Tomás Chor

> > That makes perfect sense, and it was my thought as well. However, the current behavior is that when cumulatively integrating over a periodic dimension, faces and center grids...

> Center and Face have the same size on Periodic dimensions. > > Also, I am confused whether this makes sense in a Periodic direction. As an example, `f(x) =...

> > > Center and Face have the same size on Periodic dimensions. > > > Also, I am confused whether this makes sense in a Periodic direction. As an...

> The problem is that integral of `f(x) = 1` is not periodic so it cannot be represented with a periodic `Field`. I'm aware. What I'm advocating for is that...

I don't know how to answer all those questions. I'm not that fast with software design. My only point is that it's definitely possible numerically. That said, given your questions,...

> Not sure if I am missing something but first I think the user can just write > > ```julia > averaging=(1, 2) > ``` I considered taking that approach...

Here's some validation. The animation below is from a triply-periodic 3D simulation with an initial reasonably-resolved initial condition without any forcings. Note that the IC looks a bit weird right...

It's also worth noting that right now many calculations are done more than once in each timestep. For example for each component of $M_{ij}$ I'm calculating the whole strain rate...

> One thing to note is that the current implementation appears to be very slow. While the simulation with the `SmagorinskyLilly` closure runs on my laptop in 10 seconds, it...

I have some validation scripts already that work on the CPU and indicate that the new closure is working. However, I can't make this work on the GPU. I keep...