feat: Configuration loop acceleration for SolidMechanicsLagrangeContact

[sohailwaziri]

[paveltomin]

@castelletto1 @matteofrigo5 @CusiniM this is actually ready to have a look

[rrsettgast]

@paveltomin @sohailwaziri
From the slide image and a quick glance at the code it is unclear exactly what is happening here. Can you explain in words the what is being done. For example, it seems you are trying to "accelerate" the evolution of the shear stress....is that the correct interpretation? What is the "accelerated variable" in this case? Is it the shear traction?...etc. A detailed discussion would be great. Perhaps at the Thursday meeting?

[castelletto1]

I tried the example suggested by @matteofrigo5 (ContactMechanics_SimpleCubes_smoke).

It happens that total_size computed in initializeAccelerationVariables is equal to 0, hence GEOS crashes here because m_x0 has zero size, as all the other allocated arrays needed for the nonlinear acceleration.

I suspect it is because in that example the SurfaceGenerator is used to split the mesh.

@sohailwaziri: please make sure initializeAccelerationVariables is called after the mesh has been split, when the SurfaceGenerator is used. Thanks!

[paveltomin]

I tried the example suggested by @matteofrigo5 (ContactMechanics_SimpleCubes_smoke).

It happens that total_size computed in initializeAccelerationVariables is equal to 0, hence GEOS crashes here because m_x0 has zero size, as all the other allocated arrays needed for the nonlinear acceleration.

I suspect it is because in that example the SurfaceGenerator is used to split the mesh.

@sohailwaziri: please make sure initializeAccelerationVariables is called after the mesh has been split, when the SurfaceGenerator is used. Thanks!

i got it fixed by moving initializeAccelerationVariables into setupSystem

without acceleration:

lagrangiancontact, number of time steps: 11
lagrangiancontact, number of successful outer loop iterations: 13
lagrangiancontact, number of successful nonlinear iterations: 94
lagrangiancontact, number of successful linear iterations: 94
lagrangiancontact, number of time step cuts: 1
lagrangiancontact, number of discarded outer loop iterations: 5
lagrangiancontact, number of discarded nonlinear iterations: 35
lagrangiancontact, number of discarded linear iterations: 35

with acceleration:

lagrangiancontact, number of time steps: 11
lagrangiancontact, number of successful outer loop iterations: 13
lagrangiancontact, number of successful nonlinear iterations: 95
lagrangiancontact, number of successful linear iterations: 95
lagrangiancontact, number of time step cuts: 1
lagrangiancontact, number of discarded outer loop iterations: 5
lagrangiancontact, number of discarded nonlinear iterations: 35
lagrangiancontact, number of discarded linear iterations: 35

so acceleration does not hurt but does not help either
maybe some parameter needs to be tuned as Sohail mentioned yesterday