Use AutoDiffScalar Concept and fix checkFESByAutoDiff test #329

henrij22 · 2024-09-27T13:46:29Z

No description provided.

codecov · 2024-09-27T13:57:23Z

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 80.86%. Comparing base (8e2263e) to head (492863b).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #329      +/-   ##
==========================================
- Coverage   86.66%   80.86%   -5.80%     
==========================================
  Files          69       68       -1     
  Lines        2234     2299      +65     
==========================================
- Hits         1936     1859      -77     
- Misses        298      440     +142

Flag	Coverage Δ
tests	`80.86% <100.00%> (-5.80%)`	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

to be included along with #330

tarun-mitruka · 2024-09-30T15:14:09Z

This action still failed irrespective of the fact that no floating-point exceptions were raised in this commit(note that this failure is not attributed from this PR).
The tests are hence not yet stable and I would recommend to wait before merging this branch and finding a fix before, even though it looks like the CI works at the moment

ikarus/finiteelements/mechanics/materials/vanishingstress.hh

rath3t · 2024-10-01T07:53:38Z

This action still failed irrespective of the fact that no floating-point exceptions were raised in this commit(note that this failure is not attributed from this PR). The tests are hence not yet stable and I would recommend to wait before merging this branch and finding a fix before, even though it looks like the CI works at the moment

The 3d test does not touch the vanishing stress code or did you find out something different?

The test now runs locally without any exceptions from the floating point?

tarun-mitruka · 2024-10-01T07:57:35Z

This action still failed irrespective of the fact that no floating-point exceptions were raised in this commit(note that this failure is not attributed from this PR). The tests are hence not yet stable and I would recommend to wait before merging this branch and finding a fix before, even though it looks like the CI works at the moment

The 3d test does not touch the vanishing stress code or did you find out something different?

No, it doesn't. At least for this commit, all the actions were successful.

This reverts commit c37e4c7.

This can be the cause of error due to non-linearity of NeoHooke

see [here](https://github.com/ikarus-project/ikarus/actions/runs/11107793669) for 3d

tarun-mitruka · 2024-10-07T13:24:37Z

OBSERVATIONS:
Here, I am documenting the current observations:

Only executing autoDiffTest<2>(/* ... */) always works (see here)
Only executing autoDiffTest<3>(/* ... */) always works (see here)
Executive the above two tests simultaneously results in some issues:

autoDiffTest<3>(/* ... */) fails ONLY for clang compiler when using CornerDistortionFlag::randomlyDistorted with nu !=0 and d.setRandom(nDOF)
autoDiffTest<3>(/* ... */) fails for BOTH clang and gcc compilers when using CornerDistortionFlag::unDistorted with nu !=0 and d.setRandom(nDOF) (see here)

The test fails irrespective of the number of minimum iterations (see here) allowed and presence of perturbation the stresses (see here) in vanishingstress.hh

Further debugging is necessary!!

!!! UPDATE
The autodiff<gridDim>(/* ... */) test executed immediately after autoDiffTest<2>(t, planeStressMat2, " nu = 0"); fails, irrespective of gridDim and nu, but only for non-zero displacements. with CornerDistortionFlag::unDistorted.

rath3t · 2024-10-09T06:17:02Z

Since it only fails if the tests are executed simultaneously, it sound still like an uninitialized memory problem or similar.
Therefore, I would suggest executing the test with valgrind to check for issues.

I would suggest, the following procedure, if the 3D-test would fail independently

Extract some configuration (node positions+ displacements) for which the tests fail
Modify the test to explicitly check for the same configuration
Check if NeoHooke and StVenantKirchhoff, become the same for displacements going to zero and Poisson's ratio going to zero
Reduce the formulation to a single integration point
Extract the corresponding strain for which the test fails
Create a new test without any element but only with the material and check if energy,stresses and tangent moduli are correct/the same for autodiff and implementation (This test should already exists somewhere)

use FloatCmp and change tolerance remove noexcept

tarun-mitruka · 2024-10-16T14:12:27Z

I think I found the source of the error. I ran the tests 7 times and it always works now (see this action run).

CornerDistortionFlag::randomlyDistorted adds distortions in the range [-0.2, 0.2] to the nodal positions of the reference cube element. (see here).
d.setRandom(basis.flat().dimension()) creates a vector with entities in the range [-1, 1] (from Eigen). This can sometimes leads to a heavily distorted element with a negative deformation gradient.
This breaks the sqrt function here in NeoHooke as it is not defined for negative values.
I think depending on the randomness, the observations I mentioned here was a special case of the above mentioned issue (undefined behavior).
I now scaled the displacements in checkFESByAutoDiff such that the random values in the displacement vectors are in the range [-0.2, 0.2] and also added checks in NeoHooke such that it throws a proper error message when detC < 0. May be this can be scaled down further for extreme case scenarios?

tarun-mitruka · 2024-10-16T14:18:21Z

tests/src/testnonlinearelasticityneohooke.cpp

@@ -28,8 +28,8 @@ int main(int argc, char** argv) {
  t.subTest(NonLinearElasticityLoadControlNRandTR<Grids::Yasp>(matNH1));
  t.subTest(NonLinearElasticityLoadControlNRandTR<Grids::IgaSurfaceIn2D>(matNH1));

-  autoDiffTest<2>(t, planeStressMat1, " nu != 0");
-  autoDiffTest<2>(t, planeStressMat2, " nu = 0");
+  autoDiffTest<2>(t, planeStressMat1, " nu != 0", 1e-7);


This tolerance isn't always necessary but only for certain cases depending on the randomness in the element geometry and the displacement vector (see comment). I am not sure if this is good enough or it could be resolved in a better way? May be fixing the random displacement vector in checkFESByAutoDiff such that this tolerance can be reduced further? We can also fix the tolerance while working on the Issue #211. Do you think increasing minIter to 2 or 3 for NewtonRaphson in VanishingStress would help here?

Increasing minIter might help but I'm not sure here. But using tolerances here is not to bad. But I think there is no nice way to fix this. We might refactor isApproxSame which then uses the objects to compute something reasonable: Maybe something like that https://godbolt.org/z/xEPraWKEf

Okay then I leave the tolerance here as such and we can refactor isApproxSame while perhaps working on #211

rath3t · 2024-10-16T14:59:36Z

that makes a lot of sense. I will look into it tomorrow.

ikarus/finiteelements/mechanics/materials/neohooke.hh

tarun-mitruka · 2024-10-17T15:46:51Z

The code coverage is also behaving randomly, I don't know why. Before I committed the change from MathError to InvalidStateException, codecov/project increased by +2.01% and after the commit it decreased with -6.36%. How can this happen?

rath3t · 2024-10-21T09:27:19Z

ikarus/finiteelements/mechanics/materials/neohooke.hh

+      const auto detC   = C.determinant();
+      if (Dune::FloatCmp::lt(static_cast<double>(detC), 0.0, 1e-10))
+        DUNE_THROW(Dune::InvalidStateException,
+                   "Determinant of right Cauchy Green tensor C must be greater than or equal to zero. detC = " +


It should be equal to zero. It should really be bigger. From theory.

You mean it shouldn't be equal to zero? So change FloatCmp::lt to FloatCmp::le?

Ah sorry. I missed a word. It should not be equal to zero. Therefore detC>0. The statement in the if statement is fine but the exception text is saying detC>=0.

ikarus/finiteelements/mechanics/materials/neohooke.hh

rath3t · 2024-10-21T09:30:23Z

ikarus/finiteelements/mechanics/materials/neohooke.hh

@@ -96,7 +101,12 @@ struct NeoHookeT : public Material<NeoHookeT<ST>>
    static_assert(Concepts::EigenMatrixOrVoigtNotation3<Derived>);
    if constexpr (!voigt) {
      if constexpr (!Concepts::EigenVector<Derived>) {
-        const auto logdetF = log(sqrt(C.determinant()));
+        const auto detC = C.determinant();
+        if (Dune::FloatCmp::lt(static_cast<double>(detC), 0.0, 1e-10))


Since you are repeating that functionality I would move it to a private member function and call it check positive determinant of C.

rath3t · 2024-10-21T09:32:37Z

ikarus/finiteelements/mechanics/nonlinearelastic.hh

-    if constexpr (std::is_same_v<ScalarType, double>)
-      return mat_.template tangentModuli<strainType, voigt>(strain);
-    else {
+    if constexpr (Concepts::AutodiffScalar<ScalarType>) {


Maybe we also want a private function getmaterial, which does this compile time if statement and returns the correct material object. Make sure that decltype(auto) is placed everywhere to not introduce copies

tests/src/checkfebyautodiff.hh

rath3t · 2024-10-21T09:36:14Z

tests/src/testmaterial.cpp

+
+  t.checkThrow<Dune::InvalidStateException>(
+      [&]() {
+        const auto moduli = (planeStress(matNH, 1e-8).template tangentModuli<StrainTags::greenLagrangian, true>(E));


Maybe not repeat planeStress(matNH, 1e-8) but use the same object all the time?

rath3t · 2024-10-21T09:57:34Z

The code coverage is also behaving randomly, I don't know why. Before I committed the change from MathError to InvalidStateException, codecov/project increased by +2.01% and after the commit it decreased with -6.36%. How can this happen?

This is also a mystery for me, therefore I take these changes always with a grain of salt. We might remove that even or change to another provider

rath3t · 2024-10-21T09:58:34Z

In total for me this looks quite fine and the comments might also happen in another PR except for the code duplication stuff. Still you can quickly look into if you have an idea that can be done fast.
Thanks Tarun for investigation!

ikarus/finiteelements/mechanics/materials/neohooke.hh

rath3t

Ok this is fine for me.

fix nonlinear

9e48676

henrij22 linked an issue Sep 27, 2024 that may be closed by this pull request

concept AutodiffScalar has to be used wherever applicable #328

Closed

fix eas

28c7f15

tarun-mitruka changed the title ~~Use AutoDiff Scalar Concept~~ Use AutoDiffScalar Concept Sep 30, 2024

henrij22 requested a review from rath3t September 30, 2024 09:35

tarun-mitruka force-pushed the fix/adconcept branch from 8ee9668 to 28c7f15 Compare September 30, 2024 11:59

add numerical sanitizers

ba9070e

tarun-mitruka force-pushed the fix/adconcept branch from c8c0d6a to ba9070e Compare September 30, 2024 13:59

tarun-mitruka added 4 commits September 30, 2024 16:04

check if it works in 3d (only)

8255673

fix?

884a956

fix?

75944eb

remove numerical sanitizers

a806cd3

to be included along with #330

tarun-mitruka assigned henrij22 and tarun-mitruka Sep 30, 2024

tarun-mitruka added bug Something isn't working enhancement New request to enhance existing features labels Sep 30, 2024

rath3t reviewed Sep 30, 2024

View reviewed changes

ikarus/finiteelements/mechanics/materials/vanishingstress.hh Outdated Show resolved Hide resolved

tarun-mitruka added 3 commits October 4, 2024 13:46

refactor

c37e4c7

Revert "refactor"

512fa92

This reverts commit c37e4c7.

remove perturbation on moduli

32c59a9

tarun-mitruka marked this pull request as draft October 4, 2024 12:30

tarun-mitruka added 4 commits October 4, 2024 15:11

try increasing minIter

7c0fe4f

This can be the cause of error due to non-linearity of NeoHooke

try only 2d autodiff as only 3d autodiff always works

3207d3c

see [here](https://github.com/ikarus-project/ikarus/actions/runs/11107793669) for 3d

try autodiff on 2d and 3d only

21ffe03

try changing randomlyDistorted to unDistorted

9ad6eda

tarun-mitruka changed the title ~~Use AutoDiffScalar Concept~~ Use AutoDiffScalar Concept and fix checkFESByAutoDiff test Oct 16, 2024

fix and refactor

dee9f3a

use FloatCmp and change tolerance remove noexcept

tarun-mitruka force-pushed the fix/adconcept branch from 66e4dad to dee9f3a Compare October 16, 2024 11:34

tarun-mitruka reviewed Oct 16, 2024

View reviewed changes

tarun-mitruka marked this pull request as ready for review October 16, 2024 14:18

tarun-mitruka requested a review from rath3t October 16, 2024 14:18

tarun-mitruka reviewed Oct 17, 2024

View reviewed changes

ikarus/finiteelements/mechanics/materials/neohooke.hh Outdated Show resolved Hide resolved

replace MathError with InvalidStateException

53a2c03

rath3t reviewed Oct 21, 2024

View reviewed changes

ikarus/finiteelements/mechanics/materials/neohooke.hh Outdated Show resolved Hide resolved

rath3t reviewed Oct 21, 2024

View reviewed changes

tests/src/checkfebyautodiff.hh Show resolved Hide resolved

rath3t reviewed Oct 21, 2024

View reviewed changes

tarun-mitruka added 2 commits October 21, 2024 14:18

remove code duplication

5715638

refactor nonlinearelastic

492863b

tarun-mitruka requested a review from rath3t October 21, 2024 12:36

rath3t reviewed Oct 21, 2024

View reviewed changes

ikarus/finiteelements/mechanics/materials/neohooke.hh Show resolved Hide resolved

rath3t approved these changes Oct 21, 2024

View reviewed changes

rath3t merged commit 2ff5c77 into main Oct 21, 2024
21 of 22 checks passed

rath3t deleted the fix/adconcept branch October 21, 2024 14:47

henrij22 mentioned this pull request Jan 28, 2025

[IN REVIEW] Add Hyperelastic and AutoDiff-based Material models #333

Merged

26 tasks

Use AutoDiffScalar Concept and fix checkFESByAutoDiff test #329

Use AutoDiffScalar Concept and fix checkFESByAutoDiff test #329

Uh oh!

Conversation

henrij22 commented Sep 27, 2024

Uh oh!

codecov bot commented Sep 27, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

tarun-mitruka commented Sep 30, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

rath3t commented Oct 1, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tarun-mitruka commented Oct 1, 2024

Uh oh!

tarun-mitruka commented Oct 7, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rath3t commented Oct 9, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

tarun-mitruka commented Oct 16, 2024

Uh oh!

tarun-mitruka Oct 16, 2024 • 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rath3t commented Oct 16, 2024

Uh oh!

Uh oh!

tarun-mitruka commented Oct 17, 2024 • 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!

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!

rath3t commented Oct 21, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rath3t commented Oct 21, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

rath3t left a comment • 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!

codecov bot commented Sep 27, 2024 •

edited

Loading

tarun-mitruka commented Sep 30, 2024 •

edited

Loading

rath3t commented Oct 1, 2024 •

edited

Loading

tarun-mitruka commented Oct 7, 2024 •

edited

Loading

rath3t commented Oct 9, 2024 •

edited

Loading

tarun-mitruka Oct 16, 2024 •

edited

Loading

tarun-mitruka commented Oct 17, 2024 •

edited

Loading

rath3t commented Oct 21, 2024 •

edited

Loading

rath3t commented Oct 21, 2024 •

edited

Loading

rath3t left a comment •

edited

Loading