🔢 mathematics

Distribution of mixed character sums and extremal problems for Littlewood polynomials

This paper establishes distributional results for mixed character sums and applies them to construct Littlewood polynomials with record-breaking Mahler measures and $L_1$ norms, while also resolving a conjecture regarding the asymptotic minimization of Turyn polynomials' $L_{2k}$ norms.

Original authors: Jonathan W. Bober, Oleksiy Klurman, Besfort Shala

Published 2026-03-17

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jonathan W. Bober, Oleksiy Klurman, Besfort Shala

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 are a chef trying to bake the perfect cake. In the world of mathematics, this "cake" is a special type of polynomial (a mathematical recipe made of numbers) called a Littlewood Polynomial. The ingredients are simple: you can only use $+1$ or $-1$ as your coefficients.

The goal of this paper is to figure out how to bake these cakes so they are as "flat" and "uniform" as possible, or conversely, how to make them as "rich" and "large" as possible in specific ways.

Here is a breakdown of what the authors did, using simple analogies:

1. The Problem: The "Noisy Radio" and the "Perfect Cake"

Imagine you have a radio that picks up signals from a chaotic universe.

The Signal: A mathematical sum that mixes two types of waves: one that repeats in a pattern (like a drumbeat) and one that oscillates smoothly (like a sine wave).
The Chaos: The authors are studying what happens when you randomly change the "drumbeat" pattern (using random number sequences called characters) or when you randomly change the "tuning" of the radio (the angle $\theta$ ).

Usually, when you mix random signals, they cancel each other out, leaving you with silence (or a very small number). The authors wanted to know: If we mix these signals in a specific way, do they ever create a surprisingly loud, consistent, or "flat" sound?

2. The Discovery: Finding the "Sweet Spot"

The authors proved that if you look at these mixed signals over a long period, they don't just behave randomly. They settle into a predictable statistical pattern.

Think of it like this: If you drop a million grains of sand on a beach, you can't predict where one specific grain will land. But you can predict the shape of the pile that forms. The authors found the exact "shape" of the pile formed by these mathematical signals.

They discovered that for certain settings, the signals behave like a specific random process (a mathematical model of randomness). This allowed them to predict the behavior of these complex sums with high precision.

3. Application A: The "Mahler Problem" (Baking the Richest Cake)

One famous problem in math asks: What is the "flattest" or "richest" cake you can bake using only $+1$ and $-1$ ingredients?

The Metric: Mathematicians measure the "richness" of a cake using something called the Mahler Measure. Think of this as the average "taste" of the cake. You want a cake where the taste is consistently high everywhere, not just in one spot.
The Old Record: For a long time, the best anyone could do was a "richness" score of about 0.951.
The New Record: Using their new "statistical pile" model, the authors found a way to tweak the recipe (by shifting the starting point of the ingredients). They baked a cake with a richness score of 0.954.
- Why it matters: It's a small number, but in this field, breaking the record by 0.003 is a huge deal. It proves that we can get closer to the theoretical limit of how "perfect" these cakes can be.

4. Application B: The "Gunther-Schmidt Conjecture" (Finding the Flattest Cake)

There is another problem: Where is the "flattest" part of the cake?
Imagine a cake that is mostly flat but has a bump in the middle. You want to know exactly where to cut the cake so the piece you get is perfectly flat.

The Guess: A previous mathematician guessed that the flattest spot always happens at a specific angle (1/4 of the way around the circle).
The Proof: The authors used their new statistical model to prove this guess was 100% correct for all types of these cakes. They showed that if you cut the cake at the 1/4 mark, it is indeed the flattest possible piece.

5. The Secret Weapon: The "Log-Integrability" Trick

The hardest part of their work wasn't just finding the pattern; it was dealing with the "holes" in the cake.

The Problem: Sometimes, the cake has a tiny crumb missing (a value very close to zero). In math, if you try to measure the "average taste" using logarithms, a zero value breaks the calculation (it becomes infinity).
The Solution: The authors invented a new, gentle way to handle these tiny crumbs. They created a "safety net" (a mathematical smoothing technique) that allows them to ignore the tiny zeros without messing up the rest of the calculation. This allowed them to prove their results even when the numbers got dangerously close to zero.

Summary

In short, Jonathan Bober, Oleksiy Klurman, and Besfort Shala:

Mapped the chaos: They figured out the statistical shape of complex, mixed mathematical signals.
Broke a record: They used this map to bake a "Littlewood Polynomial" that is richer and more uniform than any previously known.
Solved a mystery: They proved exactly where the "flattest" part of these mathematical cakes is located.
Invented a tool: They created a new method to handle mathematical "zeros" that was previously very difficult to manage.

They didn't just solve a puzzle; they gave future mathematicians a better map and a sharper tool to explore the landscape of numbers.

1. Problem Statement and Motivation

The paper addresses two primary interconnected problems in analytic number theory and the theory of polynomials:

Distribution of Mixed Character Sums: The authors investigate the statistical distribution of the mixed character sums:
$S(\chi, x, \theta) = \sum_{n \le x} \chi(n) e(n\theta)$
where $\chi$ is a Dirichlet character modulo $q$ , and $e(x) = e^{2\pi i x}$ . They analyze two regimes:
- Random Character: $\chi$ varies over characters mod $q$ while $\theta$ is fixed (or rational/irrational).
- Random Shift: $\chi$ is fixed (specifically quadratic) while $\theta$ is chosen uniformly at random in $[0, 1]$ .
  The focus is on the range where the sum length $x$ is proportional to the modulus $q$ (i.e., $x \approx \beta q$ ).
Extremal Problems for Littlewood Polynomials: Littlewood polynomials are polynomials with coefficients $\pm 1$ . The paper seeks to maximize the Mahler measure ( $M_0$ ) and the $L_1$ norm of these polynomials.
- Mahler Problem: Determine $\limsup_{n \to \infty} M_0(P_n)/\sqrt{n}$ . It is known that $M_0(P) < \|P\|_2 = \sqrt{n}$ , but the exact supremum is unknown. Previous records were around $0.951$.
- Günther-Schmidt Conjecture: A conjecture regarding the minimum of the normalized $L_{2k}$ norms of Turyn polynomials (a specific class of character sums) occurring at a shift $\alpha = 1/4$ .

2. Methodology

The authors develop a novel probabilistic framework combining analytic number theory with stochastic processes.

A. Poisson Summation and Random Processes

The core technical tool is the application of the quantitative Poisson summation formula to transform the discrete character sums into conditionally convergent series.

They define a normalized process $F_{k, \chi, \alpha, \beta}(t)$ derived from the sum $S(\chi, \alpha, \beta, \theta)$ where $\theta = (k+t)/q$ .
Using the method of moments, they prove that as $q \to \infty$ $q \to \infty$ , these discrete processes converge in distribution to explicit continuous random processes defined by infinite series involving random variables.
- Case 1 (Random $\chi$ ): Converges to a process $F_{k, \alpha, \beta}(t)$ involving a Steinhaus random variable (complex unit circle) and a completely multiplicative Steinhaus sequence $W(l)$ .
- Case 2 (Random $\theta$ , fixed quadratic $\chi$ ): Converges to a process $G_{\alpha, \beta}(t)$ involving independent Rademacher random variables (uniform on $\{-1, 1\}$ ).

B. Handling "Log-Integrability"

A significant hurdle in applying these distributional results to the Mahler measure is that the functional $\ell(f) = \int_0^1 \log|f(t)| dt$ is not continuous on the space of continuous functions $C[0,1]$ . Small values of $|f(t)|$ can cause the integral to diverge or behave erratically.

Proposition 1 & 8: The authors introduce a general lemma to handle this. They prove that if a function $f$ satisfies a linear differential inequality (e.g., $|\Delta f(t)| \ge 1$ for some differential operator $\Delta$ ), then the measure of the set where $|f(t)| < \epsilon$ is small ( $O(\epsilon^{1/2k})$ ).
This allows them to bound the "log-integrability" error term uniformly, justifying the interchange of limits and expectations required to compute the Mahler measure.

C. Prime Removal and Polynomial Coincidences

To ensure the convergence of moments for the random processes, the authors must avoid "pathological" primes where the multiplicative structure of $\chi$ creates unexpected algebraic coincidences (e.g., solutions to polynomial equations modulo $q$ ).

They utilize Lemma 6 and 7 to show that for sequences $k_q$ growing with $q$ , the set of primes where such coincidences occur has density zero. This allows them to restrict their convergence theorems to a "full density" subset of primes.

3. Key Results

Theorem 1 & 2: Distributional Convergence

Theorem 1: Establishes the weak convergence of the random processes $F_{k, q, \alpha, \beta}$ to limiting processes $F_{k, \alpha, \beta}$ (fixed $k$ ) or $F_{\alpha, \beta}$ (growing $k$ ) as $q \to \infty$ over primes.
Theorem 2: Proves that for a fixed quadratic character $\chi_q$ , the process $G_{q, \alpha, \beta}$ (varying shift) converges to the random process $G_{\alpha, \beta}$ .

Corollary 2 & Theorem 3: New Records for Mahler Measure

By applying the distributional results to the specific parameters $\alpha = 0.2$ and $\beta = 1.1$ (generalized Turyn polynomials), the authors compute the limiting expected Mahler measure.

Result: They construct Littlewood polynomials of arbitrarily large degree with a normalized Mahler measure $> 0.954$ .
This improves upon the previous record of $\approx 0.951$ established by Mossinghoff (2024).
They also improve the lower bound for the $L_1$ norm ratio to $0.9783\dots$ .

Theorem 4: Proof of the Günther-Schmidt Conjecture

The authors prove that for every integer $k \ge 2$ , the function $\phi_k(\alpha)$ describing the limiting normalized $L_{2k}$ norm of Turyn polynomials attains its minimum at $\alpha = 1/4$ .
This confirms a conjecture by Günther and Schmidt (2017) and explains why the Turyn polynomials at this shift are "flat" (minimizing $L_{2k}$ norms), which correlates with the high Mahler measure observed.

4. Significance and Impact

Resolution of Extremal Problems: The paper sets new benchmarks for the Mahler measure and $L_1$ norm of Littlewood polynomials, bringing the known lower bounds significantly closer to the theoretical upper bound of 1.
Methodological Innovation: The development of a general framework for "log-integrability" (Proposition 1/8) is a major contribution. It provides a robust tool for analyzing the Mahler measure of random and deterministic series where standard continuity arguments fail.
Unification of Fields: The work bridges the gap between the distribution of character sums (number theory) and the geometry of polynomials (complex analysis). It demonstrates that the statistical properties of character sums can be precisely modeled by specific random processes, which in turn dictate the extremal properties of polynomials.
Conjecture Verification: The proof of the Günther-Schmidt conjecture provides a deeper understanding of the behavior of Turyn polynomials and their merit factors, showing that the parameters maximizing the Mahler measure and minimizing the $L_{2k}$ norm are related but distinct (the former is optimized at $\alpha \approx 0.2$ , the latter at $\alpha = 0.25$ ).

In summary, this paper advances the understanding of the statistical behavior of character sums and leverages these insights to solve long-standing extremal problems in the theory of Littlewood polynomials, establishing new records and proving significant conjectures.