- Mar 17, 2021
-
-
Christian Heinigk authored
-
- Jan 27, 2020
-
-
Christian Heinigk authored
-
Christian Heinigk authored
-
- Jan 08, 2020
-
-
Christian Heinigk authored
-
Christian Heinigk authored
-
- Jan 07, 2020
-
-
René Heß authored
Merge branch 'feature/jacobian-apply-in-onestep-grid-operator' into 'master' ref:pdelab/dune-pdelab This adds the jacobian apply method to the one step grid operator. See merge request [pdelab/dune-pdelab!444] [pdelab/dune-pdelab!444]: gitlab.dune-project.org/pdelab/dune-pdelab/merge_requests/444
-
- Dec 19, 2019
-
-
Dominic Kempf authored
Merge branch 'feature/vtkwriter-lifetime-issues' into 'master' ref:pdelab/dune-pdelab This commit introduces an additional signature for addSolutionToVTKWriter that takes grid function space and data vector as shared pointers. This is necessary because of the delayed nature of VTK export: Data sets are registered in a first step and written in a second step. Currently PDELab implicitly requires both steps to happen in the same scope, as otherwise the registered function objects can be destroyed before written into a file. The additional signature allows to pass shared pointer objects that the VTK functions take ownership of. Internally, this required the BlockVector of the ISTL vector backend to store the grid function space as a shared pointer. See merge request [pdelab/dune-pdelab!434] [pdelab/dune-pdelab!434]: gitlab.dune-project.org/pdelab/dune-pdelab/merge_requests/434
-
Dominic Kempf authored
Merge branch 'feature/construct-composite-gridfunction-from-shared-ptr' into 'master' See merge request [pdelab/dune-pdelab!443] [pdelab/dune-pdelab!443]: Nonepdelab/dune-pdelab/merge_requests/443
-
- Dec 18, 2019
-
-
René Heß authored
Note: The solution might be unvalid, if you have a linear problem it never gets set!
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
Merge branch 'feature/jacobian-apply-cleanup' into 'master' ref:pdelab/dune-pdelab This merge request cleans up the existing jacobian apply engines. We remove the NonlinearJacobianApply engine and instead make sure that the JacobianApply engine can handle the linear and the non-linear case. This is done by introducing a flag isLinear to the local operators through <dune/pdelab/localoperator/flags.hh>. The JacobianApply uses this flag to decide which methods need to be called. On the GridOperator side we have two methods called jacobian_apply that take a different number of arguments for the linear and non-linear case. The method nonlinear_jacobian_apply still exists but is deprecated. If you use the wrong method you will get an error. Changes for users: If you have a non-linear residual and want to use the nonlinear_jacobian_apply method you need to set the isLinear flag in your local operator. This merge request is based on inspired by the work of @smuething and @marian in the EXADune project, checked out on commit 70c0f644dee360259b36cb17888e6baf90da5056. See merge request [pdelab/dune-pdelab!439] [pdelab/dune-pdelab!439]: gitlab.dune-project.org/pdelab/dune-pdelab/merge_requests/439
-
- Dec 16, 2019
-
-
René Heß authored
Merge branch 'feature/keep-old-newton-backward-compatible' into 'master' ref:pdelab/dune-pdelab This way it should be completely backward compatible (to the time before the new Newton was introduced) See merge request [pdelab/dune-pdelab!445] [pdelab/dune-pdelab!445]: gitlab.dune-project.org/pdelab/dune-pdelab/merge_requests/445
-
- Dec 13, 2019
-
-
René Heß authored
This way it should be completely backward compatible (to the time before the new Newton was introduced)
-
René Heß authored
Merge branch 'feature/vtkwriter-lifetime-issues' of ssh://gitlab.dune-project.org:22022/pdelab/dune-pdelab into feature/vtkwriter-lifetime-issues
-
René Heß authored
-
- Dec 12, 2019
-
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
Note: For some reason the Jacobian method does not work. This is most likely an artifact that comes from the fact that this test never assembles anything. For real problems everything seems to work fine.
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
- Dec 11, 2019
-
-
René Heß authored
-
Dominic Kempf authored
Merge branch 'feature/grid-function-probe-initialization' into 'master' ref:pdelab/dune-pdelab There is a setter for the function anyway, so it is natural to be able to construct the probe object without a function. See merge request [pdelab/dune-pdelab!438] [pdelab/dune-pdelab!438]: gitlab.dune-project.org/pdelab/dune-pdelab/merge_requests/438
-
Dominic Kempf authored
-
Dominic Kempf authored
This bug originates from a C&P bug. It never worked, but nobody notices.
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
-
René Heß authored
Merge branch 'cleanup/internal-renaming-in-engine' into 'master' ref:pdelab/dune-pdelab Before really starting [!439] it would make sense to rename some things in the engines. The proposal is as follows: - The engines have methods and members corresponding to data vectors. Current naming is like this: -------------------------- ---------------------------- residual Residual, Solution jacobian Jacobian, Solution jacobian_apply Residual, Solution nonlinear_jacobian_apply Residual, Solution, Update -------------------------- ---------------------------- - The proposal is to change those names to: -------------------------- -------------------------- residual Residual, Solution jacobian Jacobian, Solution jacobian_apply Result, Update nonlinear_jacobian_apply Result, Update, Solution -------------------------- -------------------------- The changes would be: Solution-\>Update in the linear case of jacobian_apply and Solution-\>Update, Update-\>Solution in the nonlinear case of jacobian_apply. For me it would also be fine to always call the result 'Result' and not 'Residual' or 'Jacobian'. - Why renaming? Residual form: r(u,v)=a(u,v)-b Residual vector: (R(z))_i = r(\sum_j(z)_j\phi_j, \phi_i) Jacobian: \nabla R(z) Jacobian Apply: \nabla R(z) w If you look at the jacobian and the jacobian_apply the names of the objects should be different since they have different mathematical meaning. If it stays as it is, it is really confusing in the nonlinear case since the Solution at the moment corresponds to w and the update to z and this doesn't make any sense at all. After renaming the name update actually makes sense. If you think about a Newton solver or a linear solver in residual formulation you actually evaluate the jacobian for a given solution and try to solve for an update. So calling z Solution and w update seems reasonable. See merge request [pdelab/dune-pdelab!440] [!439]: gitlab.dune-project.org/NoneNone/merge_requests/439 [pdelab/dune-pdelab!440]: gitlab.dune-project.org/pdelab/dune-pdelab/merge_requests/440
-
- Dec 10, 2019
-
-
René Heß authored
Merge branch 'feature/newton-improve-documentation' into 'master' See merge request [pdelab/dune-pdelab!441] [pdelab/dune-pdelab!441]: Nonepdelab/dune-pdelab/merge_requests/441
-