[cig-commits] [commit] devel: added two text files (97ea745)

[cig_noreply at geodynamics.org]

Repository : https://github.com/geodynamics/specfem2d

On branch  : devel
Link       : https://github.com/geodynamics/specfem2d/compare/6df54bf31534a3dbf747c199ac6669b72a826403...97ea745c241e76b7692d21be0e9f68f43d669774

>---------------------------------------------------------------

commit 97ea745c241e76b7692d21be0e9f68f43d669774
Author: Dimitri Komatitsch <komatitsch at lma.cnrs-mrs.fr>
Date:   Thu Jan 22 00:57:45 2015 +0100

    added two text files


>---------------------------------------------------------------

97ea745c241e76b7692d21be0e9f68f43d669774
 ..._on_the_adjoint_tomography_workflow_dec2014.txt | 80 ++++++++++++++++++++++
 ...current_adjoint_tomography_workflow_dec2014.txt | 37 ++++++++++
 2 files changed, 117 insertions(+)

diff --git a/Carl_Tape_his_comments_on_the_adjoint_tomography_workflow_dec2014.txt b/Carl_Tape_his_comments_on_the_adjoint_tomography_workflow_dec2014.txt
new file mode 100644
index 0000000..ffadaae
--- /dev/null
+++ b/Carl_Tape_his_comments_on_the_adjoint_tomography_workflow_dec2014.txt
@@ -0,0 +1,80 @@
+
+Subject: Re: Skype call reminder
+From: Carl Tape
+Date: 12/22/2014 06:37 AM
+To: Ryan Modrak
+CC: Dimitri Komatitsch, Daniel Peter , David Luet , Ebru Bozdag , Zhinan Xie, Matthieu Lefebvre , Jeroen Tromp , Federica Magnoni , Emanuele Casarotti
+
+Hi Ryan and all,
+
+I met with Ebru, Emanuele, Frederica, Elliott, and Lion on Thursday
+night at AGU. (We scheduled at the last minute and missed Min and
+Hejun, unfortunately.) Below are some discussion points and questions.
+You already have plenty to discuss anyway, so you don’t have to
+consider any of these, of course. I cced Emanuele and Frederica,
+though I don’t think they’ll be back by Monday.
+
++ It seems like those in our discussion were supportive of the move
+from “legacy” utilities (perl, sac, flexwin, measure_adj) to
+python-based utilities (obspy, pyflex, pyadj). Lion seemed quite
+willing to rewrite measure_adj into python (pyadj) – the multitaper
+measurement is the key piece, and there are Slepian tapers within
+python libraries, apparently.
+
++ Lion suggested that, in his opinion, the hdf5 file format would be
+optimal for seismograms and adjoint sources. I don’t know the details
+about this. But clearly this is an important decision, since it
+impacts the processing scripts, adjoint sources, the moment tensor
+inversion codes, etc.
+
++ For the early part of the adjoint tomo workflow, it seems like LASIF
+(http://krischer.github.io/LASIF/) might provide some organizational
+structure. It sounded like this was complementary to the inversion
+part of the workflow that Ryan has been discussing. I plan on
+exploring LASIF in January.
+
++ The seismogram-based calculation of numerical resolution is a minor
+part of the workflow. But also something that can be done easily, once
+the two sets of simulations are performed. I do it in matlab (with the
+waveform toolbox), but this can be done faster and better in obspy,
+too. Here is my AGU presentation slides on this topic:
+
+https://drive.google.com/file/d/0Bym088u-rKlvMnVrSk1hajhBQWc/view?usp=sharing
+
++ What constitutes a need to move a variable from constants.h to
+Par_file? Is it based on how often people use these variables? We
+discussed APPROXIMATE_HESS_KL (inversion), USER_T0 (inversion), and
+NGLLX (minimum resolvable period).
+
+constants file:
+https://github.com/geodynamics/specfem3d/blob/devel/setup/constants.h.in
+
++ Are the src/tomograhy/ files different in SPECFEM3D and SPECFEM3D_GLOBE?
+
++ My impression is that for the optimization part, there are several
+efforts, including those at Princeton, those by Dimitri, as well as
+from outside the group. (Andreas Fichtner has someone organizing some
+basic capabilities for this part of the workflow.) Furthermore, there
+are libraries within petsc (http://www.mcs.anl.gov/petsc/) that are
+broadly used in parallel solvers.
+
+I write all this from the perspective of someone who has done very
+little regarding the adjoint tomography workflow since 2009. But I
+have several projects that will move forward in 2015, lead by me
+(southern California), my student Vipul Silwal (Alaska), my postdoc
+Amir Allam (fault zones up to 5-10 Hz), and Yoshihiro Kaneko (New
+Zealand). I’d like to get these folks on the right track and am open
+to trying out new things. Hopefully we can help in this effort.
+
+I’ll talk to you soon. Thanks for including me in the call.
+
+Carl
+
+---------------------------------------------------------------
+Carl Tape
+Assistant Professor
+Geophysical Institute (office 413D)
+University of Alaska Fairbanks
+Web: http://www.giseis.alaska.edu/input/carl/
+---------------------------------------------------------------
+
diff --git a/Daniel_Peter_his_comments_on_the_current_adjoint_tomography_workflow_dec2014.txt b/Daniel_Peter_his_comments_on_the_current_adjoint_tomography_workflow_dec2014.txt
new file mode 100644
index 0000000..2c179f9
--- /dev/null
+++ b/Daniel_Peter_his_comments_on_the_current_adjoint_tomography_workflow_dec2014.txt
@@ -0,0 +1,37 @@
+
+Subject: Re: Skype call reminder
+From: daniel peter
+Date: 12/22/2014 06:01 PM
+To: Carl Tape
+CC: Ryan Modrak, Dimitri Komatitsch, David Luet, Ebru Bozdag, Zhinan Xie, Matthieu Lefebvre , Jeroen Tromp , Federica Magnoni , Emanuele Casarotti , lion.krischer
+
+Hi Carl,
+
+for 1.
+i kept the routines very similar, some differences occur when reading in mesh files and e.g. where the maximum of the gradient is taken for the update step length. in future, i want to have them also merged into the same set of SPECFEM function library.
+
+for 2.
+good point, the jacobian is used for smoothing of the kernels in the global code, but isn’t currently used for the cartesian code. that is one of the differences that has to be checked again (thus smoothing in the cartesian is based solely on point distance, not volume)
+
+best wishes,
+daniel
+
+
+> > On 22 Dec 2014, at 5:50 pm, Carl Tape wrote:
+> > 
+> > Hi all,
+> > 
+> > Thanks for the call. I wanted to ask two questions.
+> > 
+> > 1. Daniel, can you briefly describe the differences between the
+> > src/tomograhy/ files in SPECFEM3D and SPECFEM3D_GLOBE? Would it be
+> > simpler to have only one set of files?
+> > 
+> > 2. Isn't the jacobian of the mesh needed for many of the optimization
+> > operations (dot product, integration, etc)? So my impression is that
+> > you need more than just the kernel name to do something like smoothing
+> > -- you also need the jacobian. Probably this is already in the
+> > scripts, but the discussion in the call did not mention it.
+> > 
+> > Carl
+> >