Single threads or Multitheads??

[kuangtianhui]

Dear author，

The following are two scenarios using MrBayes. Only the Nchhains parameter and the number of running threads are modified. What impact will these two different operation modes have on the results? Which result is more credible?

1.Other parameters remain unchanged. Adjust Nchains to 4 and run MrBayes in a single thread.

2.When other parameters are consistent, only the Nchains parameter is changed, and then multiple threads run MrBayes.

Best wishes,
Tian-hui Kuang

According this guidance of v3.2 it just for speeding up the calculation of big data set.

[nylander]

Dear @kuangtianhui
Both the number of chains (i.e., number of "independent" starting points), and applying Metropolis-coupling (using "heated", helper chains) are means to more quickly reach the target distribution. After reaching the target, which can be more difficult and take more time for some data, especially larger, the additional samples are only gathered for increased precision. Given that you run your chain under the two scenarios outlined above, and that you access the convergence of the results, there is no reason that one should be more credible than the other. I's just a matter of waiting for chains to converge.
Yours
Johan