Here's a revised calculation, with haplotype fitnesses.
In each lake, I computed relative fitnesses of each haplotype in each effect region by adding effect
sizes (ignoring dominance), converting these to selection coefficients as before, and then computing
the relative fitness as s / (1 - exp(-2Ns), where s is the selection coefficient. I then took the weighted
average total effect size in each effect region, weighted by relative fitness, and added these across the
10 effect regions.
This gets (per-population; followed by the average across populations):
> haplotype_fitness
[,1] [,2] [,3] [,4]
(0.476,1.46] 0.083017224 2.954911 6.562728 10.84983
(1.46,2.42] 0.101076300 2.125773 7.283770 10.68220
(2.42,3.38] 0.106038983 2.358805 7.225910 10.88597
(3.38,4.34] 0.002175190 2.489439 7.085753 10.93263
(4.34,5.3] 0.012261967 2.462888 7.448830 10.93193
(5.3,6.26] 0.034985717 2.581968 7.529912 10.60745
(6.26,7.22] 0.101088190 2.337338 6.402035 10.69996
(7.22,8.18] 0.072749463 2.366462 6.853273 10.92981
(8.18,9.14] 0.016865390 2.665307 7.082309 10.80564
(9.14,10.1] 0.070724707 2.519911 7.325465 10.85935
(10.1,11.1] 0.062815327 2.598656 6.549697 10.73068
(11.1,12] 0.043337131 2.654635 6.797221 10.83219
(12,13] 0.002165483 2.638918 7.381021 10.83361
(13,13.9] 0.029420235 2.623771 7.088186 10.95417
(13.9,14.9] 0.014018106 2.579534 7.123163 10.66724
(14.9,15.9] 0.176685036 2.178223 7.547326 10.71860
(15.9,16.8] 0.042529976 2.766579 7.724924 10.79790
(16.8,17.8] 0.022120510 2.978874 7.773772 10.86157
(17.8,18.7] 0.073748001 2.403862 7.200708 10.62315
(18.7,19.7] 0.037230712 2.674534 8.234115 10.78714
(19.7,20.7] 0.102264603 2.850651 7.352217 10.55872
(20.7,21.6] 0.082073645 2.311294 6.808178 10.82064
(21.6,22.6] 0.155318107 2.330787 6.723904 10.81415
(22.6,23.5] 0.072595210 2.544701 6.505628 10.90598
(23.5,24.5] 0.011870192 2.151168 6.872798 10.78364
> colMeans(haplotype_fitness)
[1] 0.06116702 2.52595957 7.13931374 10.79496722
Note that it doesn't match Jared's observation so nicely as before - here it's suggesting that m=0.1
will top out at about -5 before needed to start recombining - but recall that this calculation is imprecise
for several reasons, notably for not including linkage.
From my email of Sept 17: