Instability with AK135

[QuLogic]

Running with the AK135 model and the latest code causes an instability and fails to run. The previously used v5.1.5 tag appeared to work correctly. A bisect shows that the commit 7b76e2e58b880cf6707510e8b62c01d6b69d23c1 ("fixed model ak135f_no_mud, which was not correct" by @komatits) is the cause. Results are as follows:

• Last good commit fe2157eed7863236b988d099f8dac594824c87ad - Works. (Note: there are a few bugs, so you need to patch it to get it to run; I can provide these changes if necessary.)
• First bad commit 7b76e2e58b880cf6707510e8b62c01d6b69d23c1 - Fails with message "forward simulation became unstable and blew up in the fluid"
• master - Fails with "forrtl: error (72): floating overflow"
• devel - Fails with "forrtl: error (72): floating overflow"

The relevant settings are (probably) NCHUNKS=6, NPROC_XI=NPROC_ETA=6, NEX_XI=NEX_ETA=240, but I can provide the full Par_file if needed.

[komatits]

It is probably a minor issue (time step slightly too high). Until that change the implementation of AK135 was not right, someone had implemented a slightly different version compared to the original AK135 paper (a few values in the inner core and outer core had been changed if I remember correctly). I fixed that, and now AK135 is really AK135. However, if some Vp values were slightly increased we may have to slightly decrease the time step accordingly.

In the routine in which the time step is computed could you put:

DT = 0.95 * DT

and try again and let us know?

On 02/06/2014 00:36, Elliott Sales de Andrade wrote:

Running with the AK135 model and the latest code causes an instability and fails to run. A bisect shows that the commit 7b76e2e https://github.com/geodynamics/specfem3d_globe/commit/7b76e2e58b880cf6707510e8b62c01d6b69d23c1 ("fixed model ak135f_no_mud, which was not correct" by @komatits https://github.com/komatits) is the cause. Results are as follows:

• Last good commit fe2157e https://github.com/geodynamics/specfem3d_globe/commit/fe2157eed7863236b988d099f8dac594824c87ad
  • Works. (Note: there are a few bugs, so you need to patch it to get it to run; I can provide these changes if necessary.)
• First bad commit 7b76e2e https://github.com/geodynamics/specfem3d_globe/commit/7b76e2e58b880cf6707510e8b62c01d6b69d23c1
  • Fails with message "forward simulation became unstable and blew up in the fluid"
• master - Fails with "forrtl: error (72): floating overflow"
• devel - Fails with "forrtl: error (72): floating overflow"

The relevant settings are (probably) |NCHUNKS=6|, |NPROC_XI=NPROC_ETA=6|, |NEX_XI=NEX_ETA=240|, but I can provide the full Par_file if needed.

— Reply to this email directly or view it on GitHub https://github.com/geodynamics/specfem3d_globe/issues/134.

Dimitri Komatitsch CNRS Research Director (DR CNRS), Laboratory of Mechanics and Acoustics, UPR 7051, Marseille, France http://komatitsch.free.fr

[QuLogic]

This appears to be working okay with ifort. I'll need to test with another compiler as well, and if it works, I'll open a PR with the fix.

[komatits]

Perfect. Please put an "if" statement before the line in which you reduce the time step to make sure it applies to the case of model AK135 only. Thanks, Dimitri.

On 02/06/2014 01:40, Elliott Sales de Andrade wrote:

This appears to be working okay with ifort. I'll need to test with another compiler as well, and if it works, I'll open a PR with the fix.

— Reply to this email directly or view it on GitHub https://github.com/geodynamics/specfem3d_globe/issues/134#issuecomment-44792970.

Dimitri Komatitsch CNRS Research Director (DR CNRS), Laboratory of Mechanics and Acoustics, UPR 7051, Marseille, France http://komatitsch.free.fr

[komatits]

Here are the only changes I made in Dec 2012, apart from adding "_NOMUD" to some flags; the old AK135 radii (right) were wrong, the new ones (left) are fine; since they have changed, it makes sense to have to slightly reduce the time step accordingly.

[QuLogic]

Fixed in above PR.