Arithmetic Progressions of Squares

Updated 8 February 2026

Arithmetic progressions of squares are sequences of perfect squares with a constant difference, illustrating deep results in number theory and Diophantine geometry.
They bridge classical techniques such as infinite descent with modern methods including elliptic curve parametrizations and additive-combinatorial constructions.
Research employs explicit parametrizations, Dirichlet series asymptotics, and modular approaches to classify progression lengths across different number fields and settings.

An arithmetic progression of squares is a sequence of numbers, each of which is a perfect square, forming an arithmetic progression: that is, a sequence $a_1^2, a_2^2, ..., a_n^2$ such that the differences $a_{j+1}^2 - a_j^2 = d$ are constant. These structures arise at the intersection of Diophantine geometry, additive combinatorics, and the arithmetic of elliptic curves, with classical results and sharp modern advances delineating both what is possible and what is forbidden in various number-theoretic settings.

1. Classical Results: Length and Parametrization over $\mathbb{Z}$ and $\mathbb{Q}$

A fundamental result due to Fermat asserts that there is no four-term arithmetic progression of integer squares (or, indeed, of nontrivial rational squares) (Granville, 2017, 0712.3850, Alvarado et al., 2012, Selder et al., 2014). This is a deeper diophantine obstruction, proven by infinite descent and, in the modern perspective, via the Mordell–Weil group of certain elliptic curves of rank zero. Specifically, if four squares are in AP, the associated equations force a descent to ever smaller solutions, contradicting the well-ordering principle unless the progression is constant.

In contrast, three-term progressions of (integer or rational) squares exist in infinite number. An explicit parametrization for rational three-term APs is

$(x_1, x_2, x_3) = (t^2 - 2t - 1,\ t^2 + 1,\ t^2 + 2t - 1)$

so that the squares $(x_1^2, x_2^2, x_3^2)$ are in arithmetic progression with common difference $4t(t^2-1)$ (Alvarado et al., 2012, Selder et al., 2014).

A classical example is $(1^2, 5^2, 7^2)$ with common difference $24$.

Table: Maximal Lengths of Nontrivial Square APs in Various Settings

Field/Domain	Maximal Length	Parameters/Remarks
$\mathbb{Z}$ , $\mathbb{Q}$	3	Infinite families (see above)
Quadratic number fields	5	Explicit families; none with 6 (Bremner et al., 2015, González-Jiménez, 3 Feb 2026)
Cubic number fields	4	Impossibility of length 5 (Bremner et al., 2015)
Rational function fields	Variable	Structure depends on genus

2. Geometric and Elliptic Curve Structure

Three-term arithmetic progressions of rational squares are closely related to the theory of congruent numbers and the geometry of elliptic curves:

The condition for three squares in AP is $x^2 + z^2 = 2y^2$ .
The set of rational solutions is parametrized via the genus 0 curve and is birational to the associated congruent number elliptic curve

$E_d : y^2 = x^3 - d^2 x.$

Every rational point on $E_d$ (excluding torsion) produces a 3-term AP of rational squares with common difference $d$ (Cushing et al., 2016, 0712.3850, Selder et al., 2014, Alvarado et al., 2012).

Parametrizations via elliptic curves also underlie the construction of infinite families and the group law yields, for suitable choices, infinitely many disjoint APs of length 3 with the same difference. This is exploited in additive-combinatorial constructions to create large families with controlled additive properties (Cushing et al., 2016).

3. Maximum Sets and Sumsets: Additive Combinatorics

Let $S(R)$ denote the set of squares in a commutative ring $R$ . Define for $A \subset S(R)$ of size $n$ :

$N_n(R):= \min_{A \subset S(R),\ |A|=n} |A+A|.$

In $\mathbb{Z}$ ,

Lower and upper bounds: $2n-1 \leq N_n(\mathbb{Z}) \leq \frac{n(n+1)}{2}$ .
For $n=3$ the AP $\{1,25,49\}$ shows $N_3 = 5 = 2\times3 - 1$ .
The minimal sumset is achieved only when $A$ is an arithmetic progression of squares.

Beyond $n=3$ , as no 4-term integer AP of squares exists, constructions require unions of disjoint 3-term APs. Over $\mathbb{Q}$ , taking unions of many 3-term APs yields $N_{3n}(\mathbb{Z}) \leq 2.5n^2$ (Cushing et al., 2016).

In finite fields $\mathbb{F}_p$ , additive structure and the Cauchy–Davenport inequality guarantee that for sufficiently large $p$ and for any $n$ ,

$N_n(\mathbb{F}_p) = 2n-1,$

with equality if and only if $A$ is an AP in the set of nonzero quadratic residues (Cushing et al., 2016).

4. Asymptotics and Lattice Point Counts: Dirichlet Series Methods

The precise asymptotics for the count of primitive 3-term APs of squares with bounded entries have been addressed using multiple Dirichlet series and Tauberian arguments (Hulse et al., 2020). Let $N(X)$ be the number of primitive 3-term APs $\{a^2, b^2, c^2\}$ with $c^2 \leq X$ . Then

$N(X) = C X^{1/2} + O(X^{3/8+\varepsilon}),$

with

$C = \frac{\sqrt{2}}{\pi^2} \log(1+\sqrt{2}).$

These methods rely on analytic continuation of associated double Dirichlet series that count such progressions and connect these to counts of rational points of bounded height on the circle $x^2 + y^2 = 2$ (Hulse et al., 2020).

5. Progressions over Number Fields and Extensions

Over quadratic fields, five-term APs of squares exist, and their classification has been recently elucidated (González-Jiménez, 3 Feb 2026). The existence/nonexistence of longer sequences:

Genus 5 curves encode the constraints for 5-term APs. A reduction to genus 1 quotients and arithmetic on elliptic curves is then used to classify all possibilities.
Under certain arithmetic hypotheses (class number one, low Mordell-Weil rank, etc.), the only non-elementary 5-term APs arise in specific exceptional cases; a unique 6-term progression exists only for $K=\mathbb{Q}(\zeta_{12}), L=K(\sqrt{5})$ .
Over cubic fields, no 5-term AP of squares exists. This is proven via an analysis of the rational points on a specific genus 3 hyperelliptic curve, establishing a strong upper bound on possible progression length (Bremner et al., 2015).

6. Perfect Powers and Sums of Squares in AP

Certain Diophantine equations involving sums of squares in AP equaling perfect powers, i.e.,

$\sum_{i=1}^d (x + ir)^2 = y^n,$

have been classified for $2 \le d \le 10$ and $r$ up to $10^4$ using factorization arguments and the Bilu–Hanrot–Voutier primitive divisors theorem. Only in the cases $d=2,3,6$ do nontrivial primitive solutions exist, with very specific patterns depending on the values of $d$ and $n$ (Koutsianas et al., 2018, Kundu et al., 2018).

7. Open Problems, Conjectures, and Modern Directions

Rudin’s conjecture asserts that the maximal number of squares within the first $N$ terms of any non-trivial integer AP is $O(\sqrt{N})$ . Its (super-)strong variant posits that for $N \geq 6$ , $Q(N) = Q(N; 24, 1)$ , i.e., the sequence $24n+1$ is uniquely extremal. This is proven for all $N \leq 52$ , with the progression $24n+1$ the maximal at all critical jump points (González-Jiménez et al., 2013, Granville, 2017).

The current understanding of longer APs of squares in number fields of larger degree, the asymptotic growth of maximal number of squares in AP as a function of progression length ( $Q(N)$ ), and the potential for extraordinary behavior in explicit families (e.g., over function fields) remain areas of fertile research. Uniform bounds are expected, but proving them would require new insights in diophantine geometry and arithmetic dynamics.

Key References: