akuhara / rf_inv

Receiver function inversion by reversible-jump Markov-chain Monte Carlo

Note

This software is no longer being updated. Please consider adopting our new software, SEIS_FILO (https://github.com/akuhara/SEIS_FILO), which offers enhanced features like joint inversion with dispersion curves.

RF_INV

Transdimensional inversion of receiver function waveforms by reversible-jump Markov-chain Monte Carlo

(c) 2018-2019 Takeshi Akuhara (Email: akuhara @ eri.u-tokyo.ac.jp)

Any bug report and suggestions are welcome!

Features

• Applicable to OBS & borehole station
  • Model can include the sea water on its top. Also the station may be buried.
• Multiple input traces
  • Can asign different ray parameter and Gaussian-filter for each trace.
• Can use S receiver functions
  • Joint inversion of P and S receiver functions is also possible.
• Parallel tempering
  • More efficient than conventional MCMC.
• Invert for velocity perturbation
  • Non-uniqueness of inversion can be mitigated by constraint from a reference velocity model.

See Wiki for more details.

Terms of Use

• Please clarify the URL of the GitHub repository (https://github.com/akuhara/RF_INV) and developer's name (Takeshi Akuhara) when you make any presentation or publish articles using this program.
• This program is licensed under the GNU General Public License v3.0.

Requirements

• FFTW library
• LAPACK library
• Open MPI

---

Manual

A manual is available here.

rf_inv's Issues

Issue in forward calculation

Erruneous forward calculation when

• Ocean layer exists
  and
• DECONV_MODE=0

Minimum layer thickness

Prohibit thin layer by wavelength criterion.

RF normalization

Should accept different ways of RF normalization.

• Relative to vertical component record or incident wave?
• Compensate for energy loss by filter or not?

Some possible compiling error and demo running error

It seems that the whole program has been revised, so I reinstalled the package for inversion.

When I compile this program with OpenMPI-5.0.0, it showed the Error, etc.

mpif-sizeof.h:66:48:

Error: Fortran 2008: The symbol ‘real128’, referenced at (1), is not in the selected standard
mpif-sizeof.h:67:14:

Error: Parameter ‘real128’ at (1) has not been declared or is a variable, which does not reduce to a constant expression

The -std option in Makefile may cause it, and it seems that f2003 is incompatible with those ‘real128’ declarations in the mpi header. Maybe we could change the FFLAGS into

FFLAGS = -g -Wall -std=f2008  -pedantic -fbounds-check -O0 -Wuninitialized \
            -ffpe-trap=invalid,zero,overflow -fbacktrace \
            -fno-range-check

The second error might be a little bit tricky. When I ran the demo, it showed the error

ERROR: cannot create ./rslt/params.in.copy

That is because there is no folder named "rslt", and it might be fixed with

mkdir rslt

before running this demo.

Problem in Parameter file (params.in) of demo

When I installed the package and ran the program with the file "params.in" in demo, I receive:

ERROR: cannot create Note:/params.in.copy

I think the reason why it happened is that : There is a mistake in params.in

# Output directory
#---------------------------------------------------------
# OUTDIR: output directory 
          Note: quotation mark is necessary
'./result'
#

There should be a # before "Note: quotation mark is necessary"

When I add the # before the "Note: quotation mark is necessary", I receive:

Sigma is solved for trace 1
At line 240 of file src/params.f90
Fortran runtime error: Bad real number in item 2 of list input

Then I check the fortran code of "params.f90", I found that "line 240" is the parameter of sigma.
According to the demo of params.in, there should be 3 line about "Min./Max. of noise sigma". In order words, This part of "params.in"

# Min./Max. of noise sigma
0.005 0.08
#

should be changed into

# Min./Max. of noise sigma
0.005 0.08
0.005 0.08
0.005 0.08
#

Besides, before running this demo, we should prepare the SAC files and "reference.mod" file.

In the view of above, I think that if there is a specific demo, this program will be used easier.

Bug: forward computation

There is a bug related to borehole option, leading to failure of forward computation.

Problem with a little big ray parameter

hello akuhara!
when I try to invert my own data setting a little bit large ray parameter like 0.15(s/km) in the param file, it just crashed and report an error like :
Program received signal SIGFPE: Floating-point exception - erroneous arithmetic operation.
but when using values around 0.06-0.1(s/km) like those in the examples, it works well.
Is it a problem of the propagator matrix itself or I set something else wrong?
yangjudiao

Bug: determinant in likelihood

Need to calculate determinant when multiple Gaussian parameters are used.