MOOSE-NEML2-Pyzag Stress agreement for complex geometries

[fusmatrs]

Hello,

I have a couple of follow up questions regarding MOOSE/NEML2/Pyzag agreement. To recap - I am trying to take the strain history of a MOOSE model using NEML2 and comparing it to the same NEML2 model run with the same strain history using Pyzag. There's agreement between the stresses output in the case of a single element that's nodes are all constrained by BCs.

For more complex geometries, where not all nodes of all elements have BCs there is still a disagreement.

1. What is MOOSE doing differently such that the stresses are different away from the BCs?
   (I follow that MOOSE is solving a different problem, but ultimately NEML2 is being passed the same strains and presumably should return the same stresses)
2. Is it possible to force / modify the way the Pyzag solves the NEML2 model is run to get agreement between MOOSE outputs?

I have been reading through as much MOOSE / NEML2 / Pyzag documentation as I can find, but I am happy to be pointed to other documentation / resources.

Any guidance is appreciated!

Thanks

R

[reverendbedford]

Sorry for the delay, I was on vacation.

So if I understand correctly:

1. All nodes of FE model constrained: MOOSE and pyzag match.
2. Some nodes of the FE model constrained: MOOSE and pyzag don't match.

When you say you're imposing the MOOSE strains into the pyzag model which strains do you mean? The element output from MOOSE? If so the problem is likely that MOOSE has multiple quadrature points per element. The local strains at these quadrature points are likely different from the average element strain history you're getting reported back from MOOSE. So in MOOSE you have eight (for a linear hex) quadrature points each with somewhat different strains and in pyzag you just have the element average strain history. Those won't match.

That also explains why it works when all the nodes are constrained: then all of the quadrature points will have the same strains.

Options to test this:

1. Apparently we can now dump out quadrature point material properties from MOOSE. See here. Check to see if all the quadrature points in your element have the same strains.
2. Change your boundary conditions so that your quadature point strains are the same. There are some BCs that will give you this condition without needing to constrain all nodes. An example is symmetry boundary conditions on the -x, -y, -z faces, and the some applied loads on the + faces pointed in the normal direction to the face.

Let me know if my guess is wrong and I can take a more detailed look!

[fusmatrs]

Thanks for the response! It looks like extracting the data at all the quadrature points is the way to go. I guess an alternative would be to use reduced integration elements (i.e HEX8 with 1qp). For now that's not an option as there is no hourglassing correction implement in MOOSE.