Hi Joao,
Thank you for your reply. I think that this kind of discussion is helpful and constructive.
With some of your comments I agree and with some other I don't.
First of all I never talked about FB elements, I was only referring to DB elements (look in my post of 30 October). So you misunderstood me in somewhat.
My main query (maybe i didn't express it clearly in my posts) it is why the user cannot use more than 2 I.P's in Seismostruct and instead of that he is obliged to use only 2 in DB elements. I think it should be a user's choice as well as the choice of the integration rule. So, if the number of I.P's is anyway 2, then the next question is: which is more preferable to use Gauss or Gaus-Lobatto rule for 2 I.P's?
In my opinion, in such case it is more preferable to use Gauss, but that depends on what is the designer's target or aim. If it is the study of element's local response I agree that DB formulation with 2 I.P's is not objective. If the designer wants to study the overal behavior of the structure then in my opinion the DB element with 2 I.P's and Gauss rule could be acceptable.
The fact that when using DB element there should be a finest discretisation along the element in order to obtain accuracy it is well known and there is no doubt about it.
Also the advantages of using FB elements instead of DB elements are also known (thought in case of localization issues one should be more careful).
Finally, I would really like to know the opinion of the programmer, I mean to know which were his intentions of using Gauss rule instead of Lobatto when using few I.P's (if it is because of my point of view or Joao's) and also why the user can't have the choice to use more than 2 I.P's for DB elements (even if it is actually useless..) and also to choose manually the integration rule.
After all, a programm should give the opportunity to the user to choose by his own the different kind of options and to judge by him self which it is suitable for each case.
Thank you.
Integration Point in DB-Element
-
seismosoft
- Posts: 1263
- Joined: 06 Jul 2007, 04:55
Re: Integration Point in DB-Element
Dear TUC,
Using more than 2 integration points in DB elements makes relatively little sense, as noted by Joao, already. This is the reason why in SeismoStruct such option is not available.
By the same token, for the very vast majority of application purposes, using a Gauss-Lobatto rule in a 2-integration points DB element brings no relevant advantage over its Gauss-Legendre counterpart.
For what concerns the issue of giving users the possibility of choosing the integration rule to be employed, we do feel such level of customisation to be slightly excessive, for a number of reasons. Hence, highly specialised users wishing to experiment with numerical models should probably make use of other currently available research software solutions (e.g. OpenSees), or, in alternative, develop their own tool.
Seismosoft Support
Using more than 2 integration points in DB elements makes relatively little sense, as noted by Joao, already. This is the reason why in SeismoStruct such option is not available.
By the same token, for the very vast majority of application purposes, using a Gauss-Lobatto rule in a 2-integration points DB element brings no relevant advantage over its Gauss-Legendre counterpart.
For what concerns the issue of giving users the possibility of choosing the integration rule to be employed, we do feel such level of customisation to be slightly excessive, for a number of reasons. Hence, highly specialised users wishing to experiment with numerical models should probably make use of other currently available research software solutions (e.g. OpenSees), or, in alternative, develop their own tool.
Seismosoft Support
Re: Integration Point in DB-Element
Hello TUC,
I think it is now clear that SeismoSoft wants to guide analysts towards clever modeling options, leaving aside those that either do not make much sense or that would simply represent a waste of computing time.
Regarding your question of whether it is preferable to use Gauss or Gauss-Lobatto rule for 2 IPs, I invite you to revisit my previous answer, point 3) (and also the reply of November 3). As I say there, Gauss-Lobatto with 2 IPs is not capable of modeling the linear elastic response. That is due to the fact that this integration scheme is only capable of accurately integrating polynomials up to degree 2n-3. So, if you have n=2 (IPs), you are only able to accurately integrate linear functions. For the linear elastic response you need at least quadratic. On the other hand, the accuracy of the Gauss-Legendre scheme is 2n-1. That is, for n=2 (IPs), you obtain cubic, which is ok.
Just a few quick extra remarks, regarding your other comments:
- To study the local response you will need to subdivide the mesh into a number of elements that is way above what is used in practice (say, you may be constrained to use 20 or 30 DB elements per member). Increasing the number of IPs won't help you on that matter (this technique will only work with FB elements). For the global behavior, it is just as you say: use a few elements per member (each with 2 Gauss IPs), and you will have an acceptable answer.
- Localization in DB elements is not any less serious than it is in FB ones, it just happens that it is slightly more masked (and that is why I consider it even more dangerous).
It was good to share these thoughts with you.
Thanks and good luck
I think it is now clear that SeismoSoft wants to guide analysts towards clever modeling options, leaving aside those that either do not make much sense or that would simply represent a waste of computing time.
Regarding your question of whether it is preferable to use Gauss or Gauss-Lobatto rule for 2 IPs, I invite you to revisit my previous answer, point 3) (and also the reply of November 3). As I say there, Gauss-Lobatto with 2 IPs is not capable of modeling the linear elastic response. That is due to the fact that this integration scheme is only capable of accurately integrating polynomials up to degree 2n-3. So, if you have n=2 (IPs), you are only able to accurately integrate linear functions. For the linear elastic response you need at least quadratic. On the other hand, the accuracy of the Gauss-Legendre scheme is 2n-1. That is, for n=2 (IPs), you obtain cubic, which is ok.
Just a few quick extra remarks, regarding your other comments:
- To study the local response you will need to subdivide the mesh into a number of elements that is way above what is used in practice (say, you may be constrained to use 20 or 30 DB elements per member). Increasing the number of IPs won't help you on that matter (this technique will only work with FB elements). For the global behavior, it is just as you say: use a few elements per member (each with 2 Gauss IPs), and you will have an acceptable answer.
- Localization in DB elements is not any less serious than it is in FB ones, it just happens that it is slightly more masked (and that is why I consider it even more dangerous).
It was good to share these thoughts with you.
Thanks and good luck
Re: Integration Point in DB-Element
Thank you Seismosoft for your intervention.
Just as notice, I didn't doubt that in DB elements there is no much sense to use more than 2 I.P's. I just tried to interpret why Seismostruct uses the Gauss integration rule and not Lobatto as it does for more than 3 I.P's. I thought that it was a special reason for that, but after your answer I understood that finally it wasn't.
Nevertheless, I am congruent with the opinion that is expressed in the paper that I had attached in a previews post about what it is more suggested to use between the two integration rules in case of using few I.P's in DB elements (although actually I don't really disagree with Joao's position except in some points).
Anyway, I really appreciate that I had the opportunity to exchange opinions about this issue from this forum and of course Joao's effort for being so comprehensive.
Joao, just as a notice: i didn’t ever have the intention to premise or promote the DB element instead of FB element. In all your posts you emphasised such thing that it was out of my question. Please try to understand this, as my question was specific and was concerning only the use of the Gauss integration rule and nothing else, although all what you said was really interesting. At the end, in your last reply you said: “"On the other hand, the accuracy of the Gauss-Legendre scheme is 2n-1. That is, for n=2 (IPs), you obtain cubic, which is ok”" that also satisfied my initial question.
That what you said, also in your last reply: “"..leaving aside those that either do not make much sense or that would simply represent a waste of computing time"” i don’t really agree with it because i have the general idea that a software must give always the opportunity to the user to choose and judge about the different kind of options by his own. Otherwise, as concern the specific issue about DB elements, someone could say that they could be excepted by the program so to use only FB elements... That is my opinion for any kind of software and not only Seismostruct of course..
About the localisation issue, i wouldn’'t like to go further because i think we will be tiring for the rest of the forum that follows our discussion as it is out of the initial subject of the request.
Thank you for your time and for exchanging all these interesting thoughts and opinions.
Regards.
P.S Seismosoft, I am lightly using OpenSees but, as you may know, there is no graphical interface so I would like to use Seismostruct "in parallel".
Just as notice, I didn't doubt that in DB elements there is no much sense to use more than 2 I.P's. I just tried to interpret why Seismostruct uses the Gauss integration rule and not Lobatto as it does for more than 3 I.P's. I thought that it was a special reason for that, but after your answer I understood that finally it wasn't.
Nevertheless, I am congruent with the opinion that is expressed in the paper that I had attached in a previews post about what it is more suggested to use between the two integration rules in case of using few I.P's in DB elements (although actually I don't really disagree with Joao's position except in some points).
Anyway, I really appreciate that I had the opportunity to exchange opinions about this issue from this forum and of course Joao's effort for being so comprehensive.
Joao, just as a notice: i didn’t ever have the intention to premise or promote the DB element instead of FB element. In all your posts you emphasised such thing that it was out of my question. Please try to understand this, as my question was specific and was concerning only the use of the Gauss integration rule and nothing else, although all what you said was really interesting. At the end, in your last reply you said: “"On the other hand, the accuracy of the Gauss-Legendre scheme is 2n-1. That is, for n=2 (IPs), you obtain cubic, which is ok”" that also satisfied my initial question.
That what you said, also in your last reply: “"..leaving aside those that either do not make much sense or that would simply represent a waste of computing time"” i don’t really agree with it because i have the general idea that a software must give always the opportunity to the user to choose and judge about the different kind of options by his own. Otherwise, as concern the specific issue about DB elements, someone could say that they could be excepted by the program so to use only FB elements... That is my opinion for any kind of software and not only Seismostruct of course..
About the localisation issue, i wouldn’'t like to go further because i think we will be tiring for the rest of the forum that follows our discussion as it is out of the initial subject of the request.
Thank you for your time and for exchanging all these interesting thoughts and opinions.
Regards.
P.S Seismosoft, I am lightly using OpenSees but, as you may know, there is no graphical interface so I would like to use Seismostruct "in parallel".