Why a std::array of MultiFabs instead of a single MultiFab with ncomp=2 in Electrostatic PIC tutorial?

[n8xm]

In the Electrostatic PIC tutorial ( authored by @atmyers ), we set up eField (which I am pretty sure is the electric field) as follows:

1. Initialize eField on line 104, as a Vector of std::array of MultiFab pointers, where the std::array is of length AMREX_SPACEDIM.
2. On lines 114 to 117, for each spatial dimension, create a new MultiFab with the ncomp argument set to 1

Presumably, the reason why we effectively have an have an array of MultiFabs of length AMREX_SPACEDIM is because the electric field is a vector field, unlike phi (electric potential?) and rhs (RHS of the Poisson problem?) which appear to be scalar fields.

What I'm wondering is: If eField is a vector field (which by definition requires one component per spatial dimension), why create multiple MultiFabs, each with ncomp=1, as opposed to a single MultiFab with ncomp=AMREX_SPACEDIM? Is it because the compute_E_nodal function assumes separate arrays for each spatial dimension? Or is there some other advantage?

[atmyers]

In that particular example, the electric field could be stored in a multi-component Multifab as you suggest, and the code would likely be a cleaner, because all the components of E have the same nodal centering. However, if you have a case where the different components might have different staggering, like on a Yee grid, then you need to store then as separate Multifabs. I think I was probably starting from a more general code that had (or could potentially have based on runtime options) staggered E fields.