Permalink
Browse files

Merge pull request #2430 from RudolfWeeber/lb_accuracy

Doc: Note about limited lb gpu precision
  • Loading branch information...
KaiSzuttor committed Jan 9, 2019
2 parents c9e20be + 039a78d commit b5e2b2659ad838b0bc8e10604bd2a743b9ad5ed5
Showing with 1 addition and 0 deletions.
  1. +1 −0 doc/sphinx/lb.rst
@@ -42,6 +42,7 @@ To use the GPU accelerated variant, replace line 5 in the example above by::

To use the (much faster) GPU implementation of the LBM, use
:class:`espressomd.lb.LBFluidGPU` in place of :class:`espressomd.lb.LBFluid`.
Please note that the GPU implementation uses single precision floating point operations. This decreases the accuracy of calculations compared to the CPU implementation. In particular, due to rounding errors, the fluid density decreases over time, when external forces, coupling to particles, or thermalization is used. The loss of density is on the order of $10^-12$ per time step.

The command initializes the fluid with a given set of parameters. It is
also possible to change parameters on the fly, but this will only rarely

0 comments on commit b5e2b26

Please sign in to comment.