NodalKernel

[friedmud]

It's new System time! I'm sure this will be the last one 😄

We need the equivalent of NodalBC for Kernels. This will enable fully-implicit, fully-coupled, spatially separated ODEs (ie, an ODE at every node). I think @YaqiWang will use this as well as anyone doing chemistry related things....

When we do this we need to make sure we get the Jacobians correct (that's one of the big downfalls of doing this using the Auxiliary system)... and as long as we're doing that we should make the Jacobians work properly for NodalBC....

[YaqiWang]

Seems to me kernel can already do this. Any specific applications for this?

[friedmud]

Ummm - you had one the other day... you wanted to do

u_dot = u + v

That's a fully implicit equation that you want to couple back into your other equations. This comes up all the time in chemistry as well....

[friedmud]

It's for when you don't have any spacial derivatives but you need to solve it at every point in the domain...

[YaqiWang]

I see. Keep in mind though the variable could be elemental.

[dschwen]

I'm not seeing why

u_dot = u + v

couldn't simply be coded up as a regular kernel.

[friedmud]

Because Kernels are evaluated at quadrature points and their residuals are
integrated over elements.

You wouldn't be solving for u_dot. You would be solving for a projection of
u_dot into the Lagrange basis. Sometimes that can work... but it's actually
unstable.... and generally doesn't represent the actual system of equations
you're trying to solve...

Derek
On Mon, Mar 30, 2015 at 11:57 PM Daniel Schwen notifications@github.com
wrote:

I'm not seeing why

u_dot = u + v

couldn't simply be coded up as a regular kernel.

—
Reply to this email directly or view it on GitHub
#3029 (comment).

[friedmud]

One more use...

https://groups.google.com/forum/#!topic/moose-users/cthPGyKPPCE