Add grad_zero for phi and test function #15205

arovinelli · 2020-05-01T23:23:22Z

Just add two constants gradient variable

@lindsayad as per mailing list discussion

arovinelli · 2020-05-01T23:24:12Z

framework/src/problems/FEProblemBase.C

+        _grad_phi_zero[tid].resize(max_qpts, {0., 0., 0.});
+        _grad_test_zero[tid].resize(max_qpts, {0., 0., 0.});


@lindsayad this init here is ugly but I can't figure out how to init with RealVectorValue...
suggestions?

lindsayad · 2020-05-02T00:18:21Z

I need to see a test to believe this is working. For one thing the outer size should be according to the number of shape functions, not the number of q-points. Perhaps you are getting away with this because for first-order Lagrange those numbers are the same in the element volume? Also my current reading of the {0,0,0} initializer list is that its going to create three RealVectorValues with each component initialized to zero. That length of three doesn't seem like it's going to match an arbitrary number of quadrature points. What are you testing this on?

moosebuild · 2020-05-02T01:03:07Z

Job Documentation on e63ae8d wanted to post the following:

View the site here

This comment will be updated on new commits.

arovinelli · 2020-05-02T01:18:49Z

I need to see a test to believe this is working. For one thing the outer size should be according to the number of shape functions, not the number of q-points. Perhaps you are getting away with this because for first-order Lagrange those numbers are the same in the element volume? Also my current reading of the {0,0,0} initializer list is that its going to create three RealVectorValues with each component initialized to zero. That length of three doesn't seem like it's going to match an arbitrary number of quadrature points. What are you testing this on?

wait, that is not how the already available _grad_zero works. If I understand the avaialble code correctly, it is simply a _grad_zero is simply a RealVectorValue(0) with a size equal to the number of _qp such that you can do something like in this example
So you just need _grad_phi_zero to be a constant reference to a RealVector.Am I wrong?

I'm not testing it right now, try to make a kernel or material for this

lindsayad · 2020-05-02T01:23:48Z

A VariableGradient is a MooseArray<RealVectorValue>. A VariablePhiGradient is a MooseArray<std::vector<RealVectorValue>. So the latter takes two indexes, the first for dofs, the second for quadrature points. The former only takes a single index for quadrature points.

arovinelli · 2020-05-02T01:41:40Z

A VariableGradient is a MooseArray<RealVectorValue>. A VariablePhiGradient is a MooseArray<std::vector<RealVectorValue>. So the latter takes two indexes, the first for dofs, the second for quadrature points. The former only takes a single index for quadrature points.

This makes a lot of sense and explains why I was not able to find the right initialization for it.
So it should something like

    _grad_test_zero[tid].resize(getMaxShapeFunctions(),
                                std::vector<RealGradient>(max_qpts, RealGradient(0.)));

lindsayad · 2020-05-03T23:55:00Z

Correct. You might actually want to look at the code around this line: https://github.com/idaholab/moose/blob/next/framework/src/problems/FEProblemBase.C#L627. We actually compute exactly what you're looking for there

moosebuild · 2020-05-12T19:00:03Z

Job Modules debug PETSc alt on 2d2630c : invalidated by @arovinelli

uhm... this should be related to this pr

arovinelli · 2020-05-12T19:06:51Z

@lindsay I'm also trying to construct a VariapliPhiValue _phi_zero. What type is and what size should a VariapliPhiValue ?

lindsayad · 2020-05-12T19:35:01Z

A VariablePhiValue is a MooseArray<std::vector<Real>.

arovinelli · 2020-05-13T18:33:26Z

@lindsay it isn't as straightforward as I thought.
first, the number of DOF might be different between nodes and elements _nl->getMaxVarNDofsPerElem _nl->getMaxVarNDofsPerNode so I have to choose which of this to use. Also when called in FEProbelmBase::createQRules they both return zero, therefore I assume FEProbelmBase::intialSetup has not been called yet.

I tried to use the code you indicated , inside FEProbelmBase::createQRules but it gives me a segfault

    MaxVarNDofsPerElem mvndpe(*this, *_nl);
    Threads::parallel_reduce(*_mesh.getActiveLocalElementRange(), mvndpe);
    auto max_var_n_dofs_per_elem = mvndpe.max();
    _communicator.max(max_var_n_dofs_per_elem);

here is the stack trace

#0  0x00007fffe8596a22 in __gnu_debug::_Error_formatter::_Parameter::_M_print_field(__gnu_debug::_Error_formatter const*, char const*) const () from /lib64/libstdc++.so.6
#1  0x00007fffe8596cfb in __gnu_debug::_Error_formatter::_Parameter::_M_print_description(__gnu_debug::_Error_formatter const*) const () from /lib64/libstdc++.so.6
#2  0x00007fffe859738f in __gnu_debug::_Error_formatter::_M_error() const () from /lib64/libstdc++.so.6
#3  0x00007ffff743ca0b in std::__debug::vector<unsigned int, std::allocator<unsigned int> >::operator[] (this=0x892f10, __n=0)
    at /opt/rh/devtoolset-7/root/usr/include/c++/7/debug/vector:424
#4  0x00007fffeef8a9e5 in libMesh::DofMap::dof_indices (this=0x892e80, elem=0x85b650, 
    di=std::__debug::vector of length 0, capacity 0, vn=0, p_level=0)
    at /home/arovinelli/projects/NEW_stuff/moose/scripts/../libmesh/src/base/dof_map.C:2093
#5  0x00007ffff5599308 in MaxVarNDofsPerElem::onElement (this=0x7fffffffb930, elem=0x85b650)
    at /home/arovinelli/projects/NEW_stuff/moose/framework/src/base/MaxVarNDofsPerElem.C:39
#6  0x00007ffff47ba86b in ThreadedElementLoopBase<libMesh::StoredRange<libMesh::MeshBase::const_element_iterator, libMesh::Elem const*> >::operator() (this=0x7fffffffb930, range=..., bypass_threading=false)
    at /home/arovinelli/projects/NEW_stuff/moose/framework/build/header_symlinks/ThreadedElementLoopBase.h:209
#7  0x00007ffff4d4b073 in libMesh::Threads::parallel_reduce<libMesh::StoredRange<libMesh::MeshBase::const_element_iterator, libMesh::Elem const*>, MaxVarNDofsPerElem> (range=..., body=...)
    at /home/arovinelli/projects/NEW_stuff/moose/scripts/../libmesh/installed/include/libmesh/threads_pthread.h:380
#8  0x00007ffff4d21b23 in FEProblemBase::createQRules (this=0x8685a0, type=libMesh::QGAUSS, order=libMesh::THIRD, 
    volume_order=libMesh::THIRD, face_order=libMesh::THIRD)
    at /home/arovinelli/projects/NEW_stuff/moose/framework/src/problems/FEProblemBase.C:4580
#9  0x00007ffff4c676a2 in SetupQuadratureAction::act (this=0x81fc90)
    at /home/arovinelli/projects/NEW_stuff/moose/framework/src/actions/SetupQuadratureAction.C:51

so I can't compute either the maxNdof per elem or node

lindsayad · 2020-05-13T18:45:50Z

It might be helpful for you to look at the order of actions in Moose.C, starting around line 245. Quadrature setup happens well before the FEProblemBase is initialized (the init_problem task calls FEProblemBase::init), which is where the libMesh EquationSystems and DofMap objects are initialized, which is where dof indices and things will get setup.

Why don't you wait to do your zeroing and setup until FEProblemBase::initialSetup which is where we already do the MaxVarNDofsPerElem calculation? You can choose to make that calculation non-optional if you want (we currently only do it if there are AD objects in the problem)

arovinelli · 2020-05-13T20:20:04Z

@lindsay, thanks a lot for the suggestion I've embraced it and I'm making good progress.
At this point, I believe the size of _second_phi_zero wrong and also initialized in the wrong place, could you check (see link)

moose/framework/src/problems/FEProblemBase.C

Lines 1612 to 1613 in 544d67e

    
           _second_phi_zero[tid].resize( 
        
               max_qpts, std::vector<RealTensor>(getMaxShapeFunctions(), RealTensor(0.)));

lindsayad · 2020-05-13T22:51:40Z

Looking at MaxQpsThread::operator(), `getMaxShapeFunctions` will definitely return the wrong value if there are second order bases.

…

On Wed, May 13, 2020 at 1:20 PM Andrea Rovinelli ***@***.***> wrote: @lindsay <https://github.com/lindsay>, thanks a lot for the suggestion I've embraced it and I'm making good progress. At this point, I believe the size of _second_phi_zero wrong and also initialized in the wrong place, could you check (see link) https://github.com/idaholab/moose/blob/544d67eb7d02227ed071bd74b677aeb5674c8d24/framework/src/problems/FEProblemBase.C#L1612-L1613 — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#15205 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACOGA4FDE3WS4NAZICSN6S3RRL6IFANCNFSM4MXNV7GQ> .

moosebuild · 2020-05-14T13:15:03Z

Job Test debug on 2a4ef51 : invalidated by @arovinelli

these failures should not be related to this PR

arovinelli · 2020-05-14T13:50:04Z

So this

moose/framework/src/problems/FEProblemBase.C

Lines 1612 to 1613 in 544d67e

    
           _second_phi_zero[tid].resize( 
        
               max_qpts, std::vector<RealTensor>(getMaxShapeFunctions(), RealTensor(0.)));

is a preexisting bug.
This PR should fix it.

Looking at MaxQpsThread::operator(), getMaxShapeFunctions will definitely return the wrong value if there are second order bases.
…

ref idaholab#15204

arovinelli · 2020-05-14T17:30:38Z

test/src/kernels/PhiZeroKernel.C

+}
+
+PhiZeroKernel::PhiZeroKernel(const InputParameters & parameters)
+  : NullKernel(parameters), _second_phi(_assembly.secondPhi(_var))


@lindsayad Why I can't get _second_phi.size() > 0? In my tests, I use second-order element and second-order lagrangian function

Just calling secondPhi in Assembly is not enough to get MOOSE to do second derivatives. You need to actually request something like _var.secondSln() in order to get it to happen. A hacky way that may also work is to directly call _assembly.feSecondPhi<Real>(FEType(LAGRANGE,FIRST))

thanks, now it works!

moosebuild · 2020-05-14T20:22:05Z

Job Create Linux Stack on 667552f : invalidated by @arovinelli

stucked

arovinelli · 2020-05-14T21:58:39Z

@lindsayad when you have time, this is ready for review

lindsayad

A nice addition. thank you @arovinelli . I'm curious what your use case is?

Mostly fixup the test specs file

lindsayad · 2020-05-14T22:38:06Z

framework/src/problems/FEProblemBase.C

-  if (haveADObjects() || (_displaced_problem && _displaced_problem->haveADObjects()))
-  {
-    CONSOLE_TIMED_PRINT("Computing max dofs per elem/node");
+  // alwasy esecute to get the max number of DoF per elment and node needed to initialized phi_zero


always execute

lindsayad · 2020-05-14T22:45:48Z

test/include/kernels/PhiZeroKernel.h

+
+#include "NullKernel.h"
+
+class PhiZeroKernel;


No need for this forward declaration

lindsayad · 2020-05-14T22:46:11Z

test/include/kernels/PhiZeroKernel.h

+/**
+ *
+ */


Either add a doxygen comment or just delete this

lindsayad · 2020-05-14T22:49:45Z

test/src/kernels/PhiZeroKernel.C

+                "_phi.size() " + std::to_string(_phi.size()) + "> _phi_zero.size() " +
+                    std::to_string(_phi_zero.size()));


I understand that you were thinking about cases in which the "zero" members were of zero size, but your message should probably just say "!=" as opposed to ">". Same for all the following assertions

lindsayad · 2020-05-14T22:50:43Z

test/tests/phi_zero/tests

+    input = 'simple_transient_diffusion.i'
+    expect_out = ' Solve Converged!'
+    issues = '#15204'
+    design = ' '


ruh roh. You should either write some documentation or find some existing documentation that roughly fits what you're working on

lindsayad · 2020-05-14T22:55:01Z

test/tests/phi_zero/tests

+    issues = '#15204'
+    design = ' '
+    requirement = 'The system shall be able to produce consestent size _phi_zero and _grad_phi_zero objects'
+    method = 'dbg'


You might as well remove this as well. I understand that your assertions won't be tested, but the problem should still run for all methods

lindsayad · 2020-05-14T22:55:23Z

test/tests/phi_zero/tests

+    cli_args = 'Mesh/elem_type=QUAD8  Variables/dummy/order=SECOND Variables/u/order=SECOND Kernels/phi_zero/second_order=true'
+    requirement = 'The system shall be able to produce consestent size _phi_zero and _grad_phi_zero'
+    'objects for quadratic shape functions'
+    expect_out = ' Solve Converged!'


Again can remove

lindsayad · 2020-05-14T22:55:36Z

test/tests/phi_zero/tests

+    'objects for quadratic shape functions'
+    expect_out = ' Solve Converged!'
+    issues = '#15204'
+    design = ' '


lindsayad · 2020-05-14T22:56:55Z

test/tests/phi_zero/tests

+    type = 'RunApp'
+    input = 'simple_transient_diffusion.i'
+    cli_args = 'Mesh/elem_type=QUAD8  Variables/dummy/order=SECOND Variables/u/order=SECOND Kernels/phi_zero/second_order=true'
+    requirement = 'The system shall be able to produce consestent size _phi_zero and _grad_phi_zero'


could use a similar requirement to the one I proposed above

lindsayad · 2020-05-14T22:57:04Z

test/tests/phi_zero/tests

+    expect_out = ' Solve Converged!'
+    issues = '#15204'
+    design = ' '
+    method = 'dbg'


arovinelli · 2020-05-15T00:07:14Z

A nice addition. thank you @arovinelli . I'm curious what your use case is?

@lindsay This is a preparation step for implementing large deformation cohesive zone modeling. The main reason I needed this is to ease the formulation of the jacobian of the traction w.r.t. the discreet displacements. It might not be optimal in terms of efficiency but make the code much more understandable. It will also allow using exactly the same formulation for 1D 2D and 3D problems, thus avoiding a bunch of if statements that are detrimental when performing maintenance.

Mostly fixup the test specs file

The last PR should address all your comments. Check the design and requirement parts to see if they comply with your specs.

Alex, thanks again for helping with this!!

moosebuild · 2020-05-15T16:24:45Z

Job Private App tests PETSc alt on bd98388 : invalidated by @lindsayad

moosebuild · 2020-05-15T18:00:13Z

Job Documentation on bd98388 : invalidated by @arovinelli

something is funny, try rerun

arovinelli · 2020-05-18T17:12:31Z

I can't put the .md file in the test doc directory, as it will not be found. Suggestions?

arovinelli · 2020-05-18T21:01:46Z

@lindsay this has passed all tests, please give it another review and let me know if there is something left to do.

lindsayad

I really like how this looks. Just the two tiny doc requests, and then I'll approve!

lindsayad · 2020-05-18T23:02:03Z

framework/doc/content/source/variables/PhiZero.md

+# PhiZero
+
+MOOSE has helper zero objects for shape functions and shape function gradients that can be accessed by all MOOSE objects. The names of the zero objects are: `_phi_zero`, `_grad_phi_zero` and `_second_phi_zero`.
+These object first two dimensions are,: (i) the maximum number of degree of freedom per variable per element, and (ii) the maximum number of quadrature points per element.


remove comma in are,

lindsayad · 2020-05-18T23:03:38Z

framework/doc/content/source/variables/PhiZero.md

+
+MOOSE has helper zero objects for shape functions and shape function gradients that can be accessed by all MOOSE objects. The names of the zero objects are: `_phi_zero`, `_grad_phi_zero` and `_second_phi_zero`.
+These object first two dimensions are,: (i) the maximum number of degree of freedom per variable per element, and (ii) the maximum number of quadrature points per element.
+The `PhiZeroKernel` checks that the first two dimensions of `_phi_zero`, `_grad_phi_zero` and `_second_phi_zero` are


Probably say the "PhiZeroKernel test object"

arovinelli commented May 1, 2020

View reviewed changes

moosebuild added the PR: Failed but allowed label May 2, 2020

moosebuild removed the PR: Failed but allowed label May 4, 2020

moosebuild added the PR: Failed but allowed label May 12, 2020

moosebuild removed the PR: Failed but allowed label May 13, 2020

arovinelli force-pushed the grad_zero_phi_test branch from 238a3de to 3ea16a8 Compare May 13, 2020 22:34

Andrea Rovinelli added 6 commits May 14, 2020 12:20

ugly init but working

d1f8c46

ref idaholab#15204

init is ok now

ed46f78

added _phi_zero

f5b3efa

added tests

0f5c3a2

fixed error generating lines

3455e68

this fixes waht went wrong during rebase

75a8f5f

arovinelli force-pushed the grad_zero_phi_test branch from 2a4ef51 to 75a8f5f Compare May 14, 2020 17:21

arovinelli commented May 14, 2020

View reviewed changes

ready for review

667552f

moosebuild added the PR: Failed but allowed label May 14, 2020

lindsayad requested changes May 14, 2020

View reviewed changes

moosebuild removed the PR: Failed but allowed label May 14, 2020

arovinelli force-pushed the grad_zero_phi_test branch from c8047b6 to df9389f Compare May 14, 2020 23:58

address comments

20c49a2

arovinelli force-pushed the grad_zero_phi_test branch from df9389f to 20c49a2 Compare May 15, 2020 12:36

arovinelli force-pushed the grad_zero_phi_test branch from bd98388 to 1a8724a Compare May 15, 2020 19:27

added documentation

3e28837

arovinelli force-pushed the grad_zero_phi_test branch from 1a8724a to 3e28837 Compare May 18, 2020 17:18

moosebuild added the PR: Failed but allowed label May 18, 2020

lindsayad requested changes May 18, 2020

View reviewed changes

address review comments

e63ae8d

moosebuild removed the PR: Failed but allowed label May 19, 2020

lindsayad approved these changes May 19, 2020

View reviewed changes

moosebuild added the PR: Failed but allowed label May 19, 2020

lindsayad merged commit cec65dc into idaholab:next May 19, 2020

		_grad_phi_zero[tid].resize(max_qpts, {0., 0., 0.});
		_grad_test_zero[tid].resize(max_qpts, {0., 0., 0.});

		"_phi.size() " + std::to_string(_phi.size()) + "> _phi_zero.size() " +
		std::to_string(_phi_zero.size()));


		#include "NullKernel.h"

		class PhiZeroKernel;

Add grad_zero for phi and test function #15205

Add grad_zero for phi and test function #15205

Conversation

arovinelli commented May 1, 2020

Choose a reason for hiding this comment

lindsayad commented May 2, 2020

moosebuild commented May 2, 2020 • edited

arovinelli commented May 2, 2020

lindsayad commented May 2, 2020

arovinelli commented May 2, 2020 • edited

lindsayad commented May 3, 2020

moosebuild commented May 12, 2020

arovinelli commented May 12, 2020

lindsayad commented May 12, 2020

arovinelli commented May 13, 2020

lindsayad commented May 13, 2020 • edited

arovinelli commented May 13, 2020

lindsayad commented May 13, 2020 via email

moosebuild commented May 14, 2020

arovinelli commented May 14, 2020 • edited

arovinelli May 14, 2020 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

moosebuild commented May 14, 2020

arovinelli commented May 14, 2020

lindsayad left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

arovinelli commented May 15, 2020

moosebuild commented May 15, 2020

moosebuild commented May 15, 2020

arovinelli commented May 18, 2020

arovinelli commented May 18, 2020

lindsayad left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

moosebuild commented May 2, 2020 •

edited

arovinelli commented May 2, 2020 •

edited

lindsayad commented May 13, 2020 •

edited

arovinelli commented May 14, 2020 •

edited

arovinelli May 14, 2020 •

edited