Metropolis not thread safe?

[simone-romiti]

If we run the metropolis with different numbers of cores the result changes.
The issue lies in the sweep function. Parallelizing in "t" only is not enough, as we should account also of nearest enighbors in the spatial directions.

[urbach]

If we run the metropolis with different numbers of cores the result changes. The issue lies in the `sweep` function. Parallelizing in "t" only is not enough, as we should account also of nearest enighbors in the spatial directions.

one needs to do even/odd in t-direction, but that should be enough. what do you mean by "the results change?" They are not expected to be identical, are they?

[christianegross]

By how much do the results change? I do not expect it to be exactly identical, because the number of random number generators we assign and use depends on the number of cores, and I am not sure if the loop iterations are distributed statically or dynamically to the threads.

[simone-romiti]

Yes, I mean "notidentical". They are fairly similar though. I am currently trying to implement a thread safe version of that.