convergence and tolerance

02-Analytical capabilities
Post Reply
user
Posts: 62
Joined: 16 Mar 2008, 09:44

convergence and tolerance

Post by user » 27 Jan 2011, 14:24

Hi,
We tried SeismoMatch in one of our projects, and noticed that it is rarely converging.
When we change the defaults especially max period, the chance of converging is very low.
Also tolerance 0.3 is very big value.
Do you have any comments on these?

User avatar
seismosoft
Posts: 565
Joined: 06 Jul 2007, 04:55

Re: convergence and tolerance

Post by seismosoft » 27 Jan 2011, 14:32

Hi,

1a) As explained in the Help System, we do not believe that the tolerance of 0.3 is necessarily too high. On the contrary it allows one to modify less the original accelerograms in question, and, through an adequate selection of records, still comply to lower tolerances on the average of the accelerograms, rather than on the single individual records.

1b) In any case, one can obviously change the tolerance to whatever value one wants to.


2a) Again, as discussed in the Help System (and cited scientific publications), if one increases the Tmax but keeps the Tmin unaltered, then the imposed spectrum matching range becomes very large and, depending on the characteristics of the original record, convergence may be difficult to achieve.

2b) If one really needs to match records for a wide range of periods, then one should select original records whose spectrum is as close as possible to the target spectrum, so as to minimise the required adjustments/changes.

2c) the above applies obviously to all cases; the less a records needs to be changed (i.e. the faster it converges), the better.

Seismosoft Support

Alban Kita
Posts: 7
Joined: 23 Jun 2018, 17:02

Re: convergence and tolerance

Post by Alban Kita » 26 Nov 2018, 12:13

Dear Seismosoft team,

my name is Alban Kita and i am working on adjusting 4 earthquake accelerograms to match a specific target response spectrum (my spectrum target is the one of Christchurch earthquake occurred on February 21th 2011).
If i check the average and maximum misfit i get very high values (from 117% up to 830%) and result of matching NOT CONVERGED. But visually i achieve a very good matching, an almost totally overlapping of the 4 spectra to the target spectrum (options Response spectra ---> Matched accelerograms).

Which results should i trust: the visually matching or the values of average and maximum misfit with non convergence result?
I thank you in advance,
Alban Kita

huffte
Posts: 827
Joined: 22 Jul 2011, 10:19
Location: Cookeville, Tennessee, USA
Contact:

Re: convergence and tolerance

Post by huffte » 26 Nov 2018, 14:36

It seems that the plot is the key, Alban Kita. You might be well advised to save the matched records, load them into SeismoSpect, and generate the elastic spectra there just to make certain the plot is correct. You could then overlay the target and ensure the match is achieved over the desired range of periods. I am not sure why the discrepancy in reported mis-match is occurring though. Does it happen with all 4 accelerograms?
Tim Huff

Alban Kita
Posts: 7
Joined: 23 Jun 2018, 17:02

Re: convergence and tolerance

Post by Alban Kita » 27 Nov 2018, 12:04

Thank you very much for your immediate response. I saved the matched records and with them i generated the elastic spectra. The plot was correct, indeed, these elastic spectra are identically to the matched spectra obtained by Seismomatch, also overlaying the target spectrum.
The discrepancy in reported mis-match is occurring in all 7 accelerogramms if the target spectrum is the spectrum of Christchurch earthquake (Case 1).
This discrepancy do not occur if the target spectrum is the one design spectrum of the New Zealand code, a very more regular spectrum (Case 2).
Last night i sent an email to support@seismosoft.com with all screenshots enclosed, like the following:
I am performing matching with min and max period equal to 0.02 and 4 sec, respectively, s. factor=1 and tolerance=0.3, and 50 iterations
Case 1: after matching, if i check the average and maximum misfit i get very high values (27%-410%) and result of matching NOT CONVERGED. Visually i achieve a very good matching (for 5 accelerograms), with an almost totally overlapping of the spectra to the target spectrum. Also, the mean matched spectrum is really near the target spectrum.
This problem doesn't happen in the Case 2, when matching is performed with respect to the spectrum prediced by New Zealand code.
Which results should i trust for Case 1: the very good visually matching Vs the bad values of average and maximum misfit with non convergence result?
Thank you very much huffte
Alban Kita

User avatar
seismosoft
Posts: 565
Joined: 06 Jul 2007, 04:55

Re: convergence and tolerance

Post by seismosoft » 27 Nov 2018, 14:04

The reasons why you are getting large divergence are (1) you have a very low start period for the matching (0.02 sec). SeismoMatch cannot do good matching in such low periods, and (2) you get a large divergence in higher periods.

Obviously it is difficult for any algorithm to adapt an existing record (note that the records are not created from scratch) in such a large period range. In general, you should focus on the period range of interest for easier convergence.

Seismosoft Support

User avatar
seismosoft
Posts: 565
Joined: 06 Jul 2007, 04:55

Re: convergence and tolerance

Post by seismosoft » 27 Nov 2018, 14:06

Note also that your target spectrum has a period for T=0 sec equal to zero, which obviously makes things even more difficult for the algorithm to converge. There was something wrong in reading the target spectrum there.

Seismosoft Support

Alban Kita
Posts: 7
Joined: 23 Jun 2018, 17:02

Re: convergence and tolerance

Post by Alban Kita » 28 Nov 2018, 17:14

Dear seismosoft

i thank you very much for your responses,
best regards
Alban Kita

Post Reply