CI: fix fluctuations in collisionXZ checksums

[EZoni]

Trying to see if we can get under control the fluctuations observed in the checksums of the collisionXZ test.

[ax3l]

That parameter should be set here already:

WarpX/Regression/prepare_file_ci.py

Line 86 in 502e5ea

'warpx.do_dynamic_scheduling=0 warpx.serialize_initial_conditions=1',

[EZoni]

That parameter should be set here already:

WarpX/Regression/prepare_file_ci.py

Line 86 in 502e5ea

'warpx.do_dynamic_scheduling=0 warpx.serialize_initial_conditions=1',

That's right, thanks for pointing it out. I always forget that we set input parameters in three different places at the moment. For now I'm just keeping this PR open (now it simply increases the relative tolerance for the checksums) until the next meeting where this issue will be discussed.

[ax3l]

Since development is failing, can you set a tolerance just high enough for it to pass and we merge it?

[EZoni]

I'm afraid 1e-1 might be the one that is just high enough. At least on my computer, the test fails with 1e-2. I think this is consistent with the "10%" error that @RemiLehe mentioned offline. I could check if this is the case also on Azure, but it is for sure on my computer. This holds for both collisionXZ and collisionXYZ.

[ax3l]

OK for me for now.

@RemiLehe @JustinRayAngus what do you think, shall we use this for now?

[roelof-groenewald]

I agree this is good to do until the underlying issue with the collision tests can be sorted out.

[JustinRayAngus]

Approve. What's different between these two tests and the other tests that use Coulomb collisions?

[ax3l]

We don't understand that yet... Maybe less particles per cell or per simulation and thus worse statistics?

[ax3l]

In #5126 I see that collisionZ fluctuates a well.

./analysis_collision_1d.py  collisionZ_plt000600
TApar at 30ps error =  0.0025701770198334937
tolerance =  0.02
ERROR: Benchmark and output file checksum have different value for key [ions,particle_momentum_x]
Benchmark: [ions,particle_momentum_x] 3.425400072687143e-16
Test file: [ions,particle_momentum_x] 3.424633029669351e-16
Absolute error: 7.67e-20
Relative error: 2.24e-04
ERROR: Benchmark and output file checksum have different value for key [ions,particle_momentum_y]
Benchmark: [ions,particle_momentum_y] 3.421937133999805e-16
Test file: [ions,particle_momentum_y] 3.429026824477811e-16
Absolute error: 7.09e-19
Relative error: 2.07e-03
ERROR: Benchmark and output file checksum have different value for key [ions,particle_momentum_z]
Benchmark: [ions,particle_momentum_z] 5.522701882677923e-16
Test file: [ions,particle_momentum_z] 5.528396735566589e-16
Absolute error: 5.69e-19
Relative error: 1.03e-03
ERROR: Benchmark and output file checksum have different value for key [ions,particle_position_x]
Benchmark: [ions,particle_position_x] 7.200011611411148e+02
Test file: [ions,particle_position_x] 7.200708684755696e+02
Absolute error: 6.97e-02
Relative error: 9.68e-05

Notable differences in runners:

• Azure Pipelines 9 (newer) in PR vs. Azure Pipelines 8 on development
• CPUs: Xeon(R) Platinum 8272CL "Cascade Lake" (~2019) on PR vs. Xeon(R) CPU E5-2673 v4 "Broadwell" (~2016) on development