Default parameters

02-Analytical capabilities
Post Reply
Zucca
Posts: 4
Joined: 17 Nov 2016, 09:57

Default parameters

Post by Zucca »

tools/program setting/matching parameters
Why is Min eigen value set to 0.1 when seismomatch is opened?
shouldn't it be better to set min eigen value to 0.0001 as suggested in the seismomatch help?

regards
User avatar
seismosoft
Posts: 1184
Joined: 06 Jul 2007, 04:55

Re: Default parameters

Post by seismosoft »

Indeed you are right, there is an inconsistency between the program and its help file.
By using a value of 0.0001 convergence is achieved faster, however with the current default value of 0.1 one can get better stability.
Best Regards,
Seismosoft Support
Zucca
Posts: 4
Joined: 17 Nov 2016, 09:57

Re: Default parameters

Post by Zucca »

Thanks for your prompt reply.
it is not just matter of time, when I set Min eigen value to 0.1 I cannot get a good spectral match.
On the other hand, by using a Min eigen value of 0.0001 I do get a very good spectral match.
could you please explain these results?
is there a way to attach figures to this forum?
User avatar
seismosoft
Posts: 1184
Joined: 06 Jul 2007, 04:55

Re: Default parameters

Post by seismosoft »

It is not a matter of speed vs. accuracy. It is speed vs. stability, i.e. with a value of 0.1 one can get better stability of the solution in most of the cases of matching.

Seismosoft Support
Zucca
Posts: 4
Joined: 17 Nov 2016, 09:57

Re: Default parameters

Post by Zucca »

i have sent you an email on support@seismosoft.com because I need to attach figures to make myself clear.
Please carry on the discussion on emails.
thanks
User avatar
seismosoft
Posts: 1184
Joined: 06 Jul 2007, 04:55

Re: Default parameters

Post by seismosoft »

Simply, with a value of 0.0001, you get more rapid convergence.
We also noticed from the screenshots that you use the default value for toerance 0.3.
By setting a tolearnce that high (the default is high for stability reason), essentially you allow for a 30% difference at any step. Hence the large differences with min Eigen = 0.1.
For more detailed information, please refer to the corresponding references.
Regards,
Seismosoft Support
Zucca
Posts: 4
Joined: 17 Nov 2016, 09:57

Re: Default parameters

Post by Zucca »

Dear support,
I understand i can get a faster convergence with min eigen value = 0.0001.

I have tried with tolerance 0.1 rather than 0.3 as you suggested and I performed the same excercise. I have sent you another email with screenshots.
As you can see, I get the same resuts even if I use tolerance 0.1 instead of 0.3.
I get very good spectral match if i use min eigen value 0.0001 and poor spectral matching with min eigen value of 0.1.
Tolerance was set to 0.1 in both cases,
Therefore, I don't think it is related to tolerance.


as a conclusion, I would use min eigen value = 0.0001 otherwise I cannot have a good spectral matching. there is no difference in using tolerance 0.1 or 0.3.
do you confirm it?
can you please advise?
User avatar
seismosoft
Posts: 1184
Joined: 06 Jul 2007, 04:55

Re: Default parameters

Post by seismosoft »

Yes, we confirm your findings. As suggested in the help system that recommended min Eigen Value is 1e-4.
However, note that these conclusions cannot be generalised; in other cases of records a low value might lead to instabilities.

Seismosoft Support
Post Reply

Return to “02-Analytical capabilities”