[cig-commits] [commit] devel, master: added Jeroen's second set of explanations about how to use PARTIAL_PHYS_DISPERSION_ONLY right to the users manual. (301191e)

Thu Nov 6 08:31:44 PST 2014

Repository : https://github.com/geodynamics/specfem3d_globe

On branches: devel,master
Link       : https://github.com/geodynamics/specfem3d_globe/compare/bc58e579b3b0838a0968725a076f5904845437ca...be63f20cbb6f462104e949894dbe205d2398cd7f

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

commit 301191ef4d74f79dd50e1d23ca9fb4d0f7f638f7
Author: Dimitri Komatitsch <komatitsch at lma.cnrs-mrs.fr>
Date:   Tue Sep 23 14:56:46 2014 +0200

    added Jeroen's second set of explanations about how to use PARTIAL_PHYS_DISPERSION_ONLY right to the users manual.


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

301191ef4d74f79dd50e1d23ca9fb4d0f7f638f7
 doc/USER_MANUAL/bibliography.bib           |  12 ++++++++++++
 doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.pdf | Bin 24378184 -> 24382097 bytes
 doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.tex |  24 ++++++++++++++++--------
 3 files changed, 28 insertions(+), 8 deletions(-)

diff --git a/doc/USER_MANUAL/bibliography.bib b/doc/USER_MANUAL/bibliography.bib
index c323a0e..1df6f51 100644
--- a/doc/USER_MANUAL/bibliography.bib
+++ b/doc/USER_MANUAL/bibliography.bib
@@ -24,6 +24,18 @@
 @String { pageoph    = {Pure Appl. Geophys.} }
 @String { pepi       = {Phys. Earth Planet. In.} }
 
+ at Article{ZhLiTr11,
+  Title                    = {Surface wave sensitivity: mode summation versus adjoint {SEM}},
+  Author                   = {Zhou, Ying and Liu, Qinya and Tromp, Jeroen},
+  Journal                  = gji,
+  Year                     = {2011}, 
+  Number                   = {3},
+  Pages                    = {1560-1576},
+  Volume                   = {187},
+  Doi                      = {10.1111/j.1365-246X.2011.05212.x},
+} 
+  
+
 @Article{Zak09novel,
   Title                    = {A novel formulation of a spectral plate element for wave propagation in isotropic structures},
   Author                   = {{\.Z}ak, A.},
diff --git a/doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.pdf b/doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.pdf
index 75c935c..8776dc7 100644
Binary files a/doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.pdf and b/doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.pdf differ
diff --git a/doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.tex b/doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.tex
index 0057de3..7c9e767 100644
--- a/doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.tex
+++ b/doc/USER_MANUAL/manual_SPECFEM3D_GLOBE.tex
@@ -903,15 +903,23 @@ requires twice as much CPU time. This feature is not used at the time
 of meshing but is required for the solver, i.e., you may change this
 parameter after running the mesher.
 \item [{\texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY or UNDO\_ATTENUATION}}] To undo attenuation for sensitivity kernel calculations or forward runs with \texttt{SAVE\_FORWARD}
-use one (and only one) of the two flags below. \texttt{UNDO\_ATTENUATION} is much better (it is exact)
-but requires a significant amount of disk space for temporary storage.
-When using \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY}, to make the approximation reasonably OK you need to take the following steps:
-%
+use one (and only one) of the two flags below. \texttt{UNDO\_ATTENUATION} is much better (it is exact) and is simpler to use
+but it requires a significant amount of disk space for temporary storage. It has the advantage of requiring only two simulations for adjoint tomography instead
+of three in the case of \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY}, i.e. for adjoint tomography it is globally significantly less expensive (each run is slightly more
+expensive, but only two runs are needed instead of three).\\
+When using \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY}, to make the approximation reasonably OK you need to take the following steps:\\
 1/ To calculate synthetic seismograms, do a forward simulation with full attenuation for the model of interest. The goal is to get synthetics that match the data as closely as possible.\\
-2/ Make measurements and produce adjoint sources by comparing the resulting synthetics with the data. In the simplest case of a cross-correlation traveltime measurement, use the time-reversed synthetic in the window of interest as the adjoint source.\\
-2/ Do a second forward calculation with \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.} and save the last snapshot.\\
-3/ Do an adjoint calculation using the adjoint source calculated in 1/, the forward wavefield reconstructed based on 2/, use \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.} for the adjoint wavefield, and save the kernel.\\
-Thus the kernel calculation uses \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.} for both the forward and the adjoint wavefields. This is in the spirit of the banana-donut kernels. But the data that are assimilated are based on the full 3D synthetic with attenuation.
+2a/ Make measurements and produce adjoint sources by comparing the resulting synthetics with the data. In the simplest case of a cross-correlation traveltime measurement, use the time-reversed synthetic in the window of interest as the adjoint source.\\
+2b/ Do a second forward calculation with \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.} and save the last snapshot.\\
+3/ Do an adjoint calculation using the adjoint source calculated in 1/, the forward wavefield reconstructed based on 2b/, use \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.} for the adjoint wavefield, and save the kernel.\\
+Thus the kernel calculation uses \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.} for both the forward and the adjoint wavefields. This is in the spirit of the banana-donut kernels. But the data that are assimilated are based on the full 3D synthetic with attenuation.\\
+%
+Another, equivalent way of explaining it is:\\
+1/ Calculate synthetics with full attenuation for the current model. Compare these to the data and measure frequency-dependent traveltime anomalies $\Delta \tau(\omega)$, e.g.. based upon multi-tapering.\\
+2/ Calculate synthetics with \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.} and construct adjoint sources by combining the seismograms from this run with the measurements from 1/. So in the expressions for the multi-taper adjoint source you use the measurements from 1/, but synthetics calculated with \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.}.\\
+3/ Construct a kernel by calculating an adjoint wavefield based on the sources constructed in 2/ and convolving it with a forward wavefield with \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.}. Again, both the forward and adjoint calculations use \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.}.\\
+Note that if you replace multi-taper measurements with cross-correlation measurements you will measure a cross-correlation traveltime anomaly in 1/, i.e., some delay time $\Delta T$. Then you would calculate an adjoint wavefield with \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.} and use the resulting time-reversed seismograms weighted by $\Delta T$ as the adjoint source. This wavefield interacts with a forward wavefield calculated with \texttt{PARTIAL\_PHYS\_DISPERSION\_ONLY = .true.}.
+If $\Delta T=1$ one gets a banana-donut kernel, i.e., a kernel for the case in which there are no observed seismograms (no data), as explained for instance on page 5 of \cite{ZhLiTr11}.
 %
 \item [{\texttt{MOVIE\_SURFACE}}] Set to \texttt{.false.}, unless you want
 to create a movie of seismic wave propagation on the Earth's surface.

