- firedrake - imperial.ac.uk

firedrake-install error
by Anna Kalogirou 23 Oct '15

23 Oct '15

Dear all, I tried to update firedrake on my Macbook using the firedrake-install script for the first time, but I got the error below. I can update successfully using the standard steps. Best, Anna. ---------------------------------------------------------- Error: No such keg: /usr/local/Cellar/hdf5 Error: undefined method `desc' for Hdf5:Class Please report this bug: http://git.io/brew-troubleshooting /usr/local/Library/Taps/homebrew/homebrew-science/hdf5.rb:2:in `<class:Hdf5>' /usr/local/Library/Taps/homebrew/homebrew-science/hdf5.rb:1:in `<top (required)>' /System/Library/Frameworks/Ruby.framework/Versions/2.0/usr/lib/ruby/2.0.0/rubygems/core_ext/kernel_require.rb:55:in `require' /System/Library/Frameworks/Ruby.framework/Versions/2.0/usr/lib/ruby/2.0.0/rubygems/core_ext/kernel_require.rb:55:in `require' /usr/local/Library/Homebrew/formulary.rb:65:in `klass' /usr/local/Library/Homebrew/formulary.rb:50:in `get_formula' /usr/local/Library/Homebrew/formulary.rb:198:in `factory' /usr/local/Library/Homebrew/extend/ARGV.rb:16:in `block in formulae' /usr/local/Library/Homebrew/extend/ARGV.rb:16:in `map' /usr/local/Library/Homebrew/extend/ARGV.rb:16:in `formulae' /usr/local/Library/Homebrew/cmd/install.rb:30:in `install' /usr/local/Library/brew.rb:136:in `<main>' Traceback (most recent call last): File "firedrake-install", line 340, in <module> brew_install("hdf5") File "firedrake-install", line 180, in brew_install check_call(["brew", "install", name]) File "firedrake-install", line 163, in check_call subprocess.check_call(arguments) File "/usr/local/Cellar/python/2.7.8_2/Frameworks/Python.framework/Versions/2.7/lib/python2.7/subprocess.py", line 540, in check_call raise CalledProcessError(retcode, cmd) subprocess.CalledProcessError: Command '['brew', 'install', 'hdf5']' returned non-zero exit status 1

2 2

Re: [firedrake] Issues with Firedrake
by David Ham 22 Oct '15

22 Oct '15

Dear William, Thanks for your email, I'm shifting this conversation onto the Firedrake mailing list, which is the appropriate place for Firedrake help requests. Please also see my responses below. On Thu, 15 Oct 2015 at 10:10 William Booker <scwb(a)leeds.ac.uk> wrote: > Dear David, > > I've been trying to get to grips with Firedrake by modifying the advection > demo. > > I'm trying to implement it as unsteady with a backwards euler timestep but > I've run into a few problems. > > First is that my initial condition does not appear to be implemented > correctly, as shown in the graph. It seems that instead of a shock, the IC > is smoothed out throughout the domain. > I've run your code, and the first thing to realise is that your first output is after the first timestep. The image you show is therefore after the tracer has begun to advect. This accounts for the "smearing out" and as far as I can see, the solution is entirely consistent with what you have specified. > > Second is that I'm not sure that my boundary conditions are actually being > set correctly, that is if I try to set the IC to be zero everywhere, I'm > not getting a value of 1 rising on the right side as I would expect. > I'm not sure why you expect this because the velocity field you have specified flows in the z direction, and the boundary condition is non-zero over the right half of the x-y (z=0) boundary. The advection equation therefore causes tracer to flux vertically through the right half of the domain until it is uniformly 1 and the left side is uniformly 0. This is what happens in the simulation, as far as I can see it's all correct. There is one small issue you may not have noticed, which is that the tracer is taking the value 0.5 when you might have expected 1.0. This is due to a subtle rounding error. When you project your tracer field, it is first being interpolated into DG1. In DG1, the top nodes of the bottom cell are (for rounding reasons) taking the value 0, which means the cell average value is 0.5. In this case, the simplest way around this is to interpolate the BC value straight into DG0. That is, replace: ic = project(inflow, V) with: ic = interpolate(inflow, V) > Finally I've tried to use a mesh made in gmsh but I've found if I don't > name the boundaries, I can't assign values using the 1,2,3,4 system. This > may be related to the previous point but if I try to assign names in gmsh, > Firedrake won't load the file with an error that it is not a valid gmsh > file. > > Firedrake is complaining that it's not a valid gmsh file because it isn't a valid gmsh file. If you check lines 8 and 9 of trapezium.msh you will see some junk which is directly quoted from the .geo file. There is clearly some issue with your .geo file. If I had to take a guess, I would say that there is a missing double quote character after Right. Regards, David > > Any help would be greatly appreciated, > > Will >

2 1

Re: [firedrake] Need guidance regarding firedrake (wave equation)
by Colin Cotter 22 Oct '15

22 Oct '15

Hi Chaitanya, Any of these timestepping schemes can be implemented in Firedrake. As for which is best, this should be determined from the properties of those schemes (perhaps it is your homework?). All the best Cjc On 22 Oct 2015 01:15, "Chaitanya Raj Goyal" <chaitanyaraj108(a)gmail.com> wrote: > Hi all, > > I am new to Fenics and recently came across Firedrake. I am trying to > solve a 2D seismic wave equation for multiple layers of soil with varying > properties with absorbing boundary conditions. > > 1. I was wondering that between the Sympletic Euler, Newmark Beta and > Crank Nicolson, (or any other) which is the best method to discretise the > wave equation {u_tt = c^2 \Delta u + f} for implementation in firedrake. > Could you please share your opinion. > > 2. If you happen to be aware of a specific good document that derives such > a time stepping scheme for wave equation, could you kindly point towards > it. > > 3. Also, would firedrake be a better platform to implement the problem > instead of fenics? > > I apologize if the questions may seem trivial, but I just started my > Masters in Civil Engineering, and it would be awesome to get first hand > opinion from math modelling experts. > > I' ll be very grateful if you could guide me on the above issues! > > Thank you, > > Sincerely yours, > Chaitanya Goyal > > > > >

1 0

Need guidance regarding firedrake (wave equation)
by Chaitanya Raj Goyal 22 Oct '15

22 Oct '15

Hi all, I am new to Fenics and recently came across Firedrake. I am trying to solve a 2D seismic wave equation for multiple layers of soil with varying properties with absorbing boundary conditions. 1. I was wondering that between the Sympletic Euler, Newmark Beta and Crank Nicolson, (or any other) which is the best method to discretise the wave equation {u_tt = c^2 \Delta u + f} for implementation in firedrake. Could you please share your opinion. 2. If you happen to be aware of a specific good document that derives such a time stepping scheme for wave equation, could you kindly point towards it. 3. Also, would firedrake be a better platform to implement the problem instead of fenics? I apologize if the questions may seem trivial, but I just started my Masters in Civil Engineering, and it would be awesome to get first hand opinion from math modelling experts. I' ll be very grateful if you could guide me on the above issues! Thank you, Sincerely yours, Chaitanya Goyal

1 0

Re: [firedrake] Faster way to do this?
by David Ham 21 Oct '15

21 Oct '15

Hi Justin, I'm kind of surprised that the vector access is the slow bit here (as opposed to the solve), but yes, you can make this much faster. There are two ways of getting access to a numpy array containing the DoF values. One is: u.vector().array() the other is: u.dat.data (or u.dat.data_ro if you want to tell PyOP2 you're not going to change the data). The two are essentially equivalent for scalar Functions. For functions in a VectorFunctionSpace the dat version returns a 2D array while vector() is flattened to 1D. Regards, David On Tue, 20 Oct 2015 at 23:07 Justin Chang <jychang48(a)gmail.com> wrote: > Hi all, > > I am currently writing my own point-wise petsc4py solver which needs data > that lives in u = Function(Q), where Q is of type CG1 (i.e., a P1 element, > same as the coordinates). > > Right now I am doing something like this: > > for i in range(mesh.num_vertices()): > FormInitialGuess(u.vector()[i]) > Solve(...) > ... > > but it seems the u.vector()[i] operation is extremely slow. Is there a > faster or more efficient way of calling or extracting pointwise values? > > Thanks, > Justin >

1 0

Faster way to do this?
by Justin Chang 20 Oct '15

20 Oct '15

Hi all, I am currently writing my own point-wise petsc4py solver which needs data that lives in u = Function(Q), where Q is of type CG1 (i.e., a P1 element, same as the coordinates). Right now I am doing something like this: for i in range(mesh.num_vertices()): FormInitialGuess(u.vector()[i]) Solve(...) ... but it seems the u.vector()[i] operation is extremely slow. Is there a faster or more efficient way of calling or extracting pointwise values? Thanks, Justin

1 0

Re: [firedrake] Matrix multiplication in bilinear form
by Colin Cotter 19 Oct '15

19 Oct '15

Yes, Lawrence was writing what you wrote, but in quantum physics notation. You can't take the physics out of the boy. --cjc On 19 October 2015 at 19:43, Anna Kalogirou <a.kalogirou(a)leeds.ac.uk> wrote: > I am not familiar with this expression |Q1> <Q2|. What I really want is > Q1*Q2^T, where Q1 and Q2 and vectors defined in my previous email. > > Actually I think it's easier if I just provide you with the actual > equation I need to solve, please see attached. > > Anna. > > > On 19/10/15 14:15, Lawrence Mitchell wrote: > > -----BEGIN PGP SIGNED MESSAGE----- > > Hash: SHA1 > > > > On 19/10/15 14:00, Anna Kalogirou wrote: > >> Dear Lawrence, > >> > >> Yes, I have a term where the test function and the basis function > >> are evaluated in separate integrals, this is why I wrote Q*mu*dx. > >> Can I assemble them separately? E.g. > >> > >> Q1 = assemble(v*dx) Q2 = assemble(mu*dx) > > The resulting values aren't unknowns (either test or trial functions) > > so I still don't quite understand. > > > > Q1 is a Co-Function (living in the dual space of V) and Q2 is a > > Function in V. You can certainly point-wise multiply the values > > together, but I don't think that's what you want, since that still > > doesn't give you an operator mapping from the trial to the test space. > > > > Do you want the rank-1 operator formed by: > > > > |Q1> <Q2| > > > > ? > > > > Cheers, > > > > Lawrence > > -----BEGIN PGP SIGNATURE----- > > Version: GnuPG v2.0.22 (GNU/Linux) > > > > iQEcBAEBAgAGBQJWJO0GAAoJECOc1kQ8PEYvtfIIAOsepbX3x0TCCrJYl3PXr3Vz > > 2AckYTLSiiUgYYFQ5IPhEOXd4yoDlBtzA9DIo6oyotwa9JltE0KpP45ZK/b0oaAO > > iAB6s73pcNKkZ328GoiT6/Jp5zu5pCQI+2X2Zhjj6EmYzEpu9J5oAGrLSKHy0nbC > > Z6gyvG72LFDHcTgNauStl+1yC2D6bTD2JOk9e5ph5ySdLbXL1kegTBt+RHYhfdI3 > > TsiHLaPZzBp7rRUQ4EGq6pDikS+qTU3IEhKz1onEuWOnHqwHCO14zaRSsiFDBDS9 > > 5ALN9kiqz1rW8XHykWrhZec0MZhxjqZ6akDbCUmdMlwdXje6nYTaYjpQrTgvyko= > > =Vvc4 > > -----END PGP SIGNATURE----- > > > > _______________________________________________ > > firedrake mailing list > > firedrake(a)imperial.ac.uk > > https://mailman.ic.ac.uk/mailman/listinfo/firedrake > > -- > > Dr Anna Kalogirou > Research Fellow > School of Mathematics > University of Leeds > > http://www1.maths.leeds.ac.uk/~matak/ > > -- http://www.imperial.ac.uk/people/colin.cotter www.cambridge.org/9781107663916

1 0

Re: [firedrake] functionspace.bt_masks: 'topological' vs 'geometric'
by Colin Cotter 19 Oct '15

19 Oct '15

A note: In terms of stability of operators, geometric boundary conditions are almost always the Wrong Thing. cheers --cjc On 19 October 2015 at 15:59, Eike Mueller <E.Mueller(a)bath.ac.uk> wrote: > Hi Miklos, > > thanks a lot for the detailled explanation, I’m much more confident that > I’m doing the correct thing now. > > Cheers, > > Eike > > > On 18 Oct 2015, at 15:16, Homolya, Miklós <m.homolya14(a)imperial.ac.uk> > wrote: > > > > Hello Eike, > > > > I don't know how you use them, but here's what topological and geometric > mean for boundary conditions: > > > > Topological boundary conditions set those degrees of freedom, which > topologically lie on the facet (more precisely, on the closure of the > facet). Geometric boundary conditions set those DoFs, which don't vanish on > the boundary. In the DGn, n >= 1 case they are the DoFs which are > geometrically on the boundary, hence the name. > > > > A few examples: > > - For CG, they mean the same thing. > > - For DG, all nodes are topologically associated with the cell interior, > so topological does nothing. Geometric sets the nodes geometrically on the > facet, except for DG0, which sets the only node (which happens to be in the > cell centre). > > - For semi-continuous function spaces, they mean different things. For > example, for RT, a topological BC sets the normal component on the facet > (or the tagential component in case of an H(curl) conforming space, > accordingly). Geometric BCs will set many of the nodes to make sure to fix > all components of the value on the boundary. (I'm not sure if you want to > do that.) > > > > I hope this helps. Topological is the only option that was supported on > extruded until now, so that must have been the one you were using. > > > > Regards, > > Miklos > > > > ________________________________________ > > From: firedrake-bounces(a)imperial.ac.uk [firedrake-bounces(a)imperial.ac.uk] > on behalf of Eike Mueller [E.Mueller(a)bath.ac.uk] > > Sent: 18 October 2015 11:36 > > To: firedrake > > Subject: [firedrake] functionspace.bt_masks: 'topological' vs 'geometric' > > > > Dear firedrake, > > > > While bringing my code up-to-date with the latest changes in > firedrake/PyOP2, I’m trying to understand the changes to > functionspace.bt_masks. In the past I just used > > > > fs.bt_masks[0] and fs.bt_masks[1] to get the top/bottom boundary nodes > of a function space. With commits > > > > a2c1add18f00eca24789f209632411f30fffba4b > > > > ba340ceb92dc1a156a73eb2ef8e60427a67fa0e7 > > > > the tuples seems to have been replaced by a dictionary, i.e. there are > now ‘toplogical’ and ‘geometric’ boundary nodes. What’s the difference > between the two? > > > > In my code, I replaced fs.bt_masks[i] -> fs.bt_masks[’topological'][i], > but I want to make sure that I’m doing the right thing. > > > > Thanks, > > > > Eike > > > > _______________________________________________ > > firedrake mailing list > > firedrake(a)imperial.ac.uk > > https://mailman.ic.ac.uk/mailman/listinfo/firedrake > > _______________________________________________ > > firedrake mailing list > > firedrake(a)imperial.ac.uk > > https://mailman.ic.ac.uk/mailman/listinfo/firedrake > > > _______________________________________________ > firedrake mailing list > firedrake(a)imperial.ac.uk > https://mailman.ic.ac.uk/mailman/listinfo/firedrake > -- http://www.imperial.ac.uk/people/colin.cotter www.cambridge.org/9781107663916

1 0

functionspace.bt_masks: 'topological' vs 'geometric'
by Eike Mueller 19 Oct '15

19 Oct '15

Dear firedrake, While bringing my code up-to-date with the latest changes in firedrake/PyOP2, I’m trying to understand the changes to functionspace.bt_masks. In the past I just used fs.bt_masks[0] and fs.bt_masks[1] to get the top/bottom boundary nodes of a function space. With commits a2c1add18f00eca24789f209632411f30fffba4b ba340ceb92dc1a156a73eb2ef8e60427a67fa0e7 the tuples seems to have been replaced by a dictionary, i.e. there are now ‘toplogical’ and ‘geometric’ boundary nodes. What’s the difference between the two? In my code, I replaced fs.bt_masks[i] -> fs.bt_masks[’topological'][i], but I want to make sure that I’m doing the right thing. Thanks, Eike

2 2

Re: [firedrake] problems with upgrading PETSc and firedrake
by Andrew McRae 19 Oct '15

19 Oct '15

Did you run 'make' in firedrake (can try 'make clean' first, too), and 'make ext' in PyOP2? If that doesn't work, try COFFEE, FFC, UFL, FIAT. I believe the output shown doesn't necessarily mean the segfault occurred during a PETSc operation; if a segfault occurs anywhere in the program (executing a kernel during assembly, say) then I believe you get the same output. On 17 October 2015 at 23:21, Eike Mueller <E.Mueller(a)bath.ac.uk> wrote: > Dear firedrake, > > I just updated petsc and petsc4py (both from mapdes, using the ‘firedrake’ > branch), but now whenever I try to create a mesh > > from firedrake import * > mesh = UnitIcosahedralSphereMesh(refinement_level=2) > > I just get a PETSc error (see below). I also upgraded both PyOP2 and > firedrake (both on ‘master’), and ran firedrake-clean, but that doesn’t > help. > > Thanks, > > Eike > > eikemueller@Eikes-MBP $ python test.py > [0]PETSC ERROR: > ------------------------------------------------------------------------ > [0]PETSC ERROR: Caught signal number 11 SEGV: Segmentation Violation, > probably memory access out of range > [0]PETSC ERROR: Try option -start_in_debugger or -on_error_attach_debugger > [0]PETSC ERROR: or see > http://www.mcs.anl.gov/petsc/documentation/faq.html#valgrind > [0]PETSC ERROR: or try http://valgrind.org on GNU/linux and Apple Mac OS > X to find memory corruption errors > [0]PETSC ERROR: configure using --with-debugging=yes, recompile, link, and > run > [0]PETSC ERROR: to get more information on the crash. > -------------------------------------------------------------------------- > MPI_ABORT was invoked on rank 0 in communicator MPI_COMM_WORLD > with errorcode 59. > > NOTE: invoking MPI_ABORT causes Open MPI to kill all MPI processes. > You may or may not see output from other processes, depending on > exactly when Open MPI kills them. > -------------------------------------------------------------------------- > > _______________________________________________ > firedrake mailing list > firedrake(a)imperial.ac.uk > https://mailman.ic.ac.uk/mailman/listinfo/firedrake >

4 3