Gretl submodule #1495

tupek2 · 2025-11-21T02:02:12Z

This adds gretl as a submodule and provides a first demo using FieldState (gretl-wrapped FiniteElementState and FiniteElementDual) for an explicit dynamics problem with dynamic checkpointing and back-propagation. Some minor differentiable_physics interface adjustments as well. Updating tribol to the latest develop branch to address a somewhat localized issue with builds that do not enable umpire.

… key things.

…al function.

…ndle to it.

Co-authored-by: Chris White <[email protected]>

…_state_advancer.hpp Co-authored-by: Chris White <[email protected]>

Co-authored-by: Chris White <[email protected]>

btalamini

The integration looks very close to what I was expecting, so LGTM.

btalamini · 2025-12-11T21:51:20Z

src/smith/differentiable_numerics/lumped_mass_explicit_newmark_state_advancer.cpp

+
+namespace smith {
+
+/// @brief uses the constrained docs on the bc_manager to zero the corresponding dofs in FieldState s.


Suggested change

/// @brief uses the constrained docs on the bc_manager to zero the corresponding dofs in FieldState s.

/// @brief uses the constrained dofs on the bc_manager to zero the corresponding dofs in FieldState s.

btalamini · 2025-12-11T22:02:08Z

src/smith/differentiable_numerics/lumped_mass_weak_form.hpp

+      return smith::tuple{get<VALUE>(Rho), tensor<double, spatial_dim>{}};
+    } else {
+      auto ones = make_tensor<lumped_dim>([](int) { return 1.0; });
+      return smith::tuple{-get<VALUE>(Rho) * ones, tensor<double, lumped_dim, spatial_dim>{}};


The fact that only the singleton case is negated is making me suspicious of a bug. Is this intentional?

its been too long. Probably it wouldn't compile without it. I mean, the ranks of the tensors are different. Maybe there is some nicer way to handle it, but I can't see it looking at it now.

I see, there was a bug, that was compensated for elsewhere. Fixed in an update.

src/smith/differentiable_numerics/differentiable_utils.hpp

btalamini · 2025-12-11T22:24:39Z

src/smith/differentiable_numerics/differentiable_utils.hpp

+namespace smith {
+
+/// Evaluates a DoubleState using a provided ScalarObjective instance, and the input arguments to that objective. This
+/// operation is tracked on the gretl graph.  ScalarObjective must remain in scope for the


looks like the last word of the comment got cut off.

btalamini · 2025-12-11T22:25:38Z

src/smith/differentiable_numerics/differentiable_utils.hpp

+DoubleState evaluateObjective(std::shared_ptr<ScalarObjective> objective, const TimeInfo& time_info,
+                              const FieldState& shape_disp, const std::vector<FieldState>& inputs);
+
+/// operation is tracked on the gretl graph.


Doc comment looks cut off. Missing the "@brief"

btalamini · 2025-12-11T22:29:59Z

src/smith/differentiable_numerics/differentiable_utils.hpp

+        auto ke = 0.5 * rho * inner(v, v);
+        auto dx_dX = get<DERIVATIVE>(U) + Identity<dim>();
+        auto J = det(dx_dX);
+        return ke * J;


Why is this multiplied by J? Rho is the density in the reference config, right? If so, no J.

well, technically, this is not a shape-aware functional implementation (this is like a 10 months old implementation at this point). So, that argument is the shape-displacement (or the displacement if the current density is passed in, e.g., for updated Lagrange things). Perhaps in a follow up PR, I can convert this to using the new ScalarObjective interface which is shape-aware always.

src/smith/numerics/functional/tensor.hpp

src/smith/physics/base_physics.hpp

btalamini · 2025-12-11T23:17:20Z

tribol

Is your tribol submodule out of date?

there was an issue in tribol where if you didn't build with umpire, some part of its build would fail, so a patch was needed.

chapman39

i know about the tribol issue, but was there something wrong with the axom and mfem submodules as well that required an update? that'd be useful to me to know so next time i update ill make sure those changes are included.

tupek2 · 2025-12-12T00:09:49Z

i know about the tribol issue, but was there something wrong with the axom and mfem submodules as well that required an update? that'd be useful to me to know so next time i update ill make sure those changes are included.

I had temporarily accidentally not merged the updated axom and mfem from dev, but I think that is fixed now? Unless I am not reading the diffs correctly?

chapman39 · 2025-12-12T00:11:57Z

src/smith/differentiable_numerics/CMakeLists.txt

+
+if(SMITH_ENABLE_TESTS)
+    add_subdirectory(tests)
+endif()


Suggested change

endif()

endif()

Is there a way our make style can just do this for us automatically? Its sort of time consuming for us to constantly check it in PRs and have to update file by file.

chapman39 · 2025-12-12T00:14:11Z

src/smith/differentiable_numerics/CMakeLists.txt

+    )
+
+blt_add_library(
+    NAME        smith_differentiable_numerics


is differentiable_numerics something that would be better moved to a subdirectory of numerics?

I think my naming may have caused confusion here. The include dependencies make it clear that differentiable_numerics does not belong in numerics. It combines things in physics, numerics, gretl, etc. to create analogous structures as exist in numerics, but which will be differentiated on the gretl graph. Its not a perfect match to numerics, so perhaps a name change could avoid confusion. differentiable_utilities was another option I was considering. If people have suggestions, I'm happy to change it, but preferably in a follow on PR.

chapman39 · 2025-12-12T00:16:50Z

@tupek2

I had temporarily accidentally not merged the updated axom and mfem from dev, but I think that is fixed now? Unless I am not reading the diffs correctly?

i see that now. ok, nevermind then! besides my other question lgtm

cmake/thirdparty/SetupSmithThirdParty.cmake

tupek2 added 30 commits August 5, 2025 14:19

Initial gretl in serac. Need to work on some tests, maybe documenting…

4ccb49e

… key things.

Revert back to original gretl.

77b56ab

--amend

f7e758b

Trying to get original checkpoint tests passing again.

c19b7ba

Operation gretl again.

4ec0b48

Make it a bit less likely toa accidentally no specify a valid zero_du…

b72a057

…al function.

Additional cleanup of compilation warnings.

ae86ce3

Add in a highly nonlinear and more complicated graph.

0746ec4

Add in a highly nonlinear and more complicated graph.

1ce7e8d

Get test tolerance in place.

2490a51

Fix style.

471dfa5

Try to fix some gcc build issues.

a928494

Fix another gcc issue.

449b438

add all headers to cmake lists.

bb271cf

Another header missing.

dbe1feb

Another gcc build fix.

f5898c5

Try again to fix gcc13 warnings.

be0ea51

Trying to address gcc issues.

96fcff9

Try to fix gcc warnings.

37c76a7

Try again to resolve gcc warning.

61fd848

Debugging gretl.

44107f6

Trying to simplify gretl.

c90d23f

Simplifying gretl implementation, no checkpointing yet.

0469141

Seemingly got gretl, no checkpoint working again.

ba58d95

Cleanup the initial, not dynamically checkpointed gretl.

02c56fd

Remove some prints, only clear shared pointer if no one else has a ha…

37c7340

…ndle to it.

style.

ddee2ce

Fix build.

80d7b4d

Fix style.

fce33b1

Fix style.

094c1d0

tupek2 and others added 19 commits December 8, 2025 14:08

Update src/smith/differentiable_numerics/field_state.cpp

a42ee65

Co-authored-by: Chris White <[email protected]>

Update src/smith/differentiable_numerics/field_state.hpp

fa4a882

Co-authored-by: Chris White <[email protected]>

Update src/smith/differentiable_numerics/lumped_mass_explicit_newmark…

18f4347

…_state_advancer.hpp Co-authored-by: Chris White <[email protected]>

Update src/smith/differentiable_numerics/lumped_mass_explicit_newmark…

d6a7d4e

…_state_advancer.hpp Co-authored-by: Chris White <[email protected]>

Update src/smith/differentiable_numerics/lumped_mass_weak_form.hpp

a6980f7

Co-authored-by: Chris White <[email protected]>

Update src/smith/differentiable_numerics/state_advancer.hpp

bbaa967

Co-authored-by: Chris White <[email protected]>

Update src/smith/differentiable_numerics/timestep_estimator.hpp

7e2cd78

Co-authored-by: Chris White <[email protected]>

Update src/smith/physics/functional_weak_form.hpp

678ec06

Co-authored-by: Chris White <[email protected]>

Update cmake/thirdparty/SetupSmithThirdParty.cmake

d127520

Co-authored-by: Chris White <[email protected]>

Merge branch 'develop' into gretl_submodule

d89ec82

Update to new uberenv.

f15a351

Fix a copy.

4252a44

Differentiable physics var can be private.

8855f26

Keep up with uberenv.

37ba752

Merge branch 'develop' into gretl_submodule

858dced

Update submodules.

5e0c330

Use correct submodules.

d189a6e

Update tribol version.

13e1b72

Merge branch 'develop' into gretl_submodule

2d929c4

tupek2 requested a review from chapman39 December 11, 2025 22:07

btalamini approved these changes Dec 11, 2025

View reviewed changes

chapman39 requested changes Dec 12, 2025

View reviewed changes

chapman39 reviewed Dec 12, 2025

View reviewed changes

cmake/thirdparty/SetupSmithThirdParty.cmake Outdated Show resolved Hide resolved

tupek2 and others added 2 commits December 11, 2025 19:55

Fix a negation bug.

46695fe

add option for Gretl

4e64694

white238 mentioned this pull request Dec 12, 2025

Gretl solid mechanics #1501

Open


		namespace smith {

		/// @brief uses the constrained docs on the bc_manager to zero the corresponding dofs in FieldState s.

	/// @brief uses the constrained docs on the bc_manager to zero the corresponding dofs in FieldState s.
	/// @brief uses the constrained dofs on the bc_manager to zero the corresponding dofs in FieldState s.

Gretl submodule #1495

Are you sure you want to change the base?

Gretl submodule #1495

Uh oh!

Conversation

tupek2 commented Nov 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

tupek2 Dec 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

chapman39 left a comment

Choose a reason for hiding this comment

Uh oh!

tupek2 commented Dec 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

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!

Choose a reason for hiding this comment

Uh oh!

chapman39 commented Dec 12, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

tupek2 commented Nov 21, 2025 •

edited

Loading

tupek2 Dec 12, 2025 •

edited

Loading

tupek2 commented Dec 12, 2025 •

edited

Loading