github.com/conservation-laws/ryujin/blob/development/source/euler/limiter.h#L347-L351
This pull request adds to more missing pieces:
HyperbolicModule
that does the bounds extension. Performance impact on the standard method should be minimalWith these two changes I now see proper convergence rates... I will submit test vectors in a separate pull request.