Feature: Extended Super Time Stepping (ExtSTS) Methods by drreynolds · Pull Request #985 · llnl/sundials

drreynolds · 2026-08-07T19:33:50Z

Updated the MRIStep module in ARKODE to support extended Super Time Stepping (ExtSTS) methods for multi-physics simulations involving parabolic components, through the MRIStepCreateExtSTS function. In addition to adding this "create" routine, this PR adds some Butcher tables and MRI coupling tables that can be used when creating ExtSTS methods -- essentially, these just correspond with embedded ImEx-MRI-GARK methods of order 2, that MRIStep did not already include.

There is one example problem that shows how these methods may be used.

The PR is relatively small, and if possible I'd like this to be included in the next release.

…om ExtSTS solver object.

…tSTS*

…h a boolean to indicate that the MRIStep is being used to implement an ExtSTS method.

…bject

…cing, thus simplifying the ExtSTS wrapper considerably. Notably, this allowed us to remove all wrappers to user-supplied functions and user-data pointers from the interface.

…TSGetSTS can return a pointer to the inner LSRKStep integrator without Python assuming ownership and causing a double free

…e-step function, instead of including entire ARKodeEvolve infrastructure. This allows an RKC or RKL failure due to an insufficient number of stages to propagate back up to MRIStep, who will limit the step size. Currently this step reduction is crude (the default reduction for a nonlinear convergence failure), but we could eventually upgrade MRIStepInnerSteppers to pass back a recommended step reduction factor when returning positive failure codes.

…d forgotten to remove it from arkode_mristep.h when removing the function).

…pace at the end of some lines; since my editor automatically removes these spaces then manual edits of the file induce a CI failure

drreynolds · 2026-08-28T17:40:04Z

I just reran the full set of tests for the ExtSTS paper, and something in the commit "Updated extSTSInnerStepper_Evolve to directly call ..." broke some of the results from the paper. I'll need to take part of this weekend to unravel (and hopefully fix) the issue.

…-step routine

…error

gardner48 · 2026-08-31T18:45:51Z


+  nb::class_<ARKodeBorrowedView>(m, "ARKodeBorrowedView")
+    .def("get", nb::overload_cast<>(&ARKodeBorrowedView::get, nb::const_),
+         nb::rv_policy::reference);


@balos1 Could you take a look at ARKodeBorrowedView and handling the void** argument in MRIStepExtSTSGetSTS?

gardner48 · 2026-08-31T20:16:48Z

+      int ark_status = ARKodeSetUserData(ark_mem, ark_mem);
+      if (ark_status != ARK_SUCCESS)
+      {
+        free(fn_table);


Suggested change

free(fn_table);

free(fn_table);

ARKodeFree(&ark_mem);

gardner48 · 2026-08-31T20:24:36Z

+    .value("ARKODE_MRI_GARK_EXP_ARS222", ARKODE_MRI_GARK_EXP_ARS222, "")
+    .value("ARKODE_IMEX_MRI_GARK_ARS222", ARKODE_IMEX_MRI_GARK_ARS222, "")
+    .value("ARKODE_MRI_GARK_EXP_GIRALDO2", ARKODE_MRI_GARK_EXP_GIRALDO2, "")
+    .value("ARKODE_IMEX_MRI_GARK_GIRALDO2", ARKODE_IMEX_MRI_GARK_GIRALDO2, "")


Is ARKODE_IMEX_MRI_GARK_GIRALDO2 different from ARKODE_IMEX_MRI_GARK_ARK2? Similarly, is ARKODE_IMEX_MRI_GARK_ASCHER_ARK2 the same as ARKODE_IMEX_MRI_GARK_ARS222? I think these might be using the same base tables with different names.

Like ARS, maybe we should use GKC for Giraldo, Kelly, and Constantinescu in the names

gardner48 · 2026-08-31T20:37:02Z

    1>(&arkode_user_supplied_fn_table::mristep_fsi, t, y, ydot, user_data);
 }

+inline int mristep_domeig_wrapper(sunrealtype t, N_Vector y, N_Vector fn,


I haven't gotten to the source updates yet, but want to leave a note to double check if this wrapper is still needed. From the PR comments, it sounded like this might not be needed any more.

gardner48 · 2026-09-01T03:36:26Z

+#. *Set vector of initial values*
+
+#. Create the MRIStep object for ExtSTS methods by calling
+   :c:func:`MRIStepExtSTSCreate`.


Maybe this should be MRIStepCreateExtSTS (and similarly MRIStepReInitExtSTS) to align with the LSRKStepCreate<X> (and LSRKStepReInit<X>) pattern

gardner48 · 2026-09-01T06:21:18Z

+  /* Call the user-supplied pre-step function (if supplied) */
+  ark_mem->tcur = ark_mem->tn;
+  if (ark_mem->ensure_ycur) { N_VScale(ONE, ark_mem->yn, ark_mem->ycur); }
+  if (retval == ARK_SUCCESS && ark_mem->PreStepFn)


retval is always ARK_SUCCESS at this point

Suggested change

if (retval == ARK_SUCCESS && ark_mem->PreStepFn)

if (ark_mem->PreStepFn)

gardner48 · 2026-09-01T06:26:20Z

+    retval = ark_mem->PreStepFn(ark_mem->tcur, ark_mem->ycur, ark_mem->nst, 1,
+                                ark_mem->user_data);
+    /* Preserve the failure flag but continue to disable forcing below. */
+    if (retval != 0) { retval = ARK_PRESTEPFN_FAIL; }


Is there a reason to not return immediately on an error here?

Suggested change

if (retval != 0) { retval = ARK_PRESTEPFN_FAIL; }

if (retval != 0) { return ARK_PRESTEPFN_FAIL; }

gardner48 · 2026-09-01T06:29:06Z

+  /* Take a single inner STS step */
+  if (retval == ARK_SUCCESS)
+  {
+    /* Let ExtSTS translate this recoverable failure after forcing is disabled. */
+    lsrkstep_mem->suppress_max_stage_limit_error = SUNTRUE;
+    retval = ark_mem->step(ark_mem, &dsm, &nflag);
+    lsrkstep_mem->suppress_max_stage_limit_error = SUNFALSE;
+    ark_mem->nst_attempts++;
+  }


If the prestep error is returned above, the retval check here can be removed. Also nst_attempts is incremented before the step is attempted in evolve.

Suggested change

/* Take a single inner STS step */

if (retval == ARK_SUCCESS)

{

/* Let ExtSTS translate this recoverable failure after forcing is disabled. */

lsrkstep_mem->suppress_max_stage_limit_error = SUNTRUE;

retval = ark_mem->step(ark_mem, &dsm, &nflag);

lsrkstep_mem->suppress_max_stage_limit_error = SUNFALSE;

ark_mem->nst_attempts++;

}

/* Take a single inner STS step */

/* Let ExtSTS translate this recoverable failure after forcing is disabled. */

ark_mem->nst_attempts++;

lsrkstep_mem->suppress_max_stage_limit_error = SUNTRUE;

retval = ark_mem->step(ark_mem, &dsm, &nflag);

lsrkstep_mem->suppress_max_stage_limit_error = SUNFALSE;

gardner48 · 2026-09-01T06:31:01Z

+    /* Preserve the failure flag but continue to disable forcing below. */
+    if (retval != 0) { retval = ARK_PRESTEPFN_FAIL; }
+  }
+


Probably need SUNLogInfo logging statements similar to ARKodeEvolve to parse the logging output correctly

gardner48 · 2026-09-01T06:38:40Z

+  return retval;
+}
+
+int extSTSInnerStepper_Free(MRIStepInnerStepper* sts_mem)


I don't see that this is called or attached anywhere

maggul and others added 30 commits December 15, 2025 13:43

power updates

290b9b8

preprocess up to a reltol

ee6479e

LSRKStep updates for complex eig support

2a9c2a6

revision

90123d0

Merge branch 'develop' into SUNDomEigEstimator_Upgrades

4712caf

swig and formatting

e23ef54

Merge branch 'develop' into SUNDomEigEstimator_Upgrades

919e839

deactivate the py interface check temporarily

2c2d8e6

removed unused zI

f93ab67

fixed ci issues

59f3f1c

zR to zRabs fix and output files

5fc27eb

format check

585e714

Merge branch 'develop' into SUNDomEigEstimator_Upgrades

96c647d

q_prev destroy

8899834

merge

ce4c6b6

updated answers with the correct CI test results

ccb80c8

Merge branch 'develop' into SUNDomEigEstimator_Upgrades

923fc66

Merge branch 'develop' into SUNDomEigEstimator_Upgrades

ee33ceb

revision

da13267

docs and stand alone module -- need a bug fix

77a99ec

bug fix

1ca8905

revert the workflow change

c707ddf

ci trigger step-1

5b64126

Merge branch 'develop' into SUNDomEigEstimator_Upgrades

976bf94

ci trigger step-2

d60f793

formatting

1fb0192

swig and litgen

d088d11

Merge branch 'develop' into SUNDomEigEstimator_Upgrades

b0f189b

Merge branch 'develop' into SUNDomEigEstimator_Upgrades

0478682

revision

91913db

drreynolds added 23 commits August 24, 2026 11:25

Removed duplicated tables

140b11c

Working my way through 'simpler' PR comments

6adebfb

Updated API for accessor function to retrieve LSRKStep sub-stepper fr…

156969c

…om ExtSTS solver object.

Updated ExtSTS function names to all have the same pattern: MRIStepEx…

77a68db

…tSTS*

Addressed another batch of simpler PR comments

b3788f1

Documentation updates in line with PR revisions

31d6bb2

Replaced pointer to extSTSInnerStepper object in ARKodeMRIStepMem wit…

4a3115b

…h a boolean to indicate that the MRIStep is being used to implement an ExtSTS method.

Renamed ExtSTS method indicator flag

7faf49f

Removed the MRIStepInnerStepper pointer from the extSTSInnerStepper o…

e16e7f2

…bject

Restructured ExtSTS solver so that LSRKStep implements polynomial for…

b0159d2

…cing, thus simplifying the ExtSTS wrapper considerably. Notably, this allowed us to remove all wrappers to user-supplied functions and user-data pointers from the interface.

Merged with develop

461ec9c

Formatting

6d4c6d8

Minor fixes

c13fd3d

Removed ExtSTS-specific routines from generated sundials4py source

0296a08

Formatting

9f3869d

Added ARKodeBorrowedView to sundials4py interface so that MRIStepExtS…

77ca6ac

…TSGetSTS can return a pointer to the inner LSRKStep integrator without Python assuming ownership and causing a double free

Formatting

d2d8a15

Swig and Python bindings

e782a47

Removed MRIStepExtSTSSetDomEigFn from Python and Fortran bindings (I'…

5f46706

…d forgotten to remove it from arkode_mristep.h when removing the function).

Added lsrkStep_f_forcing definition to arkode_lsrkstep_impl.h

7a3ec2f

Swig regenerate (apparently the generator includes extraneous white s…

6a0fe8b

…pace at the end of some lines; since my editor automatically removes these spaces then manual edits of the file induce a CI failure

Updated answer files due to PR revisions

e86eee1

drreynolds and others added 5 commits August 28, 2026 17:31

Fixed bug introduced in ExtSTS rewrite to directly call LSRKStep take…

a4add42

…-step routine

Fixed caption that was somehow causing a Sphinx latexpdf compilation …

549bcf0

…error

Merge branch 'develop' into feature/extsts

f3a3f28

Merged with develop

b49eb53

Fixed answer files due to merge issues

c8f9709

gardner48 requested changes Sep 1, 2026

View reviewed changes

Navigation Menu

Uh oh!

Feature: Extended Super Time Stepping (ExtSTS) Methods - #985

Feature: Extended Super Time Stepping (ExtSTS) Methods#985
drreynolds wants to merge 164 commits into
developfrom
feature/extsts

drreynolds commented Aug 7, 2026 •

edited

Loading

Uh oh!

drreynolds commented Aug 28, 2026 •

edited

Loading

Uh oh!

gardner48 Aug 31, 2026

Uh oh!

gardner48 Aug 31, 2026

Uh oh!

gardner48 Aug 31, 2026

Uh oh!

gardner48 Aug 31, 2026

Uh oh!

gardner48 Sep 1, 2026

Uh oh!

gardner48 Sep 1, 2026

Uh oh!

gardner48 Sep 1, 2026

Uh oh!

gardner48 Sep 1, 2026

Uh oh!

gardner48 Sep 1, 2026

Uh oh!

gardner48 Sep 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

	if (retval == ARK_SUCCESS && ark_mem->PreStepFn)
	if (ark_mem->PreStepFn)

	if (retval != 0) { retval = ARK_PRESTEPFN_FAIL; }
	if (retval != 0) { return ARK_PRESTEPFN_FAIL; }

Uh oh!

Conversation

drreynolds commented Aug 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

drreynolds commented Aug 28, 2026 • 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!

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

drreynolds commented Aug 7, 2026 •

edited

Loading

drreynolds commented Aug 28, 2026 •

edited

Loading