🔢 mathematics

Replacement dynamics of binary quadratic forms

This paper investigates the dynamics of replacing an entry in a vector with the output of a multivariate function, specifically classifying rational periodic vectors for diagonal binary quadratic forms up to period 5 and proving the non-existence of period-4 vectors for a specific non-univariate type.

Original authors: Raghav Bhutani, Frederick Saia

Published 2026-04-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Raghav Bhutani, Frederick Saia

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a pair of numbers, like a pair of dice: $(x, y)$ . Now, imagine you have a special rule, a "magic formula," that takes these two numbers, does some math with them, and spits out a single new number. Let's call this formula $f(x, y)$ .

In this paper, the authors are playing a very specific game with these numbers. Here is how the game works:

The Game: "The One-Step Replacement"

Start: You have a pair of numbers, say $(3, 5)$ .
The Rule: You apply your magic formula to get a new number. Let's say $f(3, 5) = 10$ .
The Move: You must replace exactly one of your original numbers with this new result.
- Option A (Left): Replace the first number. Your new pair is $(10, 5)$ .
- Option B (Right): Replace the second number. Your new pair is $(3, 10)$ .
Repeat: You take your new pair, apply the formula again, and replace one number again. You keep doing this forever.

The authors are asking a big question: Can you get stuck in a loop?

If you keep playing, will you eventually return to a pair of numbers you've seen before? If you do, you've found a "periodic vector." It's like a hamster running on a wheel; eventually, it comes back to the exact same spot.

The Two Types of Loops

The authors discovered that these loops come in two flavors, which they call "Univariate" and "Non-Univariate."

1. The "Univariate" Loops (The Boring, Predictable Ones)

Imagine you decide to always replace the left number. You never touch the right one.

Start: $(x, y)$
Step 1: $(f(x,y), y)$
Step 2: $(f(f(x,y), y), y)$

Notice that the right number, $y$ , never changes. It's just sitting there, watching. The left number is doing all the work, acting exactly like a single-variable math problem (like $x^2 + c$ ).

The authors found that for these types of loops, the rules are already well-known to mathematicians. In fact, for loops that take 4 or 5 steps to repeat, they don't exist for rational numbers (numbers you can write as fractions). It's like trying to find a square circle; the math just says "nope."

2. The "Non-Univariate" Loops (The Wild, New Ones)

This is where the paper gets exciting. What if you mix it up?

Step 1: Replace the Left number.
Step 2: Replace the Right number.
Step 3: Replace the Left number again.
Step 4: Replace the Right number again.

Now, both numbers are changing and influencing each other. They are dancing together. This is a "Non-Univariate" loop. The authors call the pattern of Left/Right replacements the "Type" of the loop.

The paper focuses on two specific dance patterns (Types) that had never been solved before:

The "Left-Right-Left-Right" Dance (Period 4): A 4-step loop where you alternate sides.
The "Left-Left-Right-Left-Right" Dance (Period 5): A 5-step loop with a slightly more complex rhythm.

The Big Discovery

The authors treated these dances like a treasure hunt. They built a giant mathematical map (a "moduli space") that shows every possible pair of numbers $(x, y)$ and every possible formula $(C, D)$ that could create these loops.

What they found:

For the 4-step dance (Left-Right-Left-Right):
They proved that no such loop exists using rational numbers. No matter how you tweak your formula, you cannot find a pair of fractions that will dance in this specific 4-step rhythm and return to the start. It's like trying to find a specific key that opens a lock, but the lock is welded shut.
For the 5-step dance (Left-Left-Right-Left-Right):
They couldn't prove it's impossible, but they did something even cooler. They reduced the entire infinite search for these loops down to finding a single "smooth" point on a very complicated, twisted shape (a curve with a high "genus," or number of holes).
- They used a computer to search for these points.
- They looked at millions of possibilities.
- Result: They found nothing.
- Conclusion: They strongly conjecture (strongly guess) that these loops also don't exist for rational numbers.

Why Does This Matter?

In the world of math, there is a famous guess (Poonen's Conjecture) that says: "If you have a simple quadratic formula (like $x^2 + c$ ), you can never find a rational number that loops with a period of 4 or longer."

This paper takes that idea and asks: "What happens if we have two numbers instead of one, and we swap them around?"

They found that even with this added complexity, the universe seems to agree with the original guess. The "wild" new dances they invented (the non-univariate types) also seem to be impossible to perform with rational numbers.

The Takeaway

Think of the authors as explorers mapping a new continent.

They knew the "Old World" (single number loops) had no treasure for periods 4 and 5.
They sailed to the "New World" (two-number loops).
They found two new islands (the 4-step and 5-step mixed dances).
They dug deep, used high-tech computers, and found no treasure (no rational solutions) on these islands either.

They haven't proven it with 100% absolute certainty for the 5-step island yet (that's a very hard math problem), but the evidence is so strong that they are betting their reputation on it: The treasure isn't there.

1. Problem Statement and Motivation

The paper investigates a specific generalization of arithmetic dynamics from univariate polynomials to multivariate settings.

Context: In standard arithmetic dynamics, one studies the iteration of a single-variable polynomial $f(x)$ , looking for rational periodic points where $f^{(N)}(x) = x$ . A major open problem (Poonen's Conjecture) posits that for quadratic polynomials over $\mathbb{Q}$ , there are no rational periodic points of period $N \geq 4$ .
The Gap: There is no natural notion of "iteration" for multivariate functions $f: S^m \to S$ because the output is a scalar, not a vector of the same dimension.
The Proposed Framework: The authors introduce "replacement dynamics." Given a vector $v = (v_1, \dots, v_m) \in S^m$ and a function $f$ , a single step involves replacing exactly one entry $v_j$ with the scalar output $f(v)$ . This generates a directed graph of iterates.
Objective: The paper focuses on rational periodic vectors for diagonal binary quadratic forms $f(x, y) = Cx^2 + Dy^2$ over $\mathbb{Q}$ . The goal is to classify these periodic vectors based on their "type" (the sequence of coordinates replaced) and determine existence conditions for periods $N \leq 5$ .

2. Methodology

The authors employ a combination of dynamical systems theory, algebraic geometry, and computational number theory.

A. Definition of Period Types

A periodic vector is classified by its type $t = (t_1, \dots, t_N) \in \{1, \dots, m\}^N$ , representing the sequence of indices replaced to return to the initial vector.

Equivalence: Types are considered modulo rotation (cyclic shift) and permutation of indices (symmetric group action). These equivalence classes are referred to as necklaces.
Univariate vs. Non-Univariate: A type is univariate if it is equivalent to a sequence of replacements on a single coordinate (e.g., all Left, or all Left then all Right). Non-univariate types involve alternating replacements on different coordinates.

B. Algebraic Formulation (Modular Polynomials)

To find periodic vectors, the authors construct dynamic modular polynomials $\Phi_{t, L}$ and $\Phi_{t, R}$ .

These polynomials are defined over the function field $K = \mathbb{Q}(c, d)$ (where $c, d$ are the coefficients of the quadratic form).
The condition for a vector $(x, y)$ to be periodic of type $t$ is that it must be a common root of the system:
$\Phi_{t, L}(x, y) = 0 \quad \text{and} \quad \Phi_{t, R}(x, y) = 0$
Factorization: For univariate types, these polynomials factorize into products of lower-degree polynomials corresponding to sub-periods. For non-univariate types, the authors prove these polynomials are generally irreducible over $K$ .

C. Geometric Analysis

The intersection of the varieties defined by $\Phi_{t, L}$ and $\Phi_{t, R}$ defines a curve (or zero-dimensional set) $Y_t$ .
The authors analyze the genus of these curves. If the genus is 0 or 1, rational points can often be parameterized. If the genus is high ( $\geq 2$ ), Faltings' Theorem implies finiteness, and finding rational points becomes a difficult Diophantine problem.
Computational Tools: The authors extensively use the Magma computer algebra system to compute Gröbner bases, factorize polynomials, determine curve genera, and search for rational points on high-genus curves.

3. Key Contributions and Results

The paper provides a complete classification of rational periodic vectors for diagonal binary quadratic forms for periods $N \leq 5$ .

A. Classification of Types

The authors distinguish between univariate and non-univariate types:

Univariate Types: These reduce to the dynamics of single-variable quadratic polynomials.
- Result: For periods $N \geq 4$ , no rational periodic vectors exist for univariate types (relying on known results by Morton, Flynn-Poonen-Schaefer, and Poonen).
- Periods 1, 2, 3: The authors characterize the existence of rational vectors for these periods, showing they correspond to rational points on specific families of dynamical modular curves (conics or genus 0/1 curves).
Non-Univariate Types: These are the novel cases where replacements alternate between $x$ and $y$ .
- Period 4: The only non-univariate type is $(L, R, L, R)$ .
- Period 5: The only non-univariate type is $(L, L, R, L, R)$ .

B. Main Theorems

Theorem 1.7 (Period 4 - Type $L, R, L, R$ ):

A binary quadratic form $f(x, y) = Cx^2 + Dy^2$ has a periodic vector of type $(L, R, L, R)$ if and only if $C \neq D$ .
The existence of such a vector is equivalent to the existence of a rational root of a specific degree 4 polynomial $R_{C,D}(y) \in \mathbb{Z}[y]$ .
Corollary 3.11: There are no rational periodic vectors of type $(L, R, L, R)$ $(L, R, L, R)$ over $\mathbb{Q}$ $Q$ .
- Proof Sketch: The condition for a rational root leads to a singular projective curve of geometric genus 2. The authors compute that this curve has no non-singular rational points (using LMFDB data and Magma), implying no such $C, D$ exist.

Theorem 1.7 (Period 5 - Type $L, L, R, L, R$ ):

Existence is equivalent to a root of a degree 10 polynomial $S_{C,D}(y)$ .
Conjecture 3.14: There are no rational periodic vectors of type $(L, L, R, L, R)$ $(L, L, R, L, R)$ over $\mathbb{Q}$ $Q$ .
- Evidence: The associated curve has geometric genus 11. A computational search for rational points up to height $10^7$ yielded none.

C. Structural Insights

Degree Bounds: The coordinates of a periodic vector of type $(L, R, L, R)$ generate a number field of degree at most 4. For type $(L, L, R, L, R)$ , the degree is at most 10.
Stratification: The paper establishes that determining periodic vectors for non-univariate types reduces to finding rational points on specific high-genus curves, distinct from the univariate case.

4. Significance

Extension of Arithmetic Dynamics: The paper successfully extends the study of periodic points from univariate polynomials to a well-defined multivariate "replacement" framework, creating a new class of Diophantine problems.
Resolution of Low-Period Cases: It provides a definitive classification for periods up to 5 for diagonal binary quadratic forms. It confirms that, similar to the univariate case, rational periodic points of period $\geq 4$ likely do not exist, even in this multivariate setting.
Connection to Modular Curves: The work introduces "dynamical modular curves" for multivariate systems. The non-existence proofs for periods 4 and 5 rely on the arithmetic geometry of these curves (genus 2 and 11, respectively), demonstrating the power of modern computational algebraic geometry in solving dynamical problems.
Computational Verification: The paper sets a precedent for using computational tools (Magma, LMFDB) to verify the non-existence of rational points on high-genus curves as a method for proving non-existence of periodic orbits in dynamical systems.

In summary, the paper demonstrates that while the "replacement dynamics" framework introduces new complexity (non-univariate types), the fundamental arithmetic obstruction preventing high-period rational cycles in quadratic dynamics (conjectured by Poonen) appears to persist in this multivariate generalization.