Add vertical mixing tendencies by hyungyukang · Pull Request #410 · E3SM-Project/Omega

hyungyukang · 2026-05-18T13:28:21Z

This PR adds the velocity and tracer vertical mixing tendency terms. The implicit vertical mixing is solved by using TriDiagDiffSolver and is applied once at the end of each time step.

Implemented implicit vertical mixing using TriDiagDiffSolver.
Added VertMixImplicit to VertMix.
Applied vertical mixing once at the end of each time step.
Note: Tangential velocity, which is required by ComputeGradRichardsonNum in VertMix, is temporarily handled in VertMix.

This PR includes #352 for testing purposes.

Checklist

hyungyukang · 2026-05-18T14:26:24Z

Testing

CTest unit tests

Machine: frontier
Compiler: craygnu, craygnu-mphipcc
Build type: Release
Result: All tests passed

Polaris `omega_pr` suite

Baseline workdir: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Polaris/work/pr_suite/260518_09_vMix_PR/baseline_omega
Baseline build: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Polaris/polaris_main/e3sm_submodules/Omega
PR build: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Omega_hyun_vmixHook_2
PR workdir: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Polaris/work/pr_suite/260518_09_vMix_PR/branch_omega
Machine: frontier
Partition: batch
Compiler: craygnu
Build type: Release
Log: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Polaris/work/pr_suite/260518_09_vMix_PR/branch_omega/polaris_omega_pr.o4609310
Result: All tests passed

hyungyukang · 2026-05-18T14:57:15Z

Tested this branch using ocean/column/vmix_stable from E3SM-Project/polaris#435 and compared with MPAS-O.

Here are results of vmix_stable/forward:

Temperature at day 10:

Normal velocity at day 10:

hyungyukang · 2026-05-19T16:18:07Z

@katsmith133 ,
Here're figures of VertVisc and VertDiff (vertViscTopOfCell and vertDiffTopOfCell in MPAS-O, respectively):

VertVisc at day 10:

VertDiff at day 10

I will retest your branch with this test case since you've updated it.

katsmith133 · 2026-05-19T19:00:42Z

@katsmith133 , Here're figures of VertVisc and VertDiff (vertViscTopOfCell and vertDiffTopOfCell in MPAS-O, respectively):
...
I will retest your branch with this test case since you've updated it.

@hyungyukang, thanks for the plots! I doubt the updates I just made in #352 will result in any difference here, but doesn't hurt to check. Let me pull down this branch and take a look at things and see if anything pops out to me.

hyungyukang · 2026-06-04T17:15:31Z

Test results from column/vmix_stable/forward_no_vadv in Polaris is posted here: #352 (comment)

hyungyukang · 2026-06-04T17:21:46Z

I fixed the bugs related to non-flat bottom topography and ran the overflow test to compare Omega with MPAS-O. MPAS-O was slightly modified to exclude horizontal FCT, since it has not yet been implemented in Omega.

The results below show temperature at 6 hours.

cbegeman · 2026-06-18T20:28:54Z

@hyungyukang The PR doesn't include bottom drag in the implicit mixing, correct? So we will continue to use explicit bottom drag with implicit vmix?

hyungyukang · 2026-06-18T22:17:48Z

@hyungyukang The PR doesn't include bottom drag in the implicit mixing, correct? So we will continue to use explicit bottom drag with implicit vmix?

@cbegeman , Yes, the implicit bottom drag is not included in this PR. I plan to open a separate PR for it after this one is merged.

katsmith133

My comments aren't super major, except for whether we want to split the vertical mixing stuff up in some way in anticipation of having multiple parameterizations.

katsmith133 · 2026-06-24T17:46:51Z

+   GradRichNumSmoothed = Array2DReal("GradRichNumSmoothed", Mesh->NCellsSize,
+                                     VCoord->NVertLayersP1);
+
+   // TODO: Temporary handling of TangentialVelocity


Is this temporary handling to be addressed in a later PR?

Yes.
I haven’t identified an appropriate location for it yet, but I believe we will have a suitable place for those diagnostic computations by the time the split time stepper is implemented, if not sooner.

katsmith133 · 2026-06-24T17:52:37Z


 } // end defineIOFields

+// Apply implicit velocity vertical mixing


In MPAS-O we have several files for the vertical mixing. One that handles the below calculations and several that do the computation of the coefficients (either from KPP or something else). Do we want to follow that design choice here in Omega? I feel like it might be good, as we add more mixing parameterizations, to have them split off more? Thoughts @vanroekel?

Yes I think so. This is what I'm doing in the forthcoming KPP PR. It's its own class in its own file.

katsmith133 · 2026-06-25T20:58:47Z

+             0.5_Real * (VertVisc(JCell0, K + 1) + VertVisc(JCell1, K + 1)) /
+             (LocRhoSw * SpecVolEdgeTop);
+
+         G = DT * ViscAlphaEdgeTop / PseudoThickEdgeTop;


Do we want more descriptive names for these variables (D, H, and X)?

I tried to follow the naming conventions used in TriDiagDiffSolver:

https://docs.e3sm.org/Omega/Omega/devGuide/TridiagonalSolvers.html#diffusion-type-tridiagonal-system

https://docs.e3sm.org/Omega/Omega/design/TridiagonalSolver.html#scratch-data-structs

I will consider whether more descriptive names would be appropriate. Otherwise, I can add more detailed comments to clarify the code.

I added more detailed inline comments to VelVertMixSetup and TracerVertMixSetup to explain the tridiagonal diffusion system and the roles of the G, H, and X coefficients.

cbegeman · 2026-06-28T22:05:04Z

+             (PseudoThickEdgeK + PseudoThickEdgeKp1);
+
+         const Real ViscAlphaEdgeTop =
+             0.5_Real * (VertVisc(JCell0, K + 1) + VertVisc(JCell1, K + 1)) /


Suggested change

0.5_Real * (VertVisc(JCell0, K + 1) + VertVisc(JCell1, K + 1)) /

0.5_Real * (VertVisc(JCell0, K) + VertVisc(JCell1, K)) /

I might be misremembering the indexing of the NVertLayersP1 variables. Would we be starting from K = MinLayerEdgeBot = 0 and then K + 1 corresponds to the VertVisc at the bottom of the top level? Should the index be changed or the name be changed to ViscAlphaEdgeBot?

Yes, that’s correct. The code is written for the K loop ranging from MinLayerEdgeBot to MaxLayerEdgeTop. I double checked both the equation and the implementation, and the K + 1 index is correct. As you suggested, ViscAlphaEdgeTop should be renamed to ViscAlphaEdgeBot to reflect its actual location (other variables as well).

If I use K instead of K + 1 for VertVisc, the result differs from the Omega and MPAS-O results shown above:

Thanks for checking!

vanroekel · 2026-06-30T04:22:59Z

+      VertMix *VertMixInstance = VertMix::getInstance();
+
+      if (!EosInstance) {
+         LOG_WARN("Eos has not been initialized. Skipping calculation of "


Do we only want a warning here? Would it make more sense to do this check once at init and fail the model if this happens?

Thanks for catching that. That makes much more sense.

I updated the code so that the VertMix tendency is validated once during initialization in Tendencies.cpp rather than at every time step. Also, the model will now abort if vertical mixing tendencies are enabled without both Eos and VertMix being initialized.

https://github.com/hyungyukang/Omega/blob/b44936d15950a3b4b10b450f649d6c70d325a8ff/components/omega/src/ocn/Tendencies.cpp#L335-L347

vanroekel · 2026-06-30T04:24:40Z

+
+      if (!EosInstance) {
+         LOG_WARN("Eos has not been initialized. Skipping calculation of "
+                  "PresGradZ tendency");


should this be VertMix?

The code has been removed. VertMix tendency now checks Eos only during initialization.

vanroekel · 2026-06-30T04:24:55Z

+      VertMix *VertMixInstance = VertMix::getInstance();
+
+      if (!EosInstance) {
+         LOG_WARN("Eos has not been initialized. Skipping calculation of "


again i think an error would be better.

The code has been removed, and the model will now abort if vertical mixing tendencies are enabled without both Eos and VertMix being initialized.

vanroekel · 2026-06-30T04:25:12Z

+                  "PresGradZ tendency");
+      } else if (!VertMixInstance) {
+         LOG_WARN("VertMix has not been initialized. Skipping calculation of "
+                  "VelVertMix tendency");


TracerVertMix?

The code has been removed. VertMix tendency now checks VertMix only during initialization.

- Implemented implicit vertical mixing using TriDiagDiffSolver. - Added applyVertMixImplicit to VertMix. - Applied vertical mixing once at the end of each time step. - Note: Tangential velocity is temporarily handled in VertMix.

Renaming cell-interface variables from 'Top' to 'Bot'.

- Validate the VertMix tendency once at initialization instead of during each time step. - Abort the model if vertical mixing tendencies are enabled but Eos or VertMix has not been initialized.

- Improve the inline documentation in VertMixSetup

hyungyukang · 2026-07-01T17:48:30Z

I ran into some CTest issues after rebasing, but they have been resolved. I am now running the Polaris Omega test suite.

hyungyukang · 2026-07-01T18:28:15Z

Testing

CTest unit tests

Machine: Frontier
Compiler: craygnu, craygnu-mphipcc
Build type: Release
Result: All tests passed

Polaris `omega_pr` suite

Baseline workdir: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Polaris/work/pr_suite/260701_14_vMix_PR/baseline_omega
Baseline build: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Polaris/polaris_main_260701/e3sm_submodules/Omega
PR build: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Omega_hyun_vmixHook_8
PR workdir: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Polaris/work/pr_suite/260701_14_vMix_PR/branch_omega
Machine: frontier
Partition: batch
Compiler: craygnu
Build type: Release
Log: /lustre/orion/cli115/proj-shared/hgkang/E3SM/OMEGA/Polaris/work/pr_suite/260701_14_vMix_PR/branch_omega/polaris_omega_pr.o4927573
Result: All tests passed

vanroekel

Approving based on visual inspection, developer testing, and my own testing with the single column test case.

hyungyukang force-pushed the omega/vmix-hookup branch 5 times, most recently from 7503bfb to 72ab187 Compare May 27, 2026 13:54

hyungyukang mentioned this pull request May 28, 2026

Change shear mixing to LMD94 formulation and add ri smoothing #352

Merged

17 tasks

hyungyukang force-pushed the omega/vmix-hookup branch 3 times, most recently from fd950d6 to 5abe72c Compare June 2, 2026 01:07

katsmith133 self-requested a review June 3, 2026 15:35

hyungyukang force-pushed the omega/vmix-hookup branch from f280be8 to d272fbe Compare June 5, 2026 14:34

sbrus89 requested a review from mwarusz June 17, 2026 15:18

hyungyukang force-pushed the omega/vmix-hookup branch from d272fbe to f2a132f Compare June 18, 2026 18:56

cbegeman self-requested a review June 22, 2026 15:15

hyungyukang force-pushed the omega/vmix-hookup branch from f2a132f to abbc5c2 Compare June 25, 2026 19:43

katsmith133 reviewed Jun 25, 2026

View reviewed changes

cbegeman reviewed Jun 28, 2026

View reviewed changes

vanroekel reviewed Jun 30, 2026

View reviewed changes

katsmith133 mentioned this pull request Jun 30, 2026

Re-organization of VertMix components #450

Open

katsmith133 and others added 18 commits July 1, 2026 09:40

changes shear mixing to LMD94 formulation and adds ri smoothing

d55b4e4

changes gradrichnum, vertvisc, and vertdiff to nvertlayersp1 dim

cc10103

Add ALE projection time intervals to time steppers

aa2ce7c

Implement vertical mixing tendency

fdd426f

- Implemented implicit vertical mixing using TriDiagDiffSolver. - Added applyVertMixImplicit to VertMix. - Applied vertical mixing once at the end of each time step. - Note: Tangential velocity is temporarily handled in VertMix.

Update VertMixImplicit code and linting

2589d47

Update codes and add comments

d730ab4

Fix linting issues

5e932a6

Change LayerThick to PseudoThick in VertMix tend

cd32dc9

Fix bugs after rebase

a3a6f2a

Fix tridiag policy change in applyVertMix

45fa175

Fix linting issue

fb349bc

Fix bugs after rebase

ce54fde

Fix GPU compile error

9b82c47

Fix linting issue

79c0a77

Fix bugs for non-flat bottom topography and minor updates

5ea6e16

Address reviewers’ comments

cdfbcbf

Renaming cell-interface variables from 'Top' to 'Bot'.

Address reviewers’ comments

a14740a

- Validate the VertMix tendency once at initialization instead of during each time step. - Abort the model if vertical mixing tendencies are enabled but Eos or VertMix has not been initialized.

Address reviewers’ comments

31bac0c

- Improve the inline documentation in VertMixSetup

hyungyukang force-pushed the omega/vmix-hookup branch from add9968 to 31bac0c Compare July 1, 2026 13:42

Resolve CTest issues after rebasing

102dada

vanroekel approved these changes Jul 1, 2026

View reviewed changes


		} // end defineIOFields

		// Apply implicit velocity vertical mixing

	0.5_Real * (VertVisc(JCell0, K + 1) + VertVisc(JCell1, K + 1)) /
	0.5_Real * (VertVisc(JCell0, K) + VertVisc(JCell1, K)) /

Uh oh!

Conversation

hyungyukang commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hyungyukang commented May 18, 2026

Testing

CTest unit tests

Polaris omega_pr suite

Uh oh!

hyungyukang commented May 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hyungyukang commented May 19, 2026

Uh oh!

katsmith133 commented May 19, 2026

Uh oh!

hyungyukang commented Jun 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hyungyukang commented Jun 4, 2026

Uh oh!

cbegeman commented Jun 18, 2026

Uh oh!

hyungyukang commented Jun 18, 2026

Uh oh!

katsmith133 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!

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!

hyungyukang Jun 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!

hyungyukang Jun 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!

hyungyukang Jun 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

hyungyukang commented Jul 1, 2026

Uh oh!

hyungyukang commented Jul 1, 2026

Testing

Polaris omega_pr suite

Uh oh!

vanroekel left a comment

Choose a reason for hiding this comment

Uh oh!

hyungyukang commented May 18, 2026 •

edited

Loading

Polaris `omega_pr` suite

hyungyukang commented May 18, 2026 •

edited

Loading

hyungyukang commented Jun 4, 2026 •

edited

Loading

hyungyukang Jun 30, 2026 •

edited

Loading

hyungyukang Jun 30, 2026 •

edited

Loading

hyungyukang Jun 30, 2026 •

edited

Loading

Polaris `omega_pr` suite