Dimension: Diagnostic fixes #1220
Conversation
This patch fixes the dimensional scaling of three diagnostics: - grid_Re_Ah - grid_Re_Kh - boundary_forcing_h_tendency The fix to `grid_Re_Ah` and `grid_Re_Kh` removes the `Kh_min` and `Ah_min` parameters, which could not be dimensionally scaled and replaces them with variables in the control structure. The minimum Laplacian and biharmonic viscosites are chosed to estimate the smallest viscosity which is capable to altering the velocity. We use `spacing(1.) * L^2N / T` where `spacing()` is the smallest difference in floating point. We use the minimum harmonic grid spacing, precomputed during initialization, rather than the local value, in order to avoid additional computation and memory access during timestepping. This will changes the value of grid_Re_[AK]h in certain isolated examples of negligible viscosity, but will otherwise avoid the possibility of divide-by-zero while also producing a realizable large value at near-inviscid upper limit. The second fix to `boundary_focing_h_tendency` had its T units set but not its H units. This patch fixes the dimensionality. These are required to fix the dimensional verification errors in PR 1210.
|
Sorry I just found a bug in this, missed by the testing. I will update shortly. |
The single-line if-block for reading the DT parameter did not contain the Idt = 1. / dt update. This patch amends this bug.
Codecov Report
@@ Coverage Diff @@
## dev/gfdl #1220 +/- ##
============================================
- Coverage 46.08% 46.08% -0.01%
============================================
Files 214 224 +10
Lines 69399 70605 +1206
============================================
+ Hits 31984 32535 +551
- Misses 37415 38070 +655
Continue to review full report at Codecov.
|
|
This is required for #1210 |
There was a problem hiding this comment.
I may be wrong, but it looks to me like min_grid_sp_h2 and min_grid_sp_h4 in hor_visc_init() will vary between processors (unless a Cartesian grid is used) and that they will not be invariant to domain decomposition. If so, CS%min_grid_Ah and CS%min_grid_Kh will also not be invariant to domain decomposition. Please address this concern before these changes can be approved.
An additional `min_across_PEs` call is added to the min_grid_sp_h[24] calculations, to accommodate for parallel builds and ensure consistency across layouts.
|
Thank you @Hallberg-NOAA , I have added |
Codecov Report
@@ Coverage Diff @@
## dev/gfdl #1220 +/- ##
============================================
- Coverage 46.08% 46.08% -0.01%
============================================
Files 214 224 +10
Lines 69399 70605 +1206
============================================
+ Hits 31984 32535 +551
- Misses 37415 38070 +655
Continue to review full report at Codecov.
|
|
This PR is being tested with https://gitlab.gfdl.noaa.gov/ogrp/MOM6/-/pipelines/11386. |
|
The pipeline tests and Travis tests are all passing with this PR. It can be merged in. |
This patch fixes the dimensional scaling of three diagnostics:
The fix to
grid_Re_Ahandgrid_Re_Khremoves theKh_minandAh_minparameters, which could not be dimensionally scaled andreplaces them with variables in the control structure.
The minimum Laplacian and biharmonic viscosites are chosed to estimate
the smallest viscosity which is capable to altering the velocity. We
use
spacing(1.) * L^2N / Twherespacing()is the smallestdifference in floating point. We use the minimum harmonic grid spacing,
precomputed during initialization, rather than the local value, in order
to avoid additional computation and memory access during timestepping.
This will changes the value of grid_Re_[AK]h in certain isolated
examples of negligible viscosity, but will otherwise avoid the
possibility of divide-by-zero while also producing a realizable large
value at near-inviscid upper limit.
The second fix to
boundary_focing_h_tendencyhad its T units set butnot its H units. This patch fixes the dimensionality.
These are required to fix the dimensional verification errors in PR
1210.