The dx/lx drifting away at the end is just rounding. Perhaps the 100,000 lives we start with is more like 1,000,000 but we only see it to the nearest thousand. Or, I like Calum's idea that everything is worked out from mu(x) and we work backwards, again with rounding problems.
Calum, careful with this qx = mu x idea.
I don't know what "perfectly constant" means as opposed to just constant but, if mu is constant between x and x+1 then we would get:
qx = integral beteen 1 and 0 of tpx mu(x+t) dt
= mu * integral beteen 0 and 1 of tpx dt
= mu * amount of time you expect someone to live over the next year
= mu * 0.99ish probably
<> mu
Or, qx = 1 - exp(-mu) from page 32 Tables
= mu - mu^2 / 2! + mu^3 /3! - .... (using page 2 Tables)
Also, in ELT15 we have mu(x) < qx at age 85 and mu(x) > qx at age 86 in ELT15 males. So, what you're saying here isn't really true either :-(
I agree that, if mu(x) is decreasing then qx - mu(x) will be bigger than it otherwise would have been. But this doesn't necessarily mean that qx - mu(x) is positive. Again, we can't really compare the absolute values of mux and qx because one is a FORCE, the other is a PROBABILITY.
Good luck!
John
Last edited: Apr 13, 2012