Fractions produced by mediants have some very interesting properties. We saw some of them in connection with the Stern-Brocot tree. This articles explores a more curious property, relating fractions to circles in the plane. It was discovered in 1938 by Lester R. Ford and is also mentioned in Conway and Guy's The Book of Numbers.
Let us consider a way to construct fractions. We always start out with the sequence
and then repeatedly obtain a new sequence from an existing one. Given a sequence, we pick two consequtive fractions and insert their mediant between them. Recall that the mediant of and is . We will call sequences created in this manner mediant sequences.
So the first (and only) mediant sequence possible from the initial sequence (1) is
where the new fraction is surrounded by angle brackets. From this point on, we have multiple choices each time we produce a new mediant sequence. An example of the following sequences could be
An essential property, in regard to this article, is this: For all mediant sequences, any fraction followed by fulfills . This is easily shown by induction by considering the initial mediant sequence (1) and that if for a subsequence
then it also holds after inserting the mediant
(two cases to check).
Note how this property implies that the fractions in any mediant sequence are ordered by size. But the most important use of this property will appear shortly.
We now do the following in a Cartesian coordinate system: For each fraction in a given mediant sequence, draw a circle centered at with radius . An example can be seen in Figure 1.
By construction, the x-axis is obviously a tangent to all such circles. Furthermore, two circles, corresponding to two fractions and , touch at exactly one point if and only if . This is shown by an application of the Pythagorean Theorem:
This is exactly the property that consequtive fractions in any mediant sequence fulfills.
Zooming into the dashed region in Figure 1 leads to Figure 2.
So given a mediant sequence, we can draw such circles for each fraction in the sequence, except for the final (which is just an auxillary element). The circles with diameter correspond to the seqeuence and will all touch each other. See Figure 1. Any circle with diameter less than will touch exactly two larger circles, namely the two from which it was created when considering their mediant sequence representatives. To be more specific, if we went from a mediant sequence
with to
then the three circles corresponding to the fractions of this second subsequence will all touch, and the middle one will obviously be the smallest (has the largest denominator). See, e.g., the fractions , , in Figure 2.
Zooming in once again we get Figure 3.
Note the many self-similarities.