[ESPResSo-users] GPU LB buggy since Feb

[Stefan Kesselheim]

Dear Espresso-Users,
we have noticed that there is a severe bug in the GPU LB code in version 
3.1 and git versions after Feb 1st:

Since the implementation of the lattice shift the interpolation of the 
velocity and the force assignment is not working propertly: the 
interpolation factors regularly become negative leading most importantly 
to spurious dipole-like force assignments on the LB fluid. Their sum 
however is still 1 in all cases.

As we can not estimate how severe this bug is, we recommend to check 
produced data with the current git version (master or 3.1 branch) or the 
upcoming 3.1.1 version that will be published soon. In doubt, run all 
publication relevant simulations again.

We apologize for the inconvenience and take this as the best possible 
argument for a good test system. We are very open for all simple test 
cases (preferably against analytically known solutions) that help 
avoiding these problems in the future.

Here is the code of the Buggy commit:
5fb0db5b8a07a336011426dfb717c44028e86f9c
The fix is this commit:
03b428de453652033f25a3e9ba791cd4f6c59dbb

Cheers and good luck with Espresso
Stefan Kesselheim

Institute for Computational Physics
University of Stuttgart