One other issue about equilibrium:
In our previous (unrealistic model) we found that USS initially accumulated very quickly, as singly and doubly mismatched sites were converted to perfectly matched sites. But this happened at the expense of depleting the genome of those mismatched sites, and further accumulation of perfect sites required waiting a long time for mutation of worse sites to regenerate the singly and doubly mismatched ones, which would then slowly allow further increase in the number of perfect matches. So achieving true equilibrium took a long time.
I expect this phenomenon to also apply in this new model. So I'm not at all confident that an early leveling-off of the rate of increase indicates closeness to the true equilibrium.
In the graphs to the left, the upper simulation probably hasn't reached equilibrium after 90,000 cycles (because the blue points indicating genome scores are still increasing), but the lower one has (because the blue points seem to be scattered around a stable mean).
I'm not sure why the lower run reached equilibrium so much faster than the upper one. Several factors differed - this is why I need to be more systematic. My excuse is that it's easier to motivate a systematic approach when individual tests are fast to do, and there are so many variables to test that I hate to spend a lot of time on just one. But it's time to treat this like real science.
A new kind of problem
12 hours ago in RRResearch