On 2 Apr 2017, at 04:29, Christian D. Ott cott@tapir.caltech.edu wrote:

Hi All,

I've been playing around with the GW150914 par file and have been trying to set up a pure Cartesian AMR run with the exact same grid setup in the Cartesian region.

I noticed that

Coordinates::additional_overlap_size = 3

(that appears to be required) not just increases the size of the cubed-sphere patches, but also of all patches in the Cartesian region, including refined regions. Is this a feature (if so, why?) or a bug?

Hi Christian,

I have never noticed this; I hope it's not happening, as it would be very strange! If it is happening, it definitely sounds like a bug. I'm fairly sure I have checked the refined regions in the past when using Llama, and they were the expected sizes. So if this is happening now, it is likely a change (probably to Carpet) that has happened relatively recently (the past year or two).

Can you post the parameter files and carpet-grid.asc files for the two runs which differ only in additional_overlap_size?

Is it possible that by increasing additional_overlap_size, the coarsest Cartesian grid RL0 is being moved inwards too close to the buffer zones of RL1, and RL1 is being extended to cover RL0 as a result? This would result in RL1 appearing to increase in size, but it shouldn't affect the finer levels. Do you see the finer levels also change?