[cig-commits] r18854 - seismo/3D/SPECFEM3D_GLOBE/trunk

[dkomati1 at geodynamics.org]

Author: dkomati1
Date: 2011-08-26 09:45:27 -0700 (Fri, 26 Aug 2011)
New Revision: 18854

Modified:
   seismo/3D/SPECFEM3D_GLOBE/trunk/todo_list_please_dont_remove.txt
Log:
added a second comment about fixing the RK4 scheme for attenuation one day.


Modified: seismo/3D/SPECFEM3D_GLOBE/trunk/todo_list_please_dont_remove.txt
===================================================================
--- seismo/3D/SPECFEM3D_GLOBE/trunk/todo_list_please_dont_remove.txt	2011-08-26 16:43:13 UTC (rev 18853)
+++ seismo/3D/SPECFEM3D_GLOBE/trunk/todo_list_please_dont_remove.txt	2011-08-26 16:45:27 UTC (rev 18854)
@@ -88,6 +88,17 @@
 We should fix that one day. The only thing to do would be to design a better time
 integration scheme for the attenuation equation, without that trick; the rest
 (Newmark etc) is fine and does not need to change.
+Daniel Peter did not fix that in 2011; he only fixed the old implementation in SPECFEM3D which used a
+fixed period absorption band to an adaptive one, which uses the resolved
+periods of the mesh (as is done in the global version). The time stepping of the memory variables did not change.
+Thus this IMPORTANT KNOWN BUG remains: in compute_forces_viscoelastic.f90 the calculation of
+attenuation (viscoelasticity) is slightly incorrect because the gradients
+are computed twice but *at the same time step* instead of at different
+(staggered) time steps (t_{n-1} and t_n or something like that).
+That's easy to fix but I have no time for now. Let us fix that in the future.
+It will only make a very small difference in the final seismograms (except for very long runs,
+e.g., for multi-orbit surface waves) therefore the current code with the bug can be used without any major problem.
+
 Reply from Daniel Peter (Princeton University, USA) on February 23, 2010:
 for higher orbit arrivals, we do might want to consider the symplectic time
 schemes as well. Tarje made some tests and showed some "shocking" results