For increased accuracy, particularly with lower precision formats, it can be worthwhile to use distribution shifting. This means instead of storing the PDF, we only store the deviation from the equilibrium values.
The idea was first introduced by Skordos. It is also mentioned in the book by Succi and later in a paper by Dellar. For relative deviations see the document I wrote a while ago: relative_pdfs.pdf.
In lbmpy (see #7) you can get distribution shifting by setting zero_centered=True in the LBMConfig.
For increased accuracy, particularly with lower precision formats, it can be worthwhile to use distribution shifting. This means instead of storing the PDF, we only store the deviation from the equilibrium values.
The idea was first introduced by Skordos. It is also mentioned in the book by Succi and later in a paper by Dellar. For relative deviations see the document I wrote a while ago: relative_pdfs.pdf.
In
lbmpy
(see #7) you can get distribution shifting by settingzero_centered=True
in theLBMConfig
.