## Re: [Libmesh-users] Re: wave equation and newmark scheme

 Re: [Libmesh-users] Re: wave equation and newmark scheme From: Ondrej Certik - 2005-05-17 15:46:21 ```On Tue, May 17, 2005 at 09:16:52AM -0500, John Peterson wrote: > Steffen Petersen writes: > > > > Hi Ondrej, > > > > I have just checked example 8. Everything works fine for me, > > i.e. I could not reproduce the unstable behavior you describe. > > Find attached a plot of the results at node 274 you should get > > when using the default values. > > Would you be using LasPack by any chance while Steffen is using > PETSc? Yes, I am using LasPack. I try petsc (I have a problem with large meshes on my debian, as was already reported on this list, so I must recompile it from source). If it works, there must be a bug in LasPack. ```

 [Libmesh-users] wave equation and newmark scheme From: Ondrej Certik - 2005-05-17 03:39:12 ```Hello, first I would like to thank for the library - it's wonderful. I have a question - I tried to solve the wave equation using the NewmarkSystem class in 2D and the solution always diverged in time (I based my code on example 8). I found that setting the newmark parameters alpha=1.0 and delta=2.0 helps - it seems I'am getting the right solution, so I am fine. But I thought that 0
 [Libmesh-users] Re: wave equation and newmark scheme From: Ondrej Certik - 2005-05-17 08:42:32 ```The solution in example 8 (wave equation) diverges on my computer: when I look at the pressure at the node 274 in "pressure_node.res", I got numbers like 1E80. When I make this change to the ex8.C: t_system.set_newmark_parameters(delta_t,1.0,2.0); (I set the alpha=1.0, delta=2.0), the solution seems to be ok (the pressure at the node 274 is less then 1.0 in absolute value, as should be) and the wave propagates through the pipe. I use CVS version of libmesh. Is the problem only on my computer, or is there a bug somewhere? I got the same behaviour in all my programs using the Newmark scheme (I tried both the wave and heat equation). Ondrej Certik ```
 AW: [Libmesh-users] Re: wave equation and newmark scheme From: Steffen Petersen - 2005-05-17 13:48:36 Attachments: pressure_node274.png ```Hi Ondrej, I have just checked example 8. Everything works fine for me, i.e. I could not reproduce the unstable behavior you describe. Find attached a plot of the results at node 274 you should get when using the default values. Steffen > > The solution in example 8 (wave equation) diverges on my > computer: when I look at the pressure at the node 274 in > "pressure_node.res", I got numbers like 1E80. > > When I make this change to the ex8.C: > > t_system.set_newmark_parameters(delta_t,1.0,2.0); > > (I set the alpha=1.0, delta=2.0), the solution seems to be ok > (the pressure at the node 274 is less then 1.0 in absolute > value, as should > be) and the wave propagates through the pipe. > > I use CVS version of libmesh. Is the problem only on my > computer, or is there a bug somewhere? I got the same > behaviour in all my programs using the Newmark scheme (I > tried both the wave and heat equation). > > Ondrej Certik > > > ------------------------------------------------------- > This SF.Net email is sponsored by Oracle Space Sweepstakes > Want to be the first software developer in space? > Enter now for the Oracle Space Sweepstakes! > http://ads.osdn.com/?ad_id=7412&alloc_id=16344&op=click > _______________________________________________ > Libmesh-users mailing list > Libmesh-users@... > https://lists.sourceforge.net/lists/listinfo/libmesh-users > ```
 AW: [Libmesh-users] Re: wave equation and newmark scheme From: John Peterson - 2005-05-17 14:17:09 ```Steffen Petersen writes: > > Hi Ondrej, > > I have just checked example 8. Everything works fine for me, > i.e. I could not reproduce the unstable behavior you describe. > Find attached a plot of the results at node 274 you should get > when using the default values. Would you be using LasPack by any chance while Steffen is using PETSc? -John ```
 Re: AW: [Libmesh-users] Re: wave equation and newmark scheme From: Roy Stogner - 2005-05-17 14:36:53 ```On Tue, 17 May 2005, John Peterson wrote: > Steffen Petersen writes: > > > > Hi Ondrej, > > > > I have just checked example 8. Everything works fine for me, > > i.e. I could not reproduce the unstable behavior you describe. > > Find attached a plot of the results at node 274 you should get > > when using the default values. > > Would you be using LasPack by any chance while Steffen is using > PETSc? Is there something outright broken in the laspack interface? I'd avoid laspack anyway because of the lack of some of PETSc's better algorithms, but more importantly I've seen the same simulation converge with default PETSc settings and fail badly with default laspack settings - which I thought were GMRES/ILU0 in both cases. --- Roy ```
 Re: [Libmesh-users] Re: wave equation and newmark scheme From: Ondrej Certik - 2005-05-17 15:46:21 ```On Tue, May 17, 2005 at 09:16:52AM -0500, John Peterson wrote: > Steffen Petersen writes: > > > > Hi Ondrej, > > > > I have just checked example 8. Everything works fine for me, > > i.e. I could not reproduce the unstable behavior you describe. > > Find attached a plot of the results at node 274 you should get > > when using the default values. > > Would you be using LasPack by any chance while Steffen is using > PETSc? Yes, I am using LasPack. I try petsc (I have a problem with large meshes on my debian, as was already reported on this list, so I must recompile it from source). If it works, there must be a bug in LasPack. ```
 Re: [Libmesh-users] Re: wave equation and newmark scheme From: Ondrej Certik - 2005-05-17 21:11:33 ```Thanks for the graph - I tried petsc library and got the same graph, so the problem is definitely in the laspack library. It's a good feeling that all my other programs suddenly works (with petsc). :) As a bonus I created some notes about libmesh and fem (primarily for me), but maybe others find it useful as well. There is some documentation about the newmark class (what equations it solves etc.), which may save somebody's time. http://ondrej.certik.cz/libmesh/fem.ps http://ondrej.certik.cz/libmesh/fem.tex You should be able to reproduce the problematic behaviour by running: ./ex8 pipe-mesh.unv --use-laspack (At least on my computer) Ondrej On Tue, May 17, 2005 at 03:48:10PM +0200, Steffen Petersen wrote: > > Hi Ondrej, > > I have just checked example 8. Everything works fine for me, > i.e. I could not reproduce the unstable behavior you describe. > Find attached a plot of the results at node 274 you should get > when using the default values. > > Steffen > > > > > The solution in example 8 (wave equation) diverges on my > > computer: when I look at the pressure at the node 274 in > > "pressure_node.res", I got numbers like 1E80. > > > > When I make this change to the ex8.C: > > > > t_system.set_newmark_parameters(delta_t,1.0,2.0); > > > > (I set the alpha=1.0, delta=2.0), the solution seems to be ok > > (the pressure at the node 274 is less then 1.0 in absolute > > value, as should > > be) and the wave propagates through the pipe. > > > > I use CVS version of libmesh. Is the problem only on my > > computer, or is there a bug somewhere? I got the same > > behaviour in all my programs using the Newmark scheme (I > > tried both the wave and heat equation). > > > > Ondrej Certik > > > > > > ------------------------------------------------------- > > This SF.Net email is sponsored by Oracle Space Sweepstakes > > Want to be the first software developer in space? > > Enter now for the Oracle Space Sweepstakes! > > http://ads.osdn.com/?ad_id=7412&alloc_id=16344&op=click > > _______________________________________________ > > Libmesh-users mailing list > > Libmesh-users@... > > https://lists.sourceforge.net/lists/listinfo/libmesh-users > > ```
 AW: [Libmesh-users] Re: wave equation and newmark scheme From: Steffen Petersen - 2005-05-18 06:45:19 ```> Thanks for the graph - I tried petsc library and got the same > graph, so the problem is definitely in the laspack library. > > It's a good feeling that all my other programs suddenly works > (with petsc). :) As a bonus I created some notes about > libmesh and fem (primarily for me), but maybe others find it > useful as well. There is some documentation about the newmark > class (what equations it solves etc.), which may save somebody's time. > > http://ondrej.certik.cz/libmesh/fem.ps > http://ondrej.certik.cz/libmesh/fem.tex > > You should be able to reproduce the problematic behaviour by running: > > ./ex8 pipe-mesh.unv --use-laspack > Yep, I have already noticed that laspack seems to be the problem (unfortunately, I could not find the reason for the problem right away). At this, all the laspack solvers (not only the defaults GMRES/ILU) lead to the instabilities you have mentioned. Regarding the libMesh examples, ex8 seems to be the only example not working with laspack, but Roy mentioned that he has noticed similar behavior for other applications. Steffen ```
 AW: [Libmesh-users] Re: wave equation and newmark scheme From: John Peterson - 2005-05-18 13:07:39 ```Steffen Petersen writes: > > Thanks for the graph - I tried petsc library and got the same > > graph, so the problem is definitely in the laspack library. > > > > It's a good feeling that all my other programs suddenly works > > (with petsc). :) As a bonus I created some notes about > > libmesh and fem (primarily for me), but maybe others find it > > useful as well. There is some documentation about the newmark > > class (what equations it solves etc.), which may save somebody's time. > > > > http://ondrej.certik.cz/libmesh/fem.ps > > http://ondrej.certik.cz/libmesh/fem.tex > > > > You should be able to reproduce the problematic behaviour by running: > > > > ./ex8 pipe-mesh.unv --use-laspack > > > > Yep, I have already noticed that laspack seems to be the problem > (unfortunately, I could not find the reason for the problem right away). > At this, all the laspack solvers (not only the defaults GMRES/ILU) > lead to the instabilities you have mentioned. Regarding the libMesh > examples, ex8 seems to be the only example not working with laspack, > but Roy mentioned that he has noticed similar behavior for other > applications. I think I will put a line in example 8 that prints a big warning if you are using LasPack solvers...otherwise we will have this same conversation a few months from now :) -John ```