🔢 mathematics

The divisor function for matrices

This paper introduces a matrix divisor function $\tau_n(T,M)$ that counts integer matrix factorizations of bounded height and establishes both asymptotic formulas for fixed non-singular or zero matrices and essentially sharp uniform upper bounds for arbitrary non-singular matrices using lattice point counting techniques.

Original authors: Tim Browning, Nikita P. Kalinin, Alina Ostafe, Damaris Schindler, Lena Wurzinger

Published 2026-08-27

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Tim Browning, Nikita P. Kalinin, Alina Ostafe, Damaris Schindler, Lena Wurzinger

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

In the vast landscape of number theory, mathematicians have long been fascinated by the ways integers can be broken down into smaller pieces. The most famous example is the divisor function, which simply counts how many ways a whole number can be written as the product of two other whole numbers. For the number twelve, for instance, the pairs are one times twelve, two times six, and three times four. This counting process reveals deep patterns about the structure of numbers themselves. But what happens when we move beyond single numbers and start looking at grids of numbers, known as matrices? A matrix is a rectangular array of numbers, and just like a single number, it can often be built by multiplying two other matrices together. The question of how many ways a specific matrix can be formed this way is far more complex, involving not just the values inside the grid but also the geometric relationships between the rows and columns. Understanding these factorizations helps mathematicians map the hidden architecture of higher-dimensional number systems, a field that connects pure arithmetic with the geometry of space.

A team of researchers has now taken a significant step forward in understanding this matrix version of the divisor function. They focused on counting the number of pairs of integer matrices that multiply together to produce a specific target matrix, while keeping the size of the numbers inside those matrices below a certain limit. Imagine a growing box that contains all possible integer matrices whose entries do not exceed a value $T$ . As this box gets larger, the researchers wanted to know exactly how many pairs of matrices inside it could combine to form a specific result. Their work provides a precise formula for predicting this count when the target matrix is a standard, non-zero grid, and also when the target is a grid filled entirely with zeros.

The researchers discovered that for a fixed, non-zero target matrix, the number of ways it can be formed grows at a very specific rate as the size limit increases. This growth follows a predictable power law, meaning the count increases in a steady, calculable fashion rather than randomly. The exact speed of this growth depends on the dimensions of the matrices and the specific properties of the target matrix itself. To reach this conclusion, the team treated the problem as a question of counting points on a grid within a continuous geometric space. They used advanced techniques to measure the volume of the regions where these valid matrix pairs could exist, effectively translating a difficult counting problem into a problem of measuring space. This approach allowed them to prove that the number of solutions behaves in a highly regular way, confirming that the underlying structure is stable and predictable.

The study also tackled the more chaotic case where the target matrix is filled with zeros. In this scenario, the rules change because many different types of matrix pairs can result in a zero product. The team found that even here, a clear pattern emerges. They proved that the number of solutions grows at a rate proportional to the total volume of the box of possible matrices, with a very small error margin. This result is significant because it shows that even in the most degenerate case, where the target is zero, the distribution of solutions is not random but follows a strict mathematical law. The researchers also established a firm upper limit on how large this count can ever get, regardless of the target matrix chosen. This bound is essentially the best possible, meaning the count cannot grow any faster than their formula suggests.

One of the most interesting aspects of their findings is how the behavior changes depending on the size of the matrices. For two-by-two matrices, the growth rate is well understood and matches previous calculations. However, for larger matrices, the researchers found that the growth rate is significantly slower than what some earlier, less precise estimates had suggested. They showed that for larger grids, the number of ways to form a target matrix is much more constrained than previously thought. This correction is important because it refines our understanding of how these high-dimensional systems behave. The team also demonstrated that their results hold true uniformly, meaning the formulas work consistently across different types of target matrices without needing special adjustments for each one.

The methods used to reach these conclusions were rigorous and relied on a deep connection between number theory and the geometry of groups. The researchers did not simply guess or simulate the results; they provided a complete mathematical proof. They broke the problem down into smaller, manageable pieces by looking at the lattices, or grid-like structures, that the matrices create. By counting how many of these lattices fit within certain boundaries and how many matrix pairs correspond to each lattice, they were able to reconstruct the total count. This decomposition allowed them to handle the complexity of the problem without getting lost in the sheer number of possibilities. The work stands as a definitive answer to the question of how matrix factorizations are distributed, providing a solid foundation for future research in this area.

Ultimately, this paper transforms a vague question about counting matrix pairs into a precise, predictable science. It shows that even in the high-dimensional world of integer matrices, there is an underlying order that can be described with exact formulas. The researchers have not only solved the problem for specific cases but have also provided a framework that can be applied to other similar counting problems. Their work confirms that the universe of matrix factorizations is governed by clear, unbreakable laws, offering a new perspective on how numbers interact when arranged in complex grids. For anyone interested in the hidden patterns of mathematics, this study reveals that the chaos of high-dimensional counting is, in fact, a highly organized and beautiful system.

Technical Summary: The Divisor Function for Matrices

Problem Statement
The paper introduces and analyzes a matrix analogue of the classical divisor function, denoted as $\tau_n(T, M)$ . For a fixed integer $n \ge 1$ , a height parameter $T \ge 1$ , and a target matrix $M \in M_n(\mathbb{Z})$ , the function counts the number of pairs of integer matrices $(A, B)$ such that $AB = M$ , where both $A$ and $B$ have entries bounded in absolute value by $T$ (i.e., $|A|, |B| \le T$ ).

The authors investigate the asymptotic behavior of $\tau_n(T, M)$ as $T \to \infty$ in two distinct regimes:

Fixed $M$ : Determining an asymptotic formula for a specific non-singular matrix $M$ , or for the zero matrix $M = O_n$ .
Uniform in $M$ : Establishing upper bounds for $\tau_n(T, M)$ that hold for arbitrary non-singular matrices $M$ , independent of the specific entries of $M$ .

The case $n=1$ recovers the classical divisor function (up to a factor of 2), but for $n \ge 2$ , the function exhibits growth with $T$ even when $M$ is the identity matrix, distinguishing it from the scalar case.

Methodology
The paper employs a synthesis of analytic number theory, the geometry of numbers, and the theory of Lie groups.

Lattice Point Counting on $SL_n(\mathbb{R})$ : The core of the asymptotic analysis for non-singular $M$ relies on decomposing the factorization $AB=M$ via the lattice $\Lambda = A\mathbb{Z}^n$ . This reduces the problem to counting lattice points in specific subsets of $SL_n(\mathbb{R})$ . The authors utilize a general lattice point counting theorem by Gorodnik and Nevo, which applies to locally compact second countable groups. They verify that the relevant sets $E_{X,Y}(T) = \{g \in SL_n(\mathbb{R}) : |Xg| \le T, |g^{-1}Y| \le T\}$ satisfy the necessary conditions of "well-roundedness" and stable mean ergodicity.
Volume Asymptotics: To apply the counting theorem, the authors compute the asymptotic volume of the sets $E_{X,Y}(T)$ using the Cartan decomposition of $SL_n(\mathbb{R})$ . This involves analyzing the Haar measure in terms of diagonal matrices and orthogonal groups, utilizing dominated convergence to establish the existence of a limiting constant $\kappa(X, Y)$ .
Uniform Upper Bounds: For the uniform bound, the authors decompose the set of factorizations based on the intermediate lattices $\Lambda$ satisfying $M\mathbb{Z}^n \subseteq \Lambda \subseteq \mathbb{Z}^n$ . They bound the number of such lattices using group-theoretic results on subgroups of finite abelian groups (specifically $\mathbb{Z}^n / M\mathbb{Z}^n$ ). They then apply a uniform bound for the number of lattice points in $SL_n(\mathbb{R})$ within the height constraints, derived via integration over the Stiefel manifold and estimates on successive minima.
Zero Matrix Factorizations: For $M = O_n$ , the authors adapt arguments from Katznelson regarding singular matrices. They partition the solution space based on the rank of $A$ and the associated primitive lattice $\Lambda$ containing the rows of $A$ . The columns of $B$ are then constrained to lie in the orthogonal lattice $\Lambda^\perp$ . The counting involves summing over primitive lattices of various ranks, utilizing geometry of numbers estimates for lattice points in convex bodies.

Key Contributions and Results

Asymptotic Formula for Non-Singular $M$ :
For a fixed non-singular matrix $M \in M_n(\mathbb{Z})$ , the authors prove:
$\tau_n(T, M) \sim c_M T^{\lfloor n^2/2 \rfloor}$
as $T \to \infty$ . The constant $c_M$ is explicitly defined as a sum over lattices $\Lambda$ such that $M\mathbb{Z}^n \subseteq \Lambda \subseteq \mathbb{Z}^n$ , involving the volume constants $\kappa(B_\Lambda, B_\Lambda^{-1}M)$ . In the special case where $\det(M) = \pm 1$ , the sum collapses to a single term, yielding $\tau_n(T, M) \sim 2\kappa(I_n, M)T^{\lfloor n^2/2 \rfloor}$ .
Uniform Upper Bound:
For any non-singular $M \in M_n(\mathbb{Z})$ , the authors establish an essentially sharp uniform upper bound:
$\tau_n(T, M) \ll_\varepsilon T^{\lfloor n^2/2 \rfloor + \varepsilon}$
for any $\varepsilon > 0$ . The implied constant depends only on $n$ and $\varepsilon$ , not on $M$ . The paper notes that this bound cannot be improved to depend on $|M|$ without further restrictions, as demonstrated by the case of scalar matrices $M = m^2 I_2$ .
Asymptotic Formula for the Zero Matrix:
For the zero matrix $M = O_n$ , the authors derive:
$\tau_n(T, O_n) = c_{O_n} T^{n^2} + O(T^{n^2-1})$
where $c_{O_n}$ is a positive constant derived from a sum over primitive lattices. This result confirms that the growth rate for the zero matrix is $T^{n^2}$ , significantly larger than the $T^{\lfloor n^2/2 \rfloor}$ growth for non-singular matrices.
Lower Bounds:
The paper provides lower bounds illustrating the growth of $\tau_n(T)$ (where $M=I_n$ ) as $\gg T^{\lfloor n^2/4 \rfloor}$ and establishes that for a matrix $M$ of rank $r$ , $\tau_n(T, M) \gg_M T^{\lfloor r^2/2 \rfloor + n(n-r)}$ .

Significance and Context
The paper positions its work as an extension of recent studies by Afifurrahman, who considered counting solutions to multiplicative equations $A_1 \cdots A_m = M$ . The authors note that while Afifurrahman's results provided bounds of $O(T^{n^2-n+\varepsilon})$ for non-singular $M$ , their work improves this to $O(T^{\lfloor n^2/2 \rfloor + \varepsilon})$ for $n \ge 3$ , which is a substantial refinement.

The authors explicitly state that their investigation is motivated by the desire to understand the density of solutions to Diophantine equations involving matrix products, such as $A_1B_1 + \cdots + A_mB_m = 0$ . They suggest that the uniform upper bounds derived for $\tau_n(T, M)$ could be instrumental in applying strategies similar to those of Birch, Davenport, and Lewis to such systems.

The paper acknowledges the use of generative AI tools to suggest proof outlines for the main theorems, while emphasizing that all arguments were independently verified and the final text was authored by the researchers. The results rely heavily on the interplay between the arithmetic of lattices and the analytic properties of the group $SL_n(\mathbb{R})$ .

More like this