Tupek/residual redesign by tupek2 · Pull Request #1354 · llnl/smith

tupek2 · 2025-04-04T04:39:36Z

This completes some of the follow ons from the previous review (naming, _, etc.), but also extends and simplifies the use by including a serac::FunctionalResidual, which will eliminate a TON of duplication in the downstream smith new physics. Also, now SolidResidual derives from FunctionalResidual.

src/serac/physics/functional_residual.hpp

white238 · 2025-04-10T19:20:58Z

src/serac/physics/functional_residual.hpp

+ *
+ */
+template <typename ShapeSpace, typename OutputSpace, typename... parameter_space, int... parameter_indices>
+class FunctionalResidual<ShapeSpace, OutputSpace, Parameters<parameter_space...>,


Is FunctionalResidual descriptive enough?

I hope the comments help, and I wouldn't really know what other words to use. The idea: its the 'residual' implemented using the serac::Functional utilities. Eventually we'll have a DfemResidual, or some such.

src/serac/physics/materials/solid_material.hpp

src/serac/physics/mesh.hpp

src/serac/physics/tests/test_functional_residual.cpp

white238 · 2025-04-10T19:26:28Z

src/serac/physics/tests/test_solid_residual.cpp

  }
-  for (auto& t : params) {
-    pointers.push_back(&t);
+  for (size_t n = 1; n < params.size(); ++n) {


Why is the 0 index being skipped here?

Added a comment to explain a bit more what it happening here.

Actually, I changed it to remove this confusion.

src/serac/physics/functional_residual.hpp

…lities.

… line.

…le time the spatial dimension of the problem, so that was added.

…l residual.

btalamini

I really like how streamlined the solid residual class is now!

btalamini · 2025-04-11T13:48:09Z

src/serac/physics/functional_residual.hpp

+  {
+    SLIC_ERROR_IF(vjpFields.size() != fields.size(),
+                  "Invalid number of field sensitivities relative to the number of fields");
+    SLIC_ERROR_IF(vReactions.size() != 1, "Solid mechanics nonlinear system only supports 1 output residual");


Says "Solid mechanics" again. If the size is 0, this error message is technically correct, but probably misleading. Consider checking size == 0 and size > 1 separately.

I think this is fixed now

src/serac/physics/functional_residual.hpp

src/serac/physics/tests/test_solid_residual.cpp

src/serac/physics/functional_residual.hpp

ebchin

Thanks @tupek2, these changes are a definite improvement over the current state of the code.

Two things to consider going forward: 1) how this will work with potential replacements for Functional and 2) how it will work with constraints/contact.

ebchin · 2025-04-11T20:21:27Z

src/serac/physics/tests/test_functional_residual.cpp

@@ -0,0 +1,277 @@
+// Copyright (c) 2019-2024, Lawrence Livermore National Security, LLC and


Is the naming redundant here? Maybe we can remove test_ from the name, but then it's easily confused with functional_residual.hpp. Hmm, maybe just leave it.

because its hard to tell if a file is a test, example, or library code, I've suggestion we come up with a naming convention. LiDO does something similar, we could adopt their convention, though I'm ok with this one too.

This should be a meeting topic. I have thoughts but wont die on this hill

ebchin · 2025-04-11T20:27:30Z

src/serac/physics/tests/test_functional_residual.cpp

+  using VectorSpace = serac::H1<disp_order, dim>;
+  using DensitySpace = serac::L2<disp_order - 1>;
+
+  enum STATE


enum class STATE?

enum class entries are not size_t, so it doesn't achieve what I'm trying to do here, which is simply give names to the indices to improve readability without having to add lies (aka, comments).

ebchin · 2025-04-11T20:27:40Z

src/serac/physics/tests/test_functional_residual.cpp

+    VELO
+  };
+
+  enum PAR


enum class PAR?

same, these are indented to be classical enums to index into things with names.

ebchin · 2025-04-11T20:35:02Z

src/serac/physics/tests/test_functional_residual.cpp

+        serac::StateManager::newState(VectorSpace{}, "shape_displacement", mesh->tag());
+    serac::FiniteElementState density = serac::StateManager::newState(DensitySpace{}, "density", mesh->tag());
+
+    states = {disp, velo};


Can these be moved here, or does serac::StateManager need to know they exist? Though I think there's still a copy required here, so it probably doesn't matter.

there are all shared_ptr, I'm not worried about the copy and reference counting cost at this high of a level.

ebchin · 2025-04-11T21:28:03Z

src/serac/physics/functional_residual.hpp

+ protected:
+  /// @brief Utility to evaluate residual using all fields in vector
+  template <int... i>
+  auto evaluateResidual(std::integer_sequence<int, i...>, double time, const std::vector<FieldPtr>& fs) const


Doesn't look like this is used anywhere. Should it be removed?

Good catch, removing. It was needed in an earlier iteration of the design, no longer.

ebchin · 2025-04-11T21:33:37Z

src/serac/physics/solid_residual.hpp

-class SolidResidual<order, dim, Parameters<parameter_space...>, std::integer_sequence<int, parameter_indices...>>
-    : public Residual {
+template <int order, int dim, typename... InputSpaces>
+class SolidResidual<order, dim, Parameters<InputSpaces...>>


How does this work if we create a SolidResidual built on, say, dFEM?

I suppose we could call this SolidFunctionalResidual, when we are there, there will be a DfemResidual, then maybe a SolidDfemResidual based on that. We can do the name changes at that point if we feel like it.

tupek2 force-pushed the tupek/residual_redesign branch from d19b179 to 7ae3ea8 Compare April 4, 2025 05:16

tupek2 requested review from btalamini, ebchin and white238 April 4, 2025 14:49

tupek2 commented Apr 4, 2025

View reviewed changes

src/serac/physics/functional_residual.hpp Outdated Show resolved Hide resolved

tupek2 force-pushed the tupek/residual_redesign branch from 7ae3ea8 to 3d6591b Compare April 7, 2025 17:01

tupek2 added the ready for review Ready for active inspection by reviewers label Apr 7, 2025

white238 reviewed Apr 10, 2025

View reviewed changes

src/serac/physics/functional_residual.hpp Outdated Show resolved Hide resolved

white238 reviewed Apr 10, 2025

View reviewed changes

src/serac/physics/materials/solid_material.hpp Outdated Show resolved Hide resolved

white238 reviewed Apr 10, 2025

View reviewed changes

src/serac/physics/mesh.hpp Outdated Show resolved Hide resolved

white238 reviewed Apr 10, 2025

View reviewed changes

src/serac/physics/tests/test_functional_residual.cpp Outdated Show resolved Hide resolved

white238 reviewed Apr 10, 2025

View reviewed changes

src/serac/physics/functional_residual.hpp Outdated Show resolved Hide resolved

white238 reviewed Apr 10, 2025

View reviewed changes

src/serac/physics/functional_residual.hpp Outdated Show resolved Hide resolved

tupek2 force-pushed the tupek/residual_redesign branch from 3d6591b to 4948a0b Compare April 11, 2025 00:03

btalamini reviewed Apr 11, 2025

View reviewed changes

src/serac/physics/functional_residual.hpp Outdated Show resolved Hide resolved

btalamini reviewed Apr 11, 2025

View reviewed changes

src/serac/physics/functional_residual.hpp Show resolved Hide resolved

tupek2 added 12 commits April 11, 2025 09:39

Add comments for new material model for use in new solid_residual.

cb1b025

Refactoring to include a generic residual using serac::Functional uti…

f6afcfe

…lities.

Fix style and docs.

7783fa1

Solid residual extends functional residual, less duplication down the…

b9e3e15

… line.

Fix up style and docs.

46367b3

Change the order for specifying functions.

25e92fa

Update inteface to avoid users really interacting with domains as all.

a54269f

style

08d51d8

More style

d6375a1

Fix a few of the review comments.

c003343

Fixing up some test and syntax change issues.

631890e

Rename some typedefs.

39b30af

tupek2 added 5 commits April 11, 2025 09:39

Fix style.

877ea35

Fix a few interface issues.

94326c4

Fix test usage of new names.

ba3db27

Fix docs and style.

fda665a

Address some comments, the functional residual needs to know at compi…

c8090fb

…le time the spatial dimension of the problem, so that was added.

tupek2 force-pushed the tupek/residual_redesign branch from bca7142 to c8090fb Compare April 11, 2025 16:40

tupek2 added 2 commits April 11, 2025 11:30

Fix tests after order change and new template dim usage in sfunctiona…

9ea738d

…l residual.

Fix comment.

178d2f9

btalamini approved these changes Apr 11, 2025

View reviewed changes

Refactor materials to use a more consistent interface.

db67f6c

ebchin approved these changes Apr 11, 2025

View reviewed changes

tupek2 added 2 commits April 11, 2025 16:26

Fix docs.

f41149d

Remove unused function.

7cd94e5

btalamini requested a review from lihanyu97 April 14, 2025 16:13

tupek2 merged commit 1785a6b into develop Apr 14, 2025
13 checks passed

tupek2 deleted the tupek/residual_redesign branch April 14, 2025 22:44

		@@ -0,0 +1,277 @@
		// Copyright (c) 2019-2024, Lawrence Livermore National Security, LLC and

Conversation

tupek2 commented Apr 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

btalamini left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ebchin left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

tupek2 commented Apr 4, 2025 •

edited

Loading