Access to conjugate variables at nodes with BCs

[bwspenc]

In MOOSE, boundary conditions modify the residual vector, which is driven to zero by the solver. The sum of the increments applied to the residual by those boundary conditions is the conjugate to the solution variable at those boundaries (e.g. force/displacement for solid mechanics, integrated flux/temperature in heat transfer). There is currently no way to access this because the BCs modify the residual, but the increment applied to the residual is not saved.

We need to access the conjugate to the solution variable at boundaries for many reasons, including the following examples: Contact forces are needed to implement frictional sliding and pressure-dependent heat transfer, and are often the desired outcome of contact calculations. Nodal reaction forces at Dirichlet boundary conditions are often the desired outcome of structural mechanics calculations. Similarly, fluxes are often of interest in thermal analyses. It is also often helpful to have access to the vector of nodal forces applied by pressure loads. Using the norm of the reaction force vector as a reference quantity for a relative convergence tolerance is more reliable than using the initial residual norm, which is what we currently do.

Solid mechanics codes typically provide access to a variety of force variables for output. The main ones are the reaction force (due to Dirichlet BCs), contact force (due to contact), and external force (due to force or integrated traction (Neumann) BCs). An implementation of a capability to access conjugate variables in MOOSE should provide access to these three types of conjugate variables.

[friedmud]

I'm not opposed to this... and have actually thought about it in the past. One thing though:

I don't understand how to get the reaction forces at DirichletBCs. It's not the residual there (that would be zero once the problem is solved). Can you expand on what it would be? It's actually something more like evaluating the boundary integral that comes out of the integration by parts of the stress divergence equation and reporting it's residual at the nodes... right?

[jasondhales]

Yes.

The simplest thing to do, at least initially, would be to evaluate the residual sans the kinematic bcs and write that vector of forces to the results file. This would then include external loads like pressure, reaction loads, and a bunch of zeroes in the interior.

[mooseframework]

"Nodal reaction forces at Dirichlet boundary conditions are often the desired outcome of structural mechanics calculations." I would back the comment by Ben.

In Marmot we would like to have this feature as well. As the grains evolve under applied stress we would like to monitor the reaction forces as a function of time at Dirichlet boundary conditions.

[friedmud]

In 4cb5e2a9f87973dba738584db39f7db935b65ce5:

adding the ability for kernels and bcs to save their residual contributions to an auxiliary variable closes #1030 #708 #722 #1150

[friedmud]

In 4fe4ef048c38d1234ae26a50384878f1b4340c77:

add missing files ref #1030 #708 #722 #1150

[friedmud]

In 9cf1d8b91d0edb28ed97b5ec4580126bfc46779f:

disable unstable tests ref #1030