Andrej Dujella: Newton's approximants and continued fractions

Let $ d$ be a positive integer which is not a perfect square. Let $ p_n/q_n$ denote the $ n$-th convergent and $ s(d)$ the length of the shortest period in the simple continued fraction expansion of $ \sqrt{d}$. It was proved by Mikusinski in 1954 that if $ s(d) \leq 2$, then all Newton's approximants

$\displaystyle R_n = \frac{1}{2} \Big( \frac{p_n}{q_n} + \frac{dq_n}{p_n} \Big) $

are convergents of $ \sqrt{d}$, and moreover $ R_n = p_{2n+1}/q_{2n+1}$ for all nonegative integers $ n$. If $ R_n$ is a convergent of $ \sqrt{d}$, then we say that $ R_n$ is a "good approximant". In 2001, we proved the converse of Mikusinski's result, namely that if all approximants are good, then $ s(d) \leq 2$. It is easy to see that $ R_n > \sqrt{d}$. Therefore, good approximants satisfy $ R_n = p_{2n+1+2j}/q_{2n+1+2j}$ for an integer $ j=j(d,n)$. If $ s(d) \leq 2$, then $ j(d,n) = 0$. For $ s(d) > 2$, we proved the upper bound $ \vert j(d,n)\vert \leq (s(d)-3)/2$, and we presented a sequence of $ d$'s (given in terms of Fibonacci numbers) which shows that this upper bound for $ \vert j(d,n)\vert$ is sharp. Let $ b(d)$ denote the number of good approximants among the numbers $ R_n$, $ n = 0, 1, \ldots , s(d)-1$. We will present some results and conjectures (based on experimental data) about the magnitude of $ b(d)$ compared with $ d$ and $ s(d)$.

Back to the Index

Please send comments and corrections to Thomas Klausner.