Sum of Egyptian Fractions

A friend recently posed me the following question: for an integer n, how many pairs of positive integers (a,b) exist for which

\frac{1}{n} = \frac{1}{a} + \frac{1}{b}?

Call this number F(n). Here is my solution. We will count the number of positive integers a for which there exists a corresponding b.

First, note that a b exists if and only if the numerator of \frac{a-n}{an} cancels when the fraction is put into lowest terms, in other words, if and only if a-n\vert an.

 

Lemma 1: F(n) is equal to the number of integers a>n for which a-n \vert (a,n)^2, where (a,n) denotes the greatest common divisor of a and n.

To see this fact, write a = x(a,n) and b = y(a,n). Then (x-y) \vert x y (a,n). Since x and y are relatively prime, x-y is relatively prime to both x and y, so (x-y)\vert (a,n), or (a-n)\vert (a,n)^2.

 

Lemma 2: F(n) is equal to the number of integers a>n for which a-n \vert n^2.

We will show that this set of as and the set of as given by Lemma 1 are in bijection. First, if a-n\vert (a,n)^2, then clearly a-n\vert n^2 since (a,n) is a divisor of n.

Now suppose (a-n)\vert n^2. Then a-n is equal to a divisor d of n^2. We can write d=d_1d_2 where d_1\vert n and d_2 \vert n. Therefore both d_1 and d_2 divide the right-hand side and so divide a, so both are common divisors of a and n and so d_1\vert (a,n) and d_2 \vert (a,n). Therefore d_1d_2 \vert (a,n)^2.

Since for every divisor of n^2 we can solve for a unique a>n, we have

 

Answer: F(n) = d(n^2), where the divisor function d(x) counts the number of positive divisors of x. Note that this F counts unordered pairs. There are \frac{d(n^2)+1}{2} unique ordered pairs with a \leq b. For n=10^7, there are 225 pairs (113 ordered pairs).

d(n^2) can be computed naively in O(n) by finding n's prime factorization using trial division. Possibly there are faster algorithms.

 

Footnote: Like the divisor function, F is multiplicative: if (x,y) = 1, F(xy) = F(x)F(y). Maybe there's a way to see that directly and use it to derive a formula for F in an alternate way (by first finding a formula for F(n) when n is a prime power.)