GitLab

The sinus3 source wavelet was shifted in the wrong direction.

Added references to D. Koehn's dissertation. Large parts of the theory are extracted from his dissertation.

The vector FL was allocated before the value L was exchanged.

This fix will reset the conjugate gradient history each time a frequency or workflow stage is switched.
If such a stage is switched we will have a "new" objective function, thus, it makes no sense to use the old gradients for speeding up the convergence.
This bug was the reason why the objective function could not be reduced for 2-3 iterations after a new frequency stage was appplied.

I switched the naming of the variables of the frequency filtering to a more clear and easy one.
The change affects the JSON input file and the code itself.
The following renaming is used:
FC_START -> F_LOW_PASS_START
FC_END   -> F_LOW_PASS_END
FC_INCR  -> F_LOW_PASS_INCR
FC_EXT   -> F_LOW_PASS_EXT
FC       -> F_LOW_PASS
F_HP     -> F_HIGH_PASS

However, old JSON files will also work for now. If one of the old variables is not found, IFOS will search for the corresponding old one. This option will be removed in a future release.

This option is not supported anymore. Therefore I remove it from the source code.

With the option TRKILL_STF_OFFSET_INVERT it is possible to invert the sense of the STF Offset window.
So TRKILL_STF_OFFSET_INVERT==1 means, do NOT kill the specified offset, instead kill all other offsets.

laura.gassner authored Mar 02, 2016 and Florian Wittkamp committed Mar 11, 2016

# Conflicts:
#	src/IFOS2D.c
#	src/wavelet.c

If VERBOSE==1 the seismograms (observed and synthetic) for STF will be written out.

In case of the SH waves, the STF wavelet was NOT used. To calculate the synthetic data, a new synthetic wavelet was calculated.

This reverts commit 6ae350d1.

# Conflicts:
#	src/IFOS2D.c

This reverts commit 0e8335b9.

All new files and folder in the par folder will be ignored from git.
This is usefull, as there is no big development inside the par folder. However, all production runs will produce a lot of files inside par, which are never needed to be under version control.

When L-BFGS is used and no Wolfe step length could be found, the L-BFGS needs either be resetted or the algorithm has to switch to the next workflow line.
This fix will let the algorithm switch to the next workflow line without a model update.

The STF wavelet was not filtered with the frequency filter. Therefore, the residuals were calculated between the unfiltered synthetic and the filtered observed seismograms.

At a new frequency stage, STF was deactivated after the first shot.

If TRKILL_STF_OFFSET or TRKILL_OFFSET is set to 2, the offset based TraceKill will be applied in addition to the TraceKill file.
Therefore, it is possible to exclude with TraceKill file fault receiveres and use an offset based tracekill with the build in feature at the same time.

Implemented an offset based tracekill for STF.
Makefile in par folder has now two new targets:
1. make install, do an install from precompiled files
2. make reinstall, do a clean install

(1.)
Renamed function err() to declare_error().
This was a problem, as many standard libraries also has functions err().

(2)
I fixed a lot of compiler warnings like unused variables and uninitialised variables. However, here is a lot of work to do.

(1.) IFOS Makefile
Introduce dependencies files, which contain dependencied of each objective file. Therefore, it is possible to allow a more accurate rebuilt of the objective files.
New standard target is IFOS2D. Moreover a new target install is available.

(2.) Contrib
Fixed the problem with the dependencies files if the hole IFOS folder is copied over to a different hardware architecture.
You may now the error by ".d" files.

The readmod functions verify now the dimension of the model file. Therefore IFOS will abort, if the model file is to small or too big.
Moreover, the write_matrix_disk function is used to write the models to disk.

Bevore the FWI loop the observd data was read in and was time filtered. I think this ist not necessary, as the observed data is read in for the calculation of the residuals again.
Therefore, I removed the code to save calculation time.

IFOS is now able to read in input files, that are formatted according the JSON standard. Nevertheless, the old files will work as well.

In addition, I fixed some compiler warnings.

Conflicts:
	src/exchange_par.c
	src/write_par.c