add positions_relative or similar?

[rkurchin]

The workshop discussion on dimensionless systems yesterday was inconclusive at best, but a related thing that occurred to me that might also at least partially address the whole Unitful + AD potential snarl while hopefully avoiding some of the ambiguity issues we're concerned about could be an explicit way to get relative coordinates of particles (i.e. in terms of the lattice vectors of the bounding box).

What do people think about this? And if we were to add it, do folks have thoughts about whether it should be a separate function vs. e.g. a separate dispatch of the positions function, e.g. via a flag that could be passed or something?

[jgreener64]

As I understand it you mean fractional coordinates? I could see the advantage of having that, especially since there is a sensible fallback.

[rkurchin]

Yes, that's what I meant, thanks for clarifying!

[cortner]

I have no opinion on introducing a relative coordinates interface. I can see it being useful.

But I don't think this solves the dimensionless systems request, it would feel like a hack - using a framework outside of the scope for which it was intended - with unpredictable consequences.

[mfherbst]

I think the main argument that has been put on the table against this has always been: What to do for isolated systems (cell isa IsolatedCell) ? Fractional coords don't really make sense here. I'm not sure how to handle that in a generic interface while keeping a flexibility for the cell.

[cortner]

throw an error

[mfherbst]

Ok, but my point is if this is part of the interface and I write my package against it than implicitly I am assuming that the cell is PeriodicCell. So should this not be made explicit ?

[rkurchin]

Now we go back down the proliferating type parameters rabbit hole I suppose...though I could see an argument for it in this case...

[cortner]

Just use PeriodicCell(FFF). I don't really understand the need for the IsolatedCell anyhow.