Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Here is a small discovery I stumbled across a few weeks ago. I hope at least one person will find it interesting enough to help me. The iterated continued fractions from convergents (or convergents constants) are explained at . Among other things, there you find out that a randomly selected number between 2 and 3 has a convergents constant of 2.3484074702792306..., but the same is not true for 2.1, 2.2, 2.5 and perhaps for a few other values. Also for most 0 < x < 1 the same process of iterated continued fractions returns 0.5557531042780459..., but it is not so for x =0.1, 0.11, 0.12, 0.2, 0.25, 0.34, 0.35, 0.43, 0.45, 049, 0.5, 0.65, 0.75, some values < 66/1477 and perhaps for a few other values.

I would like to know why these numbers and those like them are exceptions to the rule.

share|cite|improve this question
Removed the functional analysis tag. – Eric Naslund May 15 '11 at 2:29
See the answers for…. – Yuval Filmus Jun 2 '11 at 0:10

Possibly a bit more light occurs, if we consider to express that iteration using the matrix-representation of the continued-fractions-evaluation.
Here we insert the coefficients of the cf in a sequence of matrices, whose form is
$ \qquad M_0(a)= \begin{bmatrix} 0 & 1\\1 & a \end{bmatrix}$
and for the product of matrices according to the coefficients in the cf allow the generalized notation
$ \qquad M_0(a,b,c,...,h)= M(a)*M(b)*...*M(h)$

The iteration refers to the partial products, used for a new continued fraction:
$ \qquad M_1(a) = M_0(a); M_1(b) = M_0(a)*M_0(b) = \begin{bmatrix} 1 & b\\a & a*b+1 \end{bmatrix} ; \ldots $

This is simple to program, for instance in Pari/GP. Unfortunately this does not hold exactly: the $M_1()$ created by this procedure are not compatible with the form $ \qquad \begin{bmatrix} 0 & 1\\1 & a \end{bmatrix}$
, they have general (integer) values in all four entries; so do not reflect the required "simple-continued-fraction" representation. So for to model the process as you described it in your question we need some normalization.

What I tried next was to insert the evaluated partial continued fractions as rational (or real) numbers instead of integer numbers in the positions of the a in $M_1(a)$ - so $M_1(a) = M_0(a); M_1(b) = M_0( {ab+1 \over b}), \ldots $ according to the evaluation of the partial convergents. But still this requires then another type of re-normalization (it does not lead to the same limit value), because the fractional parts of the coefficients must now be "shifted" to the remaining cont-frac-expression.

But perhaps we find here the effect, why the leading coefficients of the iterated continued fractions converge to some constant: because after evaluations of subsequent coefficients a certain threshold for the fractional part of a coefficient can no more be overcome (in the next iteration).But I do not see really clear here...

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.