[CIG-SEISMO] Source Time function

[Dimitri Komatitsch]

Hi Kasper,

I agree; I think it is not safe to make such an assumption about 
numerical noise (because it depends on the compiler options you use and 
several other things; i.e. the assumption may be OK on a given machine 
but not on another).

I would also suggest switching to C-PML absorbing conditions instead of 
Stacey, which leads to much better sensitivity kernels (no spurious 
noise mapped back) in the case of adjoint calculations. CPML is already 
available in the 2D version of the code and will be released next week 
for the 3D code (the current 3D version had bugs, which we have now 
fixed; we will be done testing sometime next week).

Best,
Dimitri.

On 04/17/2013 03:28 AM, Jeroen Tromp wrote:
> Hi Kasper:
>
> It's possible, but I still would recommend low passing the seismograms
> to bring them within the numerical validity range. If you are doing a
> similation in a finite domain, e.g., a box with three absorbing
> boundaries and a free surface, then, yes, you also need to use a low
> frequency cut off to account for this. Thus a bandpass is the best option.
>
> Best regards,
>
> Jeroen
>
>
>
> On 4/16/13 5:27 PM, Kasper VanWijk wrote:
>> Kia Ora Jeroen!
>>
>> Greetings from Auckland. I noticed your advice to Martin about source
>> time functions. As you know, I am mostly an experimental guy, so
>> forgive me if I ask a trivial question, but I am working with a
>> student on extending Aki's SPAC method. It involves the summation of
>> many sources. When we use the Dirac option in SPECFEM2D (i.e. simulate
>> the Green function response), we get what you told Martin: good
>> signals in the range that the mesh supports. We have noticed, however,
>> that the high-frequency numerical noise seems to stack out. Is it well
>> known that this high-f noise is (apparently) uncorrelated? Or are our
>> eyes deceiving us?
>>
>> Also, are you aware of limitations on the LOW frequency information in
>> the data with the Dirac source,  or should those be  well represented?
>>
>> groet,
>> kasper
>>
>>
>>
>>
>> On Tue, Apr 16, 2013 at 4:18 AM, Jeroen Tromp <jtromp at princeton.edu
>> <mailto:jtromp at princeton.edu>> wrote:
>>
>>     Dear Martin:
>>
>>     Yes, you can. The way to do this is to calculate the imuplse response,
>>     i.e., the Green's function, and subsequently convolve with
>>     whatever you
>>     choose to be your source-time function in postprocessing. Note
>>     that you
>>     will still have to bandpass the result to bring it within the
>>     numerical
>>     validity range, i.e., make sure you are resolving all the
>>     frequencies of
>>     interest with your mesh.
>>
>>     Best regards,
>>
>>     Jeroen
>>
>>
>>     On 4/11/13 11:30 AM, Martin Martin wrote:
>>     > Dear SPECFEM2D Team,
>>     >
>>     > I would like to know if it is possible to create my own source
>>     function
>>     > - I mean other than using the five time function which is provided :
>>     > Rickett, gauss, heaviside, etc ?
>>     >
>>     > Thank you
>>     >
>>     > Best wishes,
>>     >
>>     > Martin
>>     >
>>
>>     _______________________________________________
>>     CIG-SEISMO mailing list
>>     CIG-SEISMO at geodynamics.org <mailto:CIG-SEISMO at geodynamics.org>
>>     http://geodynamics.org/cgi-bin/mailman/listinfo/cig-seismo
>>
>>
>
>
>
> _______________________________________________
> CIG-SEISMO mailing list
> CIG-SEISMO at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-seismo
>

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