Add backtracking to ElementJacobianDamper by Andreas-Lepak · Pull Request #32807 · idaholab/moose

Andreas-Lepak · 2026-04-22T19:36:03Z

Reason

ElementJacobianDamper already limits large element Jacobian changes on the displaced mesh, but
its probe path can still encounter a degenerate map when the full trial Newton update inverts an
element. In that case the solve falls back to a timestep cutback instead of trying a smaller
nonlinear update first.

This change adds an opt-in backtracking mode so the damper can reduce the probed update before FE
assembly reaches a DegenerateMap.

Design

Add optional use_backtracking, backtrack_factor, and max_backtrack_steps parameters to
ElementJacobianDamper.
Refactor the damper probe into a helper that tests a damped trial update and reports the maximum
relative JxW change.
Treat degenerate-map exceptions during the probe as a signal that the trial damping is too large,
then iteratively reduce the trial damping until the probe succeeds and the Jacobian increment is
below max_increment.
Always restore displaced node coordinates after a failed probe so the damper does not leave the
displaced mesh in a modified state.
Add a regression test covering the new backtracking path and update the damper documentation.

Impact

This is a backward-compatible enhancement to ElementJacobianDamper.

Existing input files are unchanged unless they opt into use_backtracking = true.
The new mode improves robustness for large-displacement solves that would otherwise fail probing a
nonlinear update due to temporary element inversion.
No existing APIs are removed or changed incompatibly.

Testing

Ran exception_handling.i without backtracking to confirm the existing negative-Jacobian path is
preserved.
Ran exception_handling.i with Dampers/jac/use_backtracking=true to confirm the damper now
reduces the update without emitting the negative-Jacobian error.
Ran git diff --check.

moosebuild · 2026-04-22T19:43:01Z

Job Precheck, step Clang format on b97b5c9 wanted to post the following:

Your code requires style changes.

A patch was auto generated and copied here
You can directly apply the patch by running, in the top level of your repository:

curl -s https://mooseframework.inl.gov/docs/PRs/32807/clang_format/style.patch | git apply -v

Alternatively, with your repository up to date and in the top level of your repository:

git clang-format dba0fc8d126cb8a0d6b4e8eabf8b390a09e4aa91

moosebuild · 2026-04-22T22:34:03Z

Job Documentation, step Docs: sync website on e46bda2 wanted to post the following:

View the site here

This comment will be updated on new commits.

- add opt-in backtracking parameters - retry damped trial updates after DegenerateMap - restore node coordinates after failed probes - add regression coverage and documentation refs idaholab#32806

moosebuild · 2026-04-23T18:09:40Z

Job Coverage, step Generate coverage on e46bda2 wanted to post the following:

Framework coverage

Coverage did not change

Modules coverage

Solid mechanics

	5b391d	#32807 e46bda
	Total	Total	+/-	New
Rate	85.38%	85.37%	-0.02%	100.00%
Hits	28595	28635	+40	86
Misses	4896	4909	+13	0

Diff coverage report

Full coverage report

Full coverage reports

Reports

This comment will be updated on new commits.

GiudGiud · 2026-04-30T16:19:42Z

+      "use_backtracking",
+      false,
+      "If true, iteratively cut back the probed Newton update until the displaced mesh remains "
+      "nondegenerate and the Jacobian increment stays within max_increment.");


the Jacobian update will always be within max_increment if it starts within max incremenent? How would you exceed by further cutting back?

it is for recovering from a trial that is either degenerate or still over the Jacobian-change limit.

isnt the "over the Jacobian-change limit" simply handled by scaling the update by the ratio of the limit to the current value?

GiudGiud · 2026-04-30T18:19:46Z

+
+    unsigned int has_exception = _fe_problem.hasException() ? 1 : 0;
+    _communicator.max(has_exception);
+    if (has_exception)


not sure I understand this
add a comment please? why is an exception ok for returning a damping factor

Added comment
// The exception flag will abort the nonlinear step after the damper returns, so return the
// last trial damping only to satisfy the interface.

and we know this exception is not one such that backtracking the damping would remove it?

GiudGiud · 2026-04-30T18:21:28Z

+    Real max_difference = 0.0;
+    bool valid = false;
+
+    PARALLEL_TRY { valid = probe_trial(damping, max_difference, error_message); }


valid is a boolean but this routine can be returning "_max_jacobian_diff / max_difference;" ?

probe_trial should now return bool. The lambda ends with: return !invalid_local; and !invalid_local is a boolean expression, so probe_trial(...) cannot return _max_jacobian_diff / max_difference.

GiudGiud · 2026-04-30T18:24:36Z

+    if (!valid)
+      damping *= _backtrack_factor;
+    else
+      damping =
+          std::min(damping * _backtrack_factor, damping * _max_jacobian_diff / max_difference);


let's rename valid to make it obvious it's actually a criteria for reducing the damping factor

it is now renamed trial_nondegenerate

Co-authored-by: Guillaume Giudicelli <guillaume.giudicelli@gmail.com>

GiudGiud · 2026-05-03T12:48:45Z

Test failure is unrelated
I fixed up the commit history

moosebuild · 2026-05-03T13:53:47Z

Job Test, step Results summary on e46bda2 wanted to post the following:

Framework test summary

Compared against 5b391df in job civet.inl.gov/job/3787211.

No added tests

Run time changes

Test	Base (s)	Head (s)	+/-	Base (MB)	Head (MB)
`problems/reference_residual_problem.zero_tolerance_ref`	3.97	6.00	+51.00%	147.95	132.46

Modules test summary

Compared against 5b391df in job civet.inl.gov/job/3787211.

Added tests

Test	Time (s)	Memory (MB)
`solid_mechanics/test:jacobian_damper.backtrack_negative_jacobian`	1.25	121.30

Run time changes

Test	Base (s)	Head (s)	+/-	Base (MB)	Head (MB)
`stochastic_tools/test:web_server_control.stochastic_control/batch_reset_multi`	2.33	3.85	+65.15%	247.32	252.97
`solid_mechanics/test:smeared_cracking.rz_exponential`	11.03	18.05	+63.70%	126.14	136.79
`solid_mechanics/test:rate_independent_cyclic_hardening.nonlin_isokinharden_symmetric_strain_controlled`	10.13	16.18	+59.68%	150.94	146.96
`solid_mechanics/test:rate_independent_cyclic_hardening.linear_kinharden_nonsymmetric_stress_controlled`	4.34	6.81	+56.82%	122.60	131.24
`solid_mechanics/test:combined_creep_plasticity.creepWithPlasticity`	5.24	8.18	+56.06%	121.92	120.96
`combined/test:combined_plasticity_temperature.ad_temp_dep_yield-jac`	6.55	10.17	+55.23%	173.21	179.24
`stochastic_tools/test:transfers/sampler_reporter.transfer/normal`	6.76	10.47	+54.88%	158.47	167.83
`solid_mechanics/test:beam/static.euler_finite_rot_y`	3.11	4.80	+54.22%	126.28	132.36
`solid_mechanics/test:rate_independent_cyclic_hardening.nonlin_kinharden_nonsymmetric_strain_controlled`	5.80	8.94	+54.19%	130.95	143.58
`solid_mechanics/test:rate_independent_cyclic_hardening.1D_ratcheting_nonlin_kinharden_stress_controlled`	3.73	5.72	+53.40%	116.25	120.16
`solid_mechanics/test:rate_independent_cyclic_hardening.nonlin_kinharden_symmetric_strain_controlled`	4.70	7.19	+53.05%	126.33	126.65
`solid_mechanics/test:beam/static.euler_finite_y_with_action`	3.13	4.77	+52.36%	121.64	126.71
`solid_mechanics/test:combined_creep_plasticity.combined_start_time`	5.72	8.70	+52.29%	118.07	126.05
`solid_mechanics/test:torque_reaction.disp_about_axis_axial_motion_delayed`	2.02	3.08	+52.26%	114.45	116.21
`stochastic_tools/test:transfers/sampler_reporter.transfer/distributed`	3.27	4.97	+52.13%	472.19	511.28
`solid_mechanics/test:smeared_cracking.cracking_rotation_pres_dir_z`	3.87	5.86	+51.44%	132.21	142.19
`solid_mechanics/test:smeared_cracking.cracking_rotation_pres_dir_x`	3.82	5.78	+51.35%	136.85	139.61
`solid_mechanics/test:rate_independent_cyclic_hardening.linear_kinharden_symmetric_strain_controlled`	4.83	7.30	+51.12%	144.16	132.52
`solid_mechanics/test:temperature_dependent_hardening.test`	2.17	3.27	+50.50%	121.70	120.73
`solid_mechanics/test:rate_independent_cyclic_hardening.nonlin_isoharden_symmetric_strain_controlled`	3.37	5.07	+50.33%	126.03	131.29
`solid_mechanics/test:dynamics/acceleration_bc.acceleration_bc`	3.47	5.21	+50.28%	135.66	137.12
`solid_mechanics/test:smeared_cracking.xyz`	2.82	4.24	+50.27%	129.79	137.52

Andreas-Lepak force-pushed the element-jacobian-damper-backtracking branch from b97b5c9 to 4b96447 Compare April 22, 2026 20:07

Andreas-Lepak force-pushed the element-jacobian-damper-backtracking branch from 4b96447 to 8133965 Compare April 22, 2026 23:45

Add backtracking to ElementJacobianDamper

28794bd

- add opt-in backtracking parameters - retry damped trial updates after DegenerateMap - restore node coordinates after failed probes - add regression coverage and documentation refs idaholab#32806

Andreas-Lepak force-pushed the element-jacobian-damper-backtracking branch from 8133965 to 28794bd Compare April 23, 2026 14:31

moosebuild added the PR: Failed but allowed label Apr 23, 2026

Andreas-Lepak marked this pull request as ready for review April 23, 2026 18:09

Andreas-Lepak requested review from bwspenc and dschwen as code owners April 23, 2026 18:10

GiudGiud requested changes Apr 30, 2026

View reviewed changes

moosebuild removed the PR: Failed but allowed label Apr 30, 2026

GiudGiud self-assigned this Apr 30, 2026

GiudGiud approved these changes May 1, 2026

View reviewed changes

Andreas-Lepak and others added 2 commits May 3, 2026 06:47

Fix ElementJacobianDamper review follow-ups

2977426

Co-authored-by: Guillaume Giudicelli <guillaume.giudicelli@gmail.com>

Fix MPI reduction in ElementJacobianDamper

e46bda2

GiudGiud force-pushed the element-jacobian-damper-backtracking branch from 9df72f7 to e46bda2 Compare May 3, 2026 12:48

moosebuild added the PR: Failed but allowed label May 3, 2026

Conversation

Andreas-Lepak commented Apr 22, 2026

Reason

Design

Impact

Testing

Uh oh!

moosebuild commented Apr 22, 2026

Uh oh!

moosebuild commented Apr 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

moosebuild commented Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Framework coverage

Modules coverage

Solid mechanics

Full coverage reports

Uh oh!

Uh oh!

Uh oh!

GiudGiud Apr 30, 2026 • 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!

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!

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!

GiudGiud commented May 3, 2026

Uh oh!

moosebuild commented May 3, 2026

Framework test summary

No added tests

Run time changes

Modules test summary

Added tests

Run time changes

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

moosebuild commented Apr 22, 2026 •

edited

Loading

moosebuild commented Apr 23, 2026 •

edited

Loading

GiudGiud Apr 30, 2026 •

edited

Loading