LHS of linear elastic material changes in newton raphson scheme in geomechanics application

[aronnoordam]

Either a bug or at least unexpected behaviour

In the geomechanics application, when running a dynamic calculation with newton raphson strategy and newmark scheme, the left hand side of the solution changes between solutionsteps. Currently the calculation can only be performed when the LHS matrix is rebuild every solutionstep; when only rebuilding the RHS, the solution does not converge. (in other words, using rebuildlevel = 0, does not work, rebuildlevel =1 or 2 works)

Since the material is linear elastic I expect the LHS to be constant

[loumalouomega]

Huuum, does it happen in static calculation?, maybe the problem is due to the dynamic LHS contributions.

[loumalouomega]

What elements are you using?, SmallDisplacementElement and TotalLagrangianElement integrates in the reference configuration

[Vahid-Galavi]

@aronnoordam in dynamic simulations, you have dt in LHS. Are you using the same dt in all steps? Do you have other contributions which are not constant in the time steps, damping matrix, permeability, ...?

[aronnoordam]

@loumalouomega , I will check if it also happens in static situation. I'm using SmallDisplacementElement

@Vahid-Galavi I am using a constant dt. The rest should also be constant. I have performed a drained calculation, using the GeoLinearElasticPlaneStrain2DLaw

[loumalouomega]

Okay, tell me if there are changes.

[aronnoordam]

the problem stays in quasi-static calculations

[Vahid-Galavi]

@aronnoordam There is no element with this name "SmallDisplacementElement" in GeomechanicsApp.

[aronnoordam]

i narrowed the problem down to the calling of:

BaseType::mpLinearSystemSolver->Solve(A, Dx, b);

line 476 in residualbased_block_builder_and_solver.h

[aronnoordam]

@aronnoordam There is no element with this name "SmallDisplacementElement" in GeomechanicsApp.

I've used UPwSmallStrainElement2D3N

[Vahid-Galavi]

@aronnoordam There is no element with this name "SmallDisplacementElement" in GeomechanicsApp.

I've used UPwSmallStrainElement2D3N

Do you have the same behaviour with an equivalent element from StructuralmechanicsApp? Are you using viscous boundaries?

[aronnoordam]

I did not check in other applications. I only have a line load and a non displacement boundary. But I expect indeed that its not an error in geomechanics, rather in the linear solver or my linear solver settings. Do you expect that one of the linear solver settings can change the LHS ? If so, can i change the settings such that it doesnt happen?

[aronnoordam]

I've found out that, if the linear solver scaling setting is set on false, the calculation runs correctly. When setting the scaling parameter on true, the LHS changes per iteration. @loumalouomega is this expected?

[aronnoordam]

@Vahid-Galavi , are there reasons to no set the linear solver scaling parameter in the Geomechanics application on False?

[Vahid-Galavi]

@Vahid-Galavi , are there reasons to no set the linear solver scaling parameter in the Geomechanics application on False?

This is to scale stiffness matrix when you have components with different stiffness. In your case, you do not need to set scaling to true. But it is still strange that this option changes the matrix differently per step, although the input stiffness is always the same.

[aronnoordam]

okay thanks, I will close this issue and make one for the core

[aronnoordam]

https://github.com/KratosMultiphysics/Kratos/issues/10589