OK! Noted. I would like to have your guidance on another point. I see that the system matrices and vectors required for the reduced basis and EIM computations are managed by unique pointers in the rb and eim construction classes. For hp EIM, multiple copies of these are required. This means that there will have to be multiple copies (in memory) of these matrices and vectors. From your experience, do you feel that this could cause memory shortage issues for large problems?
Best regards, Nikhil On Wed, May 1, 2019 at 4:10 PM David Knezevic <david.kneze...@akselos.com> wrote: > Hi Nikhil, > > >> I have started to implement hp-EIM. I have a question about combining hp >> EIM with the reduced basis method. In standard EIM, the theta objects from >> the eim evaluation object are attached to the rb theta expansion of the rb >> evaluation object. In hp EIM there are multiple eim evaluation objects. >> > > OK, that's good to hear! > > >> My question is this: >> >> Is it possible to attach different eim theta objects to the rb theta >> expansion depending on which EIM evaluation objects corresponds to the >> parameter values in question? Consider the greedy procedure of the rb >> construction object. Is it possible to attach different eim theta objects >> to the rb theta expansion during different iterations of the greedy >> procedure? >> > > I think you would have to make a separate set of rb objects for each > region of the hp EIM. It doesn't make sense to me to combine them into one > overall expansion, since each hp region may have a different number of > terms and hence the training needs to be performed separately for each > case. If you do it that way, then the code should look mostly like the > current EIM code, except you have one rb expansion for each hp region. At > least that's how I would approach it, I think. > > Best regards, > David > > > > > > On Tue, Apr 2, 2019 at 4:48 PM David Knezevic <david.kneze...@akselos.com> > wrote: > >> On Tue, Apr 2, 2019 at 10:39 AM Nikhil Vaidya <nikhilvaidy...@gmail.com> >> wrote: >> >>> Hi David, >>> >>> Thanks for your reply. Could you give me some tips for extending the >>> current EIM code? >>> >> >> You'd have to read through the EIM code and familiarize yourself with >> it... I don't see any blockers to supporting hp-EIM, but it would take some >> work to make sure you're managing all the data correctly. I don't have any >> specific suggestions I'm afraid, other than to make sure you understand the >> current code (I can answer questions on that if needed) and then you should >> be able to see how it could be extended to do what you need (via >> subclassing the relevant parts, for example). >> >> Best, >> David >> >>> _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
