still no power. computed log 2, log 3, and log 5 to three digits by hand. tomorrow i might learn how to churn butter idk
Conversation
This Tweet was deleted by the Tweet author. Learn more
they are several pages unfortunately. the main upshot is using the taylor series of hyperbolic arctangent to speed up convergence (it's a little more than twice as fast)
2
This Tweet was deleted by the Tweet author. Learn more
so e.g. if you wanted to compute log 2 you could do it by computing it as log 1 / (1 - 1/2) = sum (1/2)^n / n, this converges but kinda slowly. there's a much funnier way to do it which is to compare the taylor series for log 3/2 = log 1 / (1 - 1/3) and log 4/3 = log (1 - 1/3)
1
2
this gets you log 2 = log 3/2 + log 4/3 = 2 sum (1/3)^{2n-1} / (2n-1), whose terms grow like a power of 9 instead of like a power of 2 and so which converges more than 3 times as quickly (in the sense that you get more than 3 times as many digits of accuracy per term)
1
2
whoops, that should've been log 4/3 = log (1 + 1/3), the point is that the even terms in the taylor series cancel when you add them and this means you're getting the taylor series of hyperbolic arctangent
2
This Tweet was deleted by the Tweet author. Learn more
idk like theoretically a class in real analysis or numerical analysis?
4
This Tweet was deleted by the Tweet author. Learn more