disallow rescaling for traits

[petrelharp]

Currently, the rescaling factor ("Q") affects s. I think we probably need to disallow rescaling for simulations where there are traits except the classic "just fitness" trait. Unless the right way to do this is known and/or easy to see?

Considerations for when rescaling is allowed:

• at least, for mutations that directly affect fitness
• probably we want to disallow it for non-neutral traits where the trait isn't fitness?
• should we disallow it for any trait that's not the classic "just fitness" trait?

I'd need to go remember the detailed rationale for rescaling to decide on this. Notably: since each "individual" in a rescale simulation somehow represents Q individuals, would (say) "mean trait value for those Q individuals" make sense as an interpretation of the "trait value" for that "rescaled individual"?

[jeffspence]

This paper by Johri, Pouyet, and Charlesworth suggests that when a population is close to the optimum and $$V_P \ll V_S$$ one should just scale $$V_S$$ to $$V_S/C$$ where $$C$$ is the multiplicative factor multiplying the population size (this scaling is described in Section 5, which starts at the bottom of p. 13). I assume that this falls out of $$1/V_S$$ being proportional to the second derivative of the fitness function at the optimum, and so we could in principle scale the parameters of arbitrary fitness functions so that their Hessians (evaluated at the optimum) end up getting scaled by $$C$$.

There are a some caveats mentioned in the paper --- the population has to be close to the optimum, $$V_S$$ has to be small relative to $$V_P$$, and scaling affects the Bulmer effect. Are these sufficiently worrisome that we should completely disallow scaling? Or should we allow scaling $$V_S$$ but provide a (stern?) warning? I could imagine personally wanting to do scaled simulations in regimes where (I think) the scaling shouldn't be too problematic, but I don't feel strongly about whether to allow this or not.

[petrelharp]

I don't think that all of our simulations are going to satisfy those requirements, so that we should just disallow it; otherwise we need to check "is this simulation in the regime where this is reasonable". Seems like a can of worms (that someone else can open up in the future).