Inverse correlation seems plausible, literal inverse proportion seems a bit too specific a claim. I guess that's the biggest problem with your analysis here generally; it's basically impossible to know loss rate, and any assumptions about it are just that. I certainly agree it's interesting to think it through, though, starting from the simplest possible model.
Discussion
Obviously loss rate is a function of value. Not some absolute number. That's why we can be sure that the absolute rate of sats loss will diminish over time.
It's not completely obvious that the function doesn't have some floor/minimum (I don't think I can get to 10^-9 loss per year even if my life depends on it ... or whatever). But generally I'd cede the point that you're mostly right on that.
Sats don't just magically decay. If you have a seed phrase that secures 100 sats, the chance of you losing that seed phrase is no different than a seed phrase securing 1,000,000sats _if_ the price is different enough that both seed phrases represent the same value to you.
...and lightning already supports milisats. So precision isn't a big deal.
Yes the loss rate is not dependent on the numerical sat value, and *is* dependent (inversely) on the economic value. I agree that my message was blurring and confusing these two clearly correct points.
But that dependence on economic value can't be assumed to be some very simple formula such as inverse linear (l.r. = k/value say), albeit that's a sound starting point. It's true that this has a nice asymptote at l.r. = 0 but we don't know if there is some other asymptote/floor; also, we might *think* that twice as much effort expended (because twice as much value) leads to half the loss rate, but .. does it? i wouldn't be surprised to see some exponential component there. it's fascinating to analyze, but it's hard to do more than theorize, because we have a tremendous and, likely, impossible challenge in trying to measure or predict loss rates.