I know that famous numbers like $\pi$ and $e$ have multiple representations as continued fractions and I'm fascinated with the variety of representations.
My question: What continued fraction for $e$ is most computationally efficient? A proof or a link to one for why a method is optimal would be of great interest to me.
Here, my metric for "computational efficiency" is achieving the most precise decimal places with a fixed number of terms generated in the continued fraction.
If you allow generalized continued fractions $$ b_0 + \dfrac{a_1}{b_1 + \dfrac{a_2}{b_2 + \ldots}}$$ for positive integers $a_i$ and $b_i$, there is no "optimal" answer for the given criterion.
You can achieve arbitrary accuracy in any given number of terms, at the expense of having large numbers $a_i$ and $b_i$. Namely, let $r_1, r_2, \ldots$ be any sequence of positive rationals converging to $e$ with $r_i$ decreasing for even $i$ and increasing from odd $i$. We take $b_0 = 0$, $r_1 = a_1/b_1$, $a_2$ and $b_2$ so $$b_1 + \dfrac{a_2}{b_2} = \dfrac{a_1}{r_2} $$ which makes $$ r_2 = \dfrac{a_1}{b_1 + \dfrac{a_2}{b_2}}$$ $a_3$ and $b_3$ so $$ b_2 + \dfrac{a_3}{b_3} = \dfrac{a_2}{\dfrac{a_1}{r_3} - b_1} $$ which makes $$r_3 = \dfrac{a_1}{b_1 + \dfrac{a_2}{b_2 + \dfrac{a_3}{b_3}}}$$ etc.