Consider the rationals (fractions) between and with denominator bounded by . That is
These fractions are called Farey fractions of level and we denote them by
The number of fractions is
Neighbours: Any two neighbours have the property that
In fact, two fractions are adjacent to each other in some Farey sequence iff
.
In particular they are adjacent in if and only if we have
To see these conditions, observe the mediant property
,
and the fact the we can construct all the Farey fraction starting from by adding the mediant points successively — We can see this inductively, any fraction between two neighbours in will have denominator at least .
(We used inductive hypothesis to get )
Therefore the mediant is the first “new” fraction that’s going to be added. We can check that the determinant condition between the neighbours is preserved as we add the mediants,
.
Finally the condition for a neighbours in a fixed level because if the sum is smaller, the mediant is in between them and also lies in .
If are consecutive fractions in a fixed level we have the following relation
This can be rephrased in terms of the variables as
where
So by iterating we can write
.
Note that is a transformation defined on the Farey triangle
by the properties of neighbouring fractions.
Distribution/ Statistics:
How is the sequence distributed? What about the gaps and other statistics?
We can see that the minimum gap is
and the maximum gap is
.
But the average gap is .
Although the gaps are not the same and there is a lot of variation, the sequence is uniformly distributed as , that is every interval has approximately proportional number of fractions. That is
To prove this note that that , so count the integer points in the interval which are coprime to , which amount to approximately
and then summing over gives
The error terms are governed by the Weyl sums , which by Mobius inversion (to remove/detect the condition ) are given by
Thus the problem of finding the best error terms is related to bounding the Mobius sums .
Anyway note that equidistribution is a statement about of coarse scale behaviour– we take intervals of fixed size. If we take very small intervals, the error terms may dominate and the interval might have a lot more or lot less fractions. For instance has zero points, but the uniform distribution predicts for the count. The smallest scale on which the uniform distribution is valid is related to the error terms and rate of convergence of the above equidistribution statement and as we saw is related to finding cancellation in the Mobius sums- this is extremely hard and basically the Riemann hypothesis problem.
We can ask how the spacings are distributed. In fact, we want to understand the distribution of of a the denominators of adjacent fractions. The scaled denominators are uniformly distributed in the Farey triangle
Thus the spacings are distributed according to the distribution of on the Farey triangle. In general any statistic can be seen the statistics of on the Farey triangle.
Example: The average gap because sum of gaps is the length of the interval , but the uniform distribution implies that the average has to be
Computing the integral we get and this equals for large .
The expressions for the spacing distribution where computed by Hall. Let be the fraction of spacings which are at least , then we have
This distribution is given by density ,
with
This allows us to see that the most common gap is near
If we want distribution of for consecutive fraction, using the transformation that relates adjacent denominators, the distribution is given by the distribution of for a random point in the Farey triangle. Once we have this we can compute higher level spacing distributions.
Both these properties follow from the uniform distribution, but the error terms have a lot more arithmetic- for instance for , we need to understand the distribution of quantities like .