Hello, I solved the problem. During the "online" run, I was writing down the coefficients of the reduced solution u_N in data files, which were then read in a separate run to compute the error bound. I converted this writing/reading from data files to binary; the precision is now preserved.
Thanks!! Lorenzo On 5/24/16 7:19 PM, David Knezevic wrote: > Hi Lorenzo, > > It sounds like the data isn't being written to full precision in your > case. This is surprising to me, so I'd definitely like to look into > this some more. > > I suggest two options: > > 1. Make a minimal test case based on one of the RB examples in libMesh > that reproduces your issue. Then I can run that myself and look into > it. I suggest you first check if you can reproduce the issue in a > non-EIM case to check whether this issue is related to EIM or not. > > 2. Update to using RBDataSerialization. To do this you'll need to > configure libMesh wtih capnproto > <http://It%20sou%20%20This%20should%20not%20happen.%20If%20you%20can%20come%20up%20with%20a%20test%20case%20that%20reproduces%20this%20behavior,%20then%20I%20can%20look%20at%20it%20more%20carefully.%20It%20would%20also%20be%20good%20to%20isolate%20whether%20the%20issue%20is%20related%20to%20EIM%20or%20not.>. > > Switching to this approach may just fix your issue. If it doesn't then > it will still be helpful since we can directly read the capnproto > binary data using the "capnp" tool and check if it matches the data > that we expect. > > David > > > On Tue, May 24, 2016 at 12:19 PM, Zanon, Lorenzo > <za...@aices.rwth-aachen.de <mailto:za...@aices.rwth-aachen.de>> wrote: > > Hello, > > > I compute the error between the EIM-RB- and the FE-truth solution > on a training set "offline" during the RB-greedy procedure. If I > compute this same error in the "online" phase, on the same set of > training parameters, I notice a deterioration of the maximum > precision that can be reached (around 10^-8 vs. 10^-5.5). > > > My guess is that something goes lost between "offline" and > "online" during the reading/writing of the RB-data and/or of the > RB-basis functions. Would it be possibile to avoid this loss? > > > Since I started writing the code with an older libmesh version, I > am still using the functions in rb_evaluation.h: > > > legacy_write_offline_data_to_files > > legacy_read_offline_data_to_files > > > Would I get better results using the new RBDataSerialization? > > > My problem also involves EIM, but the EIM data are read-in also > before the offline RB-Greedy, so I don't think this is relevant. > > > Thanks a lot > > Lorenzo > > > -- > ============================ > Lorenzo Zanon > > Graduate School AICES > RWTH Aachen University > Schinkelstr. 2a > 52062 Aachen > Deutschland > > za...@aices.rwth-aachen.de > <mailto:za...@aices.rwth-aachen.de><mailto:za...@aices.rwth-aachen.de > <mailto:za...@aices.rwth-aachen.de>> > +49 (0)241 80 99 746 <tel:%2B49%20%280%29241%2080%2099%20746> > ============================ > > > ------------------------------------------------------------------------------ > Mobile security can be enabling, not merely restricting. Employees who > bring their own devices (BYOD) to work are irked by the imposition > of MDM > restrictions. Mobile Device Manager Plus allows you to control > only the > apps on BYO-devices by containerizing them, leaving personal data > untouched! > https://ad.doubleclick.net/ddm/clk/304595813;131938128;j > _______________________________________________ > Libmesh-users mailing list > Libmesh-users@lists.sourceforge.net > <mailto:Libmesh-users@lists.sourceforge.net> > https://lists.sourceforge.net/lists/listinfo/libmesh-users > > -- ============================ Lorenzo Zanon Graduate School AICES RWTH Aachen University Schinkelstr. 2a 52062 Aachen Deutschland za...@aices.rwth-aachen.de +49 (0)241 80 99 746 +49 (0)1577 2790439 ============================ ------------------------------------------------------------------------------ What NetFlow Analyzer can do for you? Monitors network bandwidth and traffic patterns at an interface-level. Reveals which users, apps, and protocols are consuming the most bandwidth. Provides multi-vendor support for NetFlow, J-Flow, sFlow and other flows. Make informed decisions using capacity planning reports. https://ad.doubleclick.net/ddm/clk/305295220;132659582;e _______________________________________________ Libmesh-users mailing list Libmesh-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/libmesh-users
