Closed TotoGaz closed 3 years ago
I think this is fine. As you said, this is caused by a strong absolute tolerance.
I do agree with you @WuHuiLLNL, I do not worry too much about this. It was also a way to report and track the issue.
The python warning can be ignored, it's handled in the script. As for the tolerances those look like the same tests that are failing on our GPU system (even when run CPU only). On our system some of the diffs are pretty inconsequential like the one you posted above but others are huge. I believe @rrsettgast is looking into it.
Problem is still here, for instance for beam_bending :
********************************************************************************
Error: /Problem/domain/MeshBodies/mesh1/Level0/ElementRegions/elementRegionsGroup/Region2/elementSubRegions/cb1/shale/stress
Arrays of types float64 and float64 have 122880 values of which 48166 fail both the relative and absolute tests.
Max absolute difference is at index (28, 2, 0): value = -604471801.6809957, base_value = -604471801.6808847
Max relative difference is at index (120, 2, 3): value = -1.862645149230957e-08, base_value = 0.0
Statistics of the q values greater than 1.0 defined by absolute tolerance: N = 40043
max = 10490.41748046875, mean = 649.5192945775137, std = 1021.0417646132936
Statistics of the q values greater than 1.0 defined by relative tolerance: N = 8123
max = 1964.881363738629, mean = 8.174558951984588, std = 59.06163032902932
********************************************************************************
********************************************************************************
Error: /Problem/domain/MeshBodies/mesh1/Level0/ElementRegions/elementRegionsGroup/Region2/elementSubRegions/cb1/stress
Arrays of types float64 and float64 have 122880 values of which 48166 fail both the relative and absolute tests.
Max absolute difference is at index (28, 2, 0): value = -604471801.6809957, base_value = -604471801.6808847
Max relative difference is at index (120, 2, 3): value = -1.862645149230957e-08, base_value = 0.0
Statistics of the q values greater than 1.0 defined by absolute tolerance: N = 40043
max = 10490.41748046875, mean = 649.5192945775137, std = 1021.0417646132936
Statistics of the q values greater than 1.0 defined by relative tolerance: N = 8123
max = 1964.881363738629, mean = 8.174558951984588, std = 59.06163032902932
********************************************************************************
I am trying to work on this issue, but some differences are not negligible on develop, e.g. for compositional_multiphase_wells_1d_01 :
Error: /Problem/domain/MeshBodies/mesh1/Level0/ElementRegions/elementRegionsGroup/Region1/elementSubRegions/cb1/dPhaseDensity_dGlobalCompFraction
Arrays of types float64 and float64 have 40 values of which 1 fail both the relative and absolute tests.
Max absolute difference is at index (2, 0, 0, 3): value = 3360.6991367299497, base_value = 3360.33293114939
Max relative difference is at index (2, 0, 1, 0): value = -0.0012330285399880093, base_value = -0.0018495428099812644
Statistics of the q values greater than 1.0 defined by absolute tolerance: N = 0
Statistics of the q values greater than 1.0 defined by relative tolerance: N = 1
max = 2.3148148232806087, mean = 2.3148148232806087, std = 0.0
It seems to happen for all the dPhaseDensity_dGlobalCompFraction
, even in other tests.
I don't want to have a tolerance of 1e-1.
Thoughts ?
@AntoineMazuyer do you get a diff even after rebasing and running the tests again?
@corbett5 I don't want to rebase it because it will just transfer the problem to the next one.
btw, I noticed for this specific test :
76 Attempt: 0, Newton: 0, R = 7.53007e-05
77 WARNING: Rachford-Rice Newton reached max number of iterations
78 WARNING: Rachford-Rice Newton reached max number of iterations
79 WARNING: Rachford-Rice Newton reached max number of iterations
80 WARNING: Rachford-Rice Newton reached max number of iterations
81 WARNING: Rachford-Rice Newton reached max number of iterations
82 WARNING: Rachford-Rice Newton reached max number of iterations
83 WARNING: Rachford-Rice Newton reached max number of iterations
Possible overlap with #840 and #766
We did build GEOSX with gcc 7.3.0 qnd openmpi 2.1.5. We were able to run the integrated tests on a slurm cluster.
The following tests were OK:
PASSED : 63 ( 10x10x10_LaplaceFEM_01 10x10x10_LaplaceFEM_08 10x10x10_LaplaceFEM_27 50x10x5_LaplaceFEM_01 50x10x5_LaplaceFEM_08 50x10x5_LaplaceFEM_18 DryFrac_StaticPenny_PrismElem_08 deadoil_3ph_corey_1d_03 deadoil_3ph_baker_1d_01 deadoil_3ph_baker_1d_02 deadoil_3ph_baker_1d_03 deadoil_3ph_staircase_3d_01 deadoil_3ph_staircase_3d_08 compositional_multiphase_wells_1d_01 compositional_multiphase_wells_1d_02 compositional_multiphase_wells_2d_01 compositional_multiphase_wells_2d_04 dead_oil_wells_2d_01 dead_oil_wells_2d_04 staircase_compositional_multiphase_wells_3d_01 staircase_compositional_multiphase_wells_3d_08 sedov_1 sedov_8 sedov_27 sourceFlux_1d_01 sourceFlux_1d_02 sourceFlux_1d_03 compressible_1d_01 compressible_1d_02 compressible_1d_03 incompressible_1d_01 incompressible_1d_02 incompressible_1d_03 sourceFlux_2d_01 sourceFlux_2d_04 sourceFlux_2d_09 fractureFlow_2d_01 fractureFlow_2d_02 fractureFlow_2d_04 fractureJunctionFlow_2d_01 fractureMatrixFlow_2d_01 fractureMatrixFlow_2d_04 fractureMatrixFlow_2d_09 staircase_3d_01 staircase_3d_08 staircase_3d_27 compressible_single_phase_wells_1d_01 compressible_single_phase_wells_1d_02 incompressible_single_phase_wells_2d_01 incompressible_single_phase_wells_2d_04 staircase_single_phase_wells_3d_01 staircase_single_phase_wells_3d_08 SSLE-sedov_01 SSLE-sedov_08 SSLE-sedov_27 4comp_2ph_1d_01 4comp_2ph_1d_02 4comp_2ph_1d_03 4comp_2ph_cap_1d_01 4comp_2ph_cap_1d_02 4comp_2ph_cap_1d_03 deadoil_3ph_corey_1d_01 deadoil_3ph_corey_1d_02 )
But the following ones were not:
For most of them (i.e. not SurfaceGenerator/SurfaceGenerator_08, SurfaceGenerator/DryFrac_ThreeNodesPinched_HorizontalFrac_08 and SurfaceGenerator/DryFrac_ThreeNodesPinched_SlantFrac_08, see other issue https://github.com/GEOSX/GEOSX/issues/629), it appears that the convergence required by the test is too strong.
Here is an example
If the absolute tolerance is modified from 1e-08 to 1e-07, then comparison is OK. Except for the 3 other cases mentioned above, we are in the same case/pattern.
What is your point of view on this?
Thanks,
PS: Do you know about the python RuntimeWarning?