# What function describes the frequency for each unique ratio for all possible expansions n over d where n<d?

I am hoping to solve the following problem for a scientific investigation, which relies on the probabilites of all possible expansions.

What function $f(r)$ describes the frequency for each ratio for all possible combinations of positive integers*

$$\frac{n}{d}$$

where $n$ and $d$ are finite integers $$n\wedge d \in\mathbb Z$$ ($n$ is the numerator and $d$ is the denominator) such that:

$$n<d$$

Example

When $d_{max} = 16$ the ratio $\frac{n}{d}=0.5$ can be expressed as $\frac{8}{16},\frac{4}{8},\frac{2}{4},\frac{1}{2}$ as well as $\frac{3}{6},\frac{5}{10},\frac{6}{12},\frac{7}{14}$

The number of possible expansions sum to $8$. Thus, the frequency of possible expansions when $r=0.5$ is $$f(0.5)=8$$

Given $n<d$ and $d_{max}=16$

What is $f(r)$?

*For this particular problem a solution with $d_{max}=16$ is sufficient

Background Reading for the Mathematically Inclined

I found this paper by Everett (1946), where f-expansions for the special case $x/p$ are mentioned, but it is too complex for someone with my limited background to digest. It does describe "correspondence from real numbers to sequences of integers mod p" (p.1).

Programmatical Solution in MATLAB

EDIT

Inspired by the reply to Number of common divisors between two given numbers I modified the code to work with MATLAB. The resulting output is presented in the figure below the code.

% MATLAB CODE
% Based on C++ math.stackexchange.com question 8611 (VelvetThunder)

dmax = 16;

distribution = zeros(dmax,1);

for j = 1:(dmax-1)

[N,C,D] = gcd(j,dmax);
sqt = sqrt(N);
for i = 1:sqt
if mod(N,i) == 0
end
if i == N/i
end
end

fprintf('%d\\%d = %d\n', j, dmax, answer);

bar(1:length(distribution),distribution);

end Commentary

The programmatical (inductive) solution appears to yield, given the mentioned constraints, a modal peak when $n = \frac{d_{max}}{2}$ and a stepwise up-and-down progression. However, for $d_{max}=128$ the pattern becomes more intricate: For inductive purposes, let us not forget to change the radix from 2, by setting $d_{max} = 96$. The $odd=1$ pattern is now gone, but the maximum number of expansions still appears to be located at the half-width point: The question how to express this - formally - remains.