The paper studies two ways of counting integer partitions, called f_{0,1,4}(n) and f_{0,2,3}(n). Each counts the number of ways to write a positive integer n as a sum of smaller integers, but with restrictions: the parts must belong to specific remainder classes when divided by 5, and no part can be used more than twice. The first function uses parts that leave remainders 0, 1, or 4 when divided by 5, while the second uses parts leaving remainders 0, 2, or 3. These two functions are related to classical objects in number theory called Rogers-Ramanujan identities, and the paper investigates how these counting functions behave modulo 3, meaning whether they give values that are divisible by 3 or not.
The central discovery is a striking symmetry governed by a concept from number theory called quadratic residues. For any prime p of the form 4k+3 (such as 3, 7, 11, 19, ...), the paper shows that the two partition functions either stay the same or swap roles when the input n is scaled and shifted in a precise way depending on p. Which behavior occurs, preservation or swapping, is determined entirely by whether 5 is a quadratic residue modulo p, meaning whether some integer squared gives 5 when divided by p leaves no remainder. This connection to quadratic reciprocity, a deep classical theorem about prime numbers, is what the authors call a "quadratic-residue dichotomy." The simplest example is that f_{0,1,4}(9m) equals f_{0,2,3}(m) modulo 3, so multiplying the input by 9 switches which function you are effectively computing.
The proof works by simplifying the generating functions that encode these partition counts, exploiting a basic modular arithmetic fact that (1-x) to a prime power p behaves simply modulo p, together with the Jacobi triple product identity, a classical tool for manipulating infinite product formulas. This reduces the problem to analyzing quadratic expressions and their symmetries relative to the prime p. As bonus results, the authors derive infinite families of self-similarity relations, where knowing the partition counts on one arithmetic progression tells you the counts on a related but much sparser progression, as well as congruences that hold everywhere except on a thin set of exceptions.