AD Strong DG - Implicit Inviscid - Computes Jacobians #261

AlexanderCicchino · 2024-06-27T17:17:07Z

Strong DG now uses AD to compute Jacobians for dR/du and dR/dX for pdes that don't have diffusion. Implicit strong DG implemented for convective pdes. Operators templated.

Implementation Changes:
-Operators: removed the real template and instead templated the function calls, like matrix_vector_mult. This change was made because the class doesn't need to know the template since the basis functions are double, and the template only needs to be used when the operator is applied on a vector.
-DG Structure: The tape was pulled out of the functions in DGWeak into DGBase. The tape is defined in assemble_cell_residual_and_ad_derivatives, with sensitivities computed in assemble_volume_codi_taped_derivatives_ad, and face and boundary terms. For DGStrong, we reuse the metric terms from the interior cell on the face, so we set the tape position and compute the terms before the volume residual call in build_volume_metric_operators.
-DGStrong/Weak: The functions left are the residual evaluations, which are templated by real2 in DGWeak or adtype in DGStrong.
-Small changes: dealii::FullMatrix doesn't accept an AD argument so I changed the structure for the Hadamard products etc in DGStrong to use an std:::vector.

…or face term.

…tions using vectors instead of dealii::FullMatrix because FullMatrix cannot use AD variables.

…g for DG strong to us AD variables.

…utes Jacobian DRDX and DRDW. Have it passing selected previous ctests.

… the regression unit tests for advection and the implicit solve for Euler Gaussian bump with an entorpy stable scheme.

AlexanderCicchino · 2024-06-27T17:18:54Z

Currently running full ctest and will post log file when it is completed.

AlexanderCicchino · 2024-06-27T18:30:53Z

Added finite difference vs Ad Jacobian unit tests. dR/dU works/validated, but dR/dX does not pass for strong currently.

AlexanderCicchino · 2024-06-27T19:54:02Z

dR/dX works now for dg strong and added sensitivity unit tests verifying. Also added unit tests for d2R

PranshulThakur · 2024-07-02T14:36:15Z

dR/dX works now for dg strong and added sensitivity unit tests verifying. Also added unit tests for d2R

Thanks @AlexanderCicchino . The current PR of computing sensitivities with strong DG will facilitate several future avenues for research. There is currently an issue on narval with cmake when compiling master and is not specific to this PR. I will post the results of running ctest on narval soon once the issue is fixed.

cpethrick · 2024-07-02T21:08:32Z

I just tried compiling in debug mode and there are some issues with the Assert statements and it doesn't build. Commenting so that the issue is recorded!

Thanks for the description of the implementation changes, @AlexanderCicchino, they are quite helpful :)

AlexanderCicchino · 2024-07-03T16:31:33Z

I just tried compiling in debug mode and there are some issues with the Assert statements and it doesn't build. Commenting so that the issue is recorded!

Thanks for the description of the implementation changes, @AlexanderCicchino, they are quite helpful :)

Fixed the debug mode Assert issues.

cpethrick

I've done one pass over the PR. It seems like a tremendous amount of work and I'm glad to see it working! I've left a few comments and questions. AD is not a familiar topic for me so my comments are mainly about readability. I can go through it again in more detail if needed.

cpethrick · 2024-07-02T19:29:39Z

src/dg/dg_base.cpp

-            for(int idim=0; idim<dim; idim++){
-                rhs_aux[idim][current_dofs_indices[i]] += current_cell_rhs_aux[i][idim];
+        for(int idim=0; idim<dim; idim++){
+            for (unsigned int i=0; i<n_dofs_curr_cell; ++i) {


maybe idof would be more consistent with other loops in DG?

cpethrick · 2024-07-02T19:31:59Z

src/dg/dg_base.cpp

+    dealii::Tensor<1,dim,std::vector<real>> current_cell_rhs_aux;// Defaults to 0.0 initialization
+    if(compute_auxiliary_right_hand_side){
+        for(int idim=0; idim<dim; idim++){
+            current_cell_rhs_aux[idim].resize(n_dofs_curr_cell);


Did the old implementation allocate rhs_aux for any PDE, even ones without dissipation? I.e., does this version have a lower memory use for inviscid PDEs?

cpethrick · 2024-07-02T19:39:34Z

src/dg/dg_base.cpp

+            for (unsigned int itest_int=0; itest_int<n_soln_dofs_int; ++itest_int) {
+                current_cell_rhs_aux[idim][itest_int] += getValue<adtype>(aux_rhs_int[idim][itest_int]);
+            }
+        //    for (unsigned int itest_ext=0; itest_ext<n_soln_dofs_ext; ++itest_ext) {


Commented code

cpethrick · 2024-07-02T19:41:20Z

src/dg/dg_base.cpp

+        x_end   = x_start + n_metric_dofs;
+        w_start = x_end;
+        w_end   = w_start + 0;
+    } //else {


Commented code (also in the next function)

cpethrick · 2024-07-02T19:48:44Z

src/dg/dg_base.hpp

        const bool                                                         compute_auxiliary_right_hand_side,//flag on whether computing the Auxiliary variable's equations' residuals
        dealii::LinearAlgebra::distributed::Vector<double>                 &rhs,
        std::array<dealii::LinearAlgebra::distributed::Vector<double>,dim> &rhs_aux);

+    template <typename adtype>


Can you add comments about this function and following ones?

cpethrick · 2024-07-02T19:52:36Z

src/dg/strong_dg.cpp

+        const unsigned int ishape = this->fe_collection[poly_degree].system_to_component_index(idof).second;
+        if(ishape == 0)
+            soln_coeff[istate].resize(n_shape_fns);
+        soln_coeff[istate][ishape] = soln_coeffs[idof];


Suggested change

soln_coeff[istate][ishape] = soln_coeffs[idof];

soln_coeff[istate][ishape] = soln_coeffs[idof];

cpethrick · 2024-07-02T19:53:58Z

src/dg/strong_dg.cpp

+        soln_coeff[istate][ishape] = soln_coeffs[idof];
+        for(int idim=0; idim<dim; idim++){
+            if(ishape == 0)
+                aux_soln_coeff[istate][idim].resize(n_shape_fns);


Does this need to be done in inviscid cases?

cpethrick · 2024-07-04T14:51:28Z

src/dg/strong_dg.hpp

-        typename dealii::DoFHandler<dim>::active_cell_iterator /*cell*/,
+        adtype                                                 &dual_dot_residual);
+
+    void assemble_volume_term_and_build_operators_ad(


Add documentation to several functions in this header

Maybe highlight the difference between the similar functions when adding comments? I see that these take different types as arguments. It would be a bit easier to read if these are written explicitly in the documentation :)

cpethrick · 2024-07-04T18:30:19Z

tests/unit_tests/sensitivities/d2RdWdW_fd_vs_ad_strong.cpp

Is it possible to use the same code for strong and weak? I see that this code is very similar to d2RdWdW_fd_vs_ad.cpp. Could you either unify these tests and have the same test run twice (once for strong, once for weak), or have an input variable/PRM file?

If it's too much work to unify, it's also okay to leave as-is.

cpethrick · 2024-07-04T18:32:38Z

tests/unit_tests/sensitivities/CMakeLists.txt

Could you add "_weak" to the names of the old sensitivity tests? I would also prefer that they use upper-case letters to be consistent with other tests, though I understand that the current naming was chosen for consistency with the old tests.

PranshulThakur and others added 30 commits May 17, 2024 17:28

Added assemble_cell_residual_derivative()

02c2450

created cell_residual_derivatives

61cfe46

Added code of volume term.

9f911d8

Added code.

1b4f2aa

Added more code.

8f2e84d

Added code.

6d64a72

Added code.

6551bf9

Added code.

b4179ae

Added code.

de75177

Added code.

a4275e4

Added code.

338c744

Added code.

31d9035

Added code.

cf6e592

Restructured code.

d93adc9

Fixed compile errors.

2f888a7

Added code.

50b0448

Working AD which is moved out.

fc758fd

Added test to look at derivatives. To be removed later.

e0c2a59

Updated DGBase with position-based AD tape evaluation.

451673b

Removed old test files.

a85a877

Renamed functions.

3f088e8

Merged latest master.

7878659

Moved volume and boundary fe values in weak dg. Still need to do it f…

419d41d

…or face term.

Added additional parameters to volume residual ad function

9465e53

Moved fe_values in weakDG.

4b460c1

Renamed tape position.

488f9d5

Templated operators for all AD type and restructured it for the opera…

9d7c2f2

…tions using vectors instead of dealii::FullMatrix because FullMatrix cannot use AD variables.

Restructured header functions in header files for DG/DG_Weak/DG_Stron…

fca9e77

…g for DG strong to us AD variables.

Restructured DG/DG Strong for AD variables with operators class--comp…

6698ae3

…utes Jacobian DRDX and DRDW. Have it passing selected previous ctests.

Added a few tests showcasing the AD Jacobian works for DG strong with…

e4b18de

… the regression unit tests for advection and the implicit solve for Euler Gaussian bump with an entorpy stable scheme.

AlexanderCicchino marked this pull request as draft June 27, 2024 17:18

Added strong form finite difference vs AD unit tests.

d4d8185

dRdX for strong works now with sensitivity unit tests to verify.

c73c82f

AlexanderCicchino added 2 commits June 27, 2024 16:11

D2R for strong dg.

288b4c7

Cleaned up DG/DGStrong/DGWeak from deprecated functions.

eeb1977

AlexanderCicchino changed the title ~~AD Strong DG - Computes Jacobians~~ AD Strong DG - Implicit Inviscid - Computes Jacobians Jun 27, 2024

AlexanderCicchino marked this pull request as ready for review June 27, 2024 21:17

PranshulThakur requested review from jbrillon, ssaishruthi98 and cpethrick July 2, 2024 14:36

AlexanderCicchino and others added 3 commits July 3, 2024 11:40

Merge branch 'dougshidong:master' into ad_strong_weak_dg

f72b78b

Fixed issue with auxiliary RHS test.

817730e

Fixed assert issues when compile with debug.

6ee3660

cpethrick reviewed Jul 4, 2024

View reviewed changes

PranshulThakur marked this pull request as draft November 21, 2024 21:44

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

AD Strong DG - Implicit Inviscid - Computes Jacobians #261

AD Strong DG - Implicit Inviscid - Computes Jacobians #261

AlexanderCicchino commented Jun 27, 2024 •

edited

Loading

AlexanderCicchino commented Jun 27, 2024

AlexanderCicchino commented Jun 27, 2024

AlexanderCicchino commented Jun 27, 2024

PranshulThakur commented Jul 2, 2024

cpethrick commented Jul 2, 2024

AlexanderCicchino commented Jul 3, 2024

cpethrick left a comment

cpethrick Jul 2, 2024

cpethrick Jul 4, 2024

cpethrick Jul 2, 2024

cpethrick Jul 2, 2024

cpethrick Jul 2, 2024

cpethrick Jul 2, 2024

cpethrick Jul 2, 2024

cpethrick Jul 2, 2024

cpethrick Jul 4, 2024

cpethrick Jul 4, 2024

cpethrick Jul 4, 2024

cpethrick Jul 4, 2024

	soln_coeff[istate][ishape] = soln_coeffs[idof];
	soln_coeff[istate][ishape] = soln_coeffs[idof];

AD Strong DG - Implicit Inviscid - Computes Jacobians #261

Are you sure you want to change the base?

AD Strong DG - Implicit Inviscid - Computes Jacobians #261

Conversation

AlexanderCicchino commented Jun 27, 2024 • edited Loading

AlexanderCicchino commented Jun 27, 2024

AlexanderCicchino commented Jun 27, 2024

AlexanderCicchino commented Jun 27, 2024

PranshulThakur commented Jul 2, 2024

cpethrick commented Jul 2, 2024

AlexanderCicchino commented Jul 3, 2024

cpethrick 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

Choose a reason for hiding this comment

Choose a reason for hiding this comment

AlexanderCicchino commented Jun 27, 2024 •

edited

Loading