From: Paul T. B. <ptb...@gm...> - 2016-03-08 17:41:41
|
Precision of the output in xda format? On Tue, Mar 8, 2016 at 12:27 PM, Derek Gaston <ga...@mi...> wrote: > (Note: this is cross-posted between MOOSE and libMesh devel lists because > it's interesting to both!) > > I had a user here at MIT bring me this interesting problem. She developed > her code last year and got good results (i.e.: it worked last year). She > did an internship for 6 months and came back to the code only recently came > back to the code and her same stuff doesn't work anymore. > > What she is doing is writing out XDA format meshes from one simulation and > reading them into the next simulation. What's happening is that the XDA > format meshes are "corrupt" for some reason. We finally tracked it down to > happening when her mesh gets really fine due to adaptivity. I've been able > to replicate this in a small MOOSE input file (runs with moose_test) that I > have attached. > > The original mesh looks like this: > > [image: initial_mesh.png] > > After 5 adaptivity steps it looks like this: > > [image: corruption.png] > > To generate that I outputted the file in XDA format and then just read it > in and converted it to Exodus. > > The really weird thing about this problem is that it runs just fine... and > the Exodus output looks just fine: > > [image: solution.png] > > So the mesh is NOT actually corrupt in memory. In addition, earlier XDA > files look just fine. Here it is after 4 adaptivity steps: > > [image: four.png] > > Anyone have an idea of what's going on? > > Derek > > > > > > > ------------------------------------------------------------------------------ > Transform Data into Opportunity. > Accelerate data analysis in your applications with > Intel Data Analytics Acceleration Library. > Click to learn more. > http://makebettercode.com/inteldaal-eval > _______________________________________________ > Libmesh-devel mailing list > Lib...@li... > https://lists.sourceforge.net/lists/listinfo/libmesh-devel > > |