Different physics solved in different "growing/merging" domains

[keassyguang]

Check these boxes if you have followed the posting rules.

• Q&A General is the most appropriate section for my question
• I have consulted the posting Guidelines on the Discussions front page
• I have searched the Discussions forum and my question has not been asked before
• I have searched the MOOSE website and the documentation does not answer my question
• I have formatted my post following the posting guidelines (screenshots as a last resort, triple back quotes around pasted text)

Question

Hi everyone,

I am dealing with a photopolymerization based printing process simulation to predict the distortions of the printed part. As shown below, a capital letter "H" is selected as the print part, three physics are considered, namely the chemical, thermal, and structural. Is it possible to solve different physics in different domains? For example, chemical is solved in regions 2 and 3; thermal is solved in regions 1, 2, and 3; structural is solved in regions 2 and 3. There could be potential couplings between different physics. The main concerns are as follow:

1. The moving/merging domains
2. The moving/growing node sets

For the first, the number of domains could change over time. Domains could be merged into one or more. Is it possible to use MOOSE's built-in mesh generator to achieve this, or to use a third party software to pre-define all the blocks? For these three physics, the blocks that variables to be solved is changing during the simulation due to the geometry. In addition, the variables may only be solved within the last 50 or 100 layers, while the variables for the previous layers are treated as constant.

For the second one, I need use a lot of node sets to define the boundary conditions or initializations for the chemical and structural physics, and these node sets could change over time.

Any possible solutions, problems, challenges related to the implementation are welcome. Many thanks.

[GiudGiud]

Hello

s it possible to solve different physics in different domains

Certainly. You just use block restriction on each variable.

For the first, the number of domains could change over time. Domains could be merged into one or more. Is it possible to use MOOSE's built-in mesh generator to achieve this, or to use a third party software to pre-define all the blocks?

The easiest way so far to do this has been to have every element you need, already in the mesh, just assigned to an inactive block. Then when the equations are supposed to be solved on these elements, you change their subdomain to one of the active blocks.

Note that you can pre-declare blocks that do not exist on the mesh currnetly but will later in the simulation in the Mesh block

[keassyguang]

May I know how the MOOSE deals with boundary conditions (e.g. Dirichlet boundary condition) numerically? As far as I know, the Lagrange Multiplier method and the Penalty method are two effective solutions to deal with boundary conditions in FEM. Does MOOSE use either of them?

Back to the problem of different physics solved in different domains, assuming that we know the solution for the yellow region for the chemical physics (we get the time- and space-dependent solution from somewhere, e.g. MATLAB), I want to assign these known solutions directly to the nodes via an ADFunctionDirichletBC so that to reduce the computational cost. However, after doing so, the total computation time increased a lot instead of decreased. The number of linear iterations used for each nonlinear step increased significantly, while the number of nonlinear iterations for each time step seemed to decrease slightly. Is it possible to improve the computational efficiency of MOOSE using this approach? If it is possible, I would like to know which step is wrong.

Is it possible to count the time taken for each part of the simulation? For example, the assembly of the stiff matrix, solving the system KU=F

[GiudGiud]

May I know how the MOOSE deals with boundary conditions (e.g. Dirichlet boundary condition) numerically? As far as I know, the Lagrange Multiplier method and the Penalty method are two effective solutions to deal with boundary conditions in FEM. Does MOOSE use either of them?

Both of these methods are available by selecting certain objects. For some variables (nodal) we should just set the DOF to the value of the BC

Is it possible to count the time taken for each part of the simulation? For example, the assembly of the stiff matrix, solving the system KU=F

We solve something closer to Jacobian dSolution = Residual actually. It's a Newton method.
You can set perf_graph = true in the Outputs block to get a summary. You can use a PerfGraphOutput object in the Outputs block to set more options

[keassyguang]

Both of these methods are available by selecting certain objects.

Which is set to default? Or do we have a default setting?

For some variables (nodal) we just set the DOF to the value of the BC

Do you mean the variables defined with order = FIRST family = LAGRANGE? If so, then after I set up a large number of nodes with ADFunctionDirichletBC, , the total computation time should decrease significantly. But in my case the calculation time has increased a lot. This is too weird.

[GiudGiud]

Penalty and Lagrange are not default. You have to pick a PenaltyDirichletBC for example

Yes the first order Lagrange are nodal, in that the DOFs are located at the nodes

It's not obvious the system will be easier to solve as the solution is now different and could be further away from the initial condition in terms of the region of attraction of Newton s method.

I think you should look at the libmesh documentation for a more comprehensive answer on how dirichlet BCs are applied by default

[keassyguang]

Penalty and Lagrange are not default. You have to pick a PenaltyDirichletBC for example
Yes the first order Lagrange are nodal, in that the DOFs are located at the nodes

Thanks a lot. But I think the documentation of libmesh is too difficult for me.