A point process on the unit circle with mirror-type interactions

This paper investigates a point process on the unit circle with mirror-type interactions, characterizing the diverse asymptotic behaviors of its smooth linear statistics as the number of points grows, which can exhibit fluctuations of varying orders and distributions including Bernoulli, Gaussian, or mixed forms, while also deriving precise large-$n$ asymptotics for the normalization constant.

The Gist

Imagine you are hosting a party on a giant, circular dance floor (the unit circle). You invite $n$ guests, and they all want to dance. But there's a strange rule governing how they move: Mirror Magic.

In this paper, mathematician Christophe Charlier studies a very specific type of party where the guests don't just interact with each other; they interact with their reflections in a giant mirror placed along the floor's diameter (the real line).

Here is the breakdown of what happens at this party, explained simply.

1. The Setup: The Mirror Dance

Usually, in physics and math, particles (or dancers) repel each other. If you have two people on a dance floor, they tend to spread out to avoid bumping into one another. This is like the famous "Circular Beta Ensemble" where everyone tries to be as far apart as possible.

But in this paper, the rule is different.

• The Rule: A guest at position $A$ feels a strong "push" away from their reflection $A'$ (the mirror image across the floor).
• The Catch: They don't push away from each other directly. They only push away from their mirrors.

The Analogy: Imagine you are standing on a stage. You have a twin standing exactly opposite you, reflected in a mirror. You hate your twin's reflection, so you try to get as far away from the mirror as possible. But you don't care about the other people on stage; you only care about your own reflection.

2. The Big Surprise: The "All-or-Nothing" Party

The most fascinating discovery in the paper is what happens when the number of guests ($n$) gets very large.

You might expect the guests to spread out evenly around the circle to maximize their distance from their reflections. That is not what happens.

Instead, the party splits into two extreme scenarios, and the system flips a coin to decide which one happens:

• Scenario A: Everyone clusters tightly around the top of the circle (the "North Pole").
• Scenario B: Everyone clusters tightly around the bottom of the circle (the "South Pole").

The Metaphor: Imagine a room full of people holding magnets. If they are all North poles, they repel each other. But here, imagine everyone is holding a magnet that repels only the reflection in a mirror. The only way for everyone to be happy (maximize their distance from their reflection) is for the whole group to huddle together at the very top or the very bottom.

The paper proves that as the party gets bigger, the chance of finding a "mixed" crowd (some at the top, some at the bottom) becomes zero. It's either 100% Top or 100% Bottom, with a 50/50 chance for each.

3. The Fluctuations: The "Bernoulli" and the "Gaussian"

The author studies "linear statistics," which is a fancy way of asking: "If we measure a specific property of the party (like the average height of the dancers), how much does it wiggle around?"

The paper finds that the answer depends on what you are measuring, leading to four different types of "wiggles" (fluctuations):

1. The Giant Jump (Order $n$): If you measure something that is different at the top vs. the bottom (like "Are we at the North Pole?"), the result jumps wildly. It's like a light switch. It's either "On" (everyone at the top) or "Off" (everyone at the bottom). This is called a Bernoulli fluctuation. It's huge and unpredictable.
2. The Gentle Wobble (Order 1, Gaussian): If you measure something that is the same at the top and bottom, but the guests wiggle slightly around their spot, the result is a smooth, bell-curve wobble. This is a Gaussian fluctuation (like the normal distribution you see in test scores).
3. The Mixed Bag: Sometimes, the result is a mix. Imagine a coin flip (Bernoulli) that decides which bell curve you get to stand on. If the coin is Heads, you get Wobble A; if Tails, you get Wobble B.

Why is this cool? In most physics systems, as you add more particles, the randomness smooths out and becomes predictable (like a calm sea). Here, because of the mirror rule, the randomness never smooths out completely. The system stays "jumpy" and chaotic even with thousands of guests.

4. The Mathematical Magic Trick

How did the author prove this?
He used a method inspired by counting regular graphs (a problem in computer science).

• The Trick: He realized that the complex math describing the party could be simplified by changing the "coordinates" of the dance floor.
• The Decoupling: Imagine the dancers are all tangled in a knot. The author found a mathematical "untangler" (a change of variables) that straightened the knot, allowing him to calculate the probability of the "All-Top" vs. "All-Bottom" scenarios with high precision.

Summary

This paper describes a strange world where particles interact with their reflections instead of each other.

• The Result: The particles don't spread out; they clump together entirely at one of two spots.
• The Behavior: The system acts like a giant coin flip. It's either all here or all there.
• The Math: The author developed a new way to calculate the odds of these clumps and the tiny wiggles within them, showing that nature can be surprisingly "jumpy" even with huge numbers of particles.

It's a beautiful example of how changing a single rule (from "repel neighbors" to "repel reflections") completely transforms the behavior of a system, turning a calm, predictable crowd into a chaotic, all-or-nothing phenomenon.

Technical Summary

1. Problem Statement and Context

The paper investigates a specific point process on the unit circle defined by the probability measure:
$$\frac{1}{Z_n} \prod_{1 \le j < k \le n} |e^{i\theta_j} - e^{-i\theta_k}|^\beta \prod_{j=1}^n d\theta_j, \quad \theta_j \in (-\pi, \pi], \beta > 0.$$
Unlike the standard Circular $\beta$-Ensemble (C$\beta$E), where particles repel each other ($|e^{i\theta_j} - e^{i\theta_k}|^\beta$), this process features mirror-type interactions. Here, each particle $e^{i\theta_j}$ interacts with the "mirror image" of every other particle $e^{-i\theta_k}$ (reflected across the real axis).

Key Characteristics of the Model:

• Attractive Nature: The interaction is attractive. The probability density is maximized only when all points coincide at specific locations. Specifically, for $n \ge 3$, the global maxima occur when all points cluster at $i$ ($\theta = \pi/2$) or all points cluster at $-i$ ($\theta = -\pi/2$).
• Non-Ergodicity: Unlike repulsive ensembles where points spread out uniformly, this system exhibits a "bimodal" behavior. As $n \to \infty$, the system does not converge to a deterministic empirical measure. Instead, it converges in distribution to a random measure that is either a Dirac mass at $\pi/2$ or a Dirac mass at $-\pi/2$, each with probability $1/2$.
• Goal: The primary objective is to analyze the asymptotic fluctuations of smooth linear statistics of the form $S_n = \sum_{j=1}^n g(\theta_j)$ as $n \to \infty$, where $g$ is a $2\pi$-periodic smooth function. The paper also derives precise asymptotics for the normalization constant $Z_n$.

2. Methodology

The proof strategy is inspired by the work of McKay and Wormald [12] on the enumeration of regular graphs, adapted to handle the specific structure of this $n$-fold integral. The methodology proceeds in several rigorous steps:

1. Localization of the Integral:
   The authors first prove that the main contribution to the partition function integral $I(f)$ comes from two small neighborhoods of size $O(n^{-1/2+\epsilon})$ around the configurations $(\pi/2, \dots, \pi/2)$ and $(-\pi/2, \dots, -\pi/2)$. All other configurations are shown to be exponentially suppressed.

2. Taylor Expansion:
   Within these neighborhoods, the integrand is expanded. The interaction term $|e^{i\eta_j} + e^{-i\eta_k}|^\beta$ (after shifting variables $\eta_j = \theta_j - \pi/2$) is expanded using the approximation $| \cos \frac{\eta_j+\eta_k}{2} |^\beta \approx \exp(-\frac{\beta}{8}(\eta_j+\eta_k)^2 + \dots)$. This transforms the problem into analyzing an integral of a Gaussian-like function with higher-order corrections.

3. Decoupling via Linear Transformation:
   A critical technical hurdle is the quadratic term $\sum_{j<k} (\eta_j + \eta_k)^2$, which couples all variables. The authors apply a linear change of variables $\eta = Ty$, where $T = I_n - \gamma J_n/n$ (with $J_n$ being the all-ones matrix).

   • This transformation decouples the quadratic terms, converting the sum into a form involving moments of the new variables $y_j$ (specifically $\sum y_j^2$).
   • Crucially, for this mirror-type interaction, the determinant of the transformation matrix is $\det(T) = 1-\gamma \approx 1/\sqrt{2}$, which remains non-singular as $n \to \infty$. This contrasts with the companion paper on antipodal interactions where the determinant vanishes.

4. Asymptotic Evaluation:
   The transformed integral is evaluated using Lemma 2.4 (an extension of results from [12]), which provides asymptotic expansions for integrals of the form $\int \exp(P(y)) dy$ where $P(y)$ is a polynomial in the moments of $y$. The authors carefully bound the error terms to ensure the expansion holds uniformly.

3. Key Contributions and Results

A. Localization Theorem (Theorem 1.1)

The paper rigorously proves that for large $n$, the system is almost surely in one of two states:

• All points are within distance $O(n^{-1/2+\epsilon})$ of $i$.
• All points are within distance $O(n^{-1/2+\epsilon})$ of $-i$.
  The probability of any other configuration is exponentially small ($O(e^{-cn^2\epsilon})$).

B. Asymptotics of the Partition Function (Theorem 1.2)

The authors derive a precise asymptotic formula for the normalization constant $Z_n$ (and generalized integrals $I(f)$) up to the order $O(1)$ term. The formula involves a sum of two dominant terms corresponding to the two clusters ($\pi/2$ and $-\pi/2$), each containing exponential factors dependent on the derivatives of the test function $f$ at these points.

C. Fluctuations of Linear Statistics (Theorem 1.5 & 1.6)

The most significant contribution is the characterization of the fluctuations of $S_n = \sum g(\theta_j)$. The behavior depends heavily on the values of $g$ and its derivatives at the critical points $\pm \pi/2$.

The paper identifies four distinct asymptotic scenarios for the fluctuations:

1. Generic Case ($g(\pi/2) \neq g(-\pi/2)$):

   • Leading Order: Fluctuations are of order $n$ and purely Bernoulli.
   • Subleading Order: Fluctuations are of order $1$ and take the form $B N_1 + (1-B) N_2$, where $B \sim \text{Bernoulli}(1/2)$ and $N_1, N_2$ are independent Gaussians.
   • Interpretation: The system chooses one of two macroscopic states (all near $\pi/2$ or all near $-\pi/2$), causing a massive jump in the sum.

2. Case $g(\pi/2) = g(-\pi/2)$ but $g'(\pi/2) \neq g'(-\pi/2)$:

   • The $O(n)$ term vanishes.
   • Fluctuations are of order $1$ and are a mixture of two Gaussians ($B N_1 + (1-B) N_2$).

3. Case $g(\pi/2) = g(-\pi/2)$, $g'(\pi/2) \neq 0$, $g'(-\pi/2) = 0$:

   • Fluctuations are of order $1$.
   • The limit is a mixture of a Gaussian and a constant (deterministic shift).

4. Case $g(\pi/2) = g(-\pi/2)$ and $g'(\pi/2) = g'(-\pi/2) = 0$:

   • If second derivatives differ, fluctuations are of order $1$ and purely Bernoulli (scaled by the difference in second derivatives).
   • If second derivatives are also equal, the fluctuations are of order $o(1)$ (smaller than 1), requiring higher-order analysis not fully resolved here.

Key Distinction: Unlike standard repulsive ensembles (like C$\beta$E) where linear statistics typically have Gaussian fluctuations of order $\sqrt{n}$ or $1$, this attractive mirror process exhibits Bernoulli-dominated fluctuations of order $n$ in the generic case due to the bimodal nature of the underlying measure.

4. Significance and Comparison

• New Class of Point Processes: This is the first rigorous analysis of a point process with only mirror-type interactions. It fills a gap between purely repulsive systems (C$\beta$E) and systems with mixed interactions.
• Non-Standard Limit Theorems: The results challenge the intuition that linear statistics of point processes always converge to Gaussian limits. Here, the limit is a mixture of discrete (Bernoulli) and continuous (Gaussian) components, driven by the "phase transition" between the two energy minima.
• Technical Innovation: The paper demonstrates the adaptability of the McKay-Wormald method to complex-valued integrands (via characteristic functions) and attractive potentials. It highlights a crucial technical difference between mirror interactions (non-singular transformation matrix) and antipodal interactions (singular transformation matrix), the latter being studied in a companion paper [4].
• Applications: While the paper focuses on mathematical theory, the model serves as a prototype for systems with competing attractive forces or symmetry-breaking interactions, relevant in statistical mechanics and random matrix theory.

In summary, the paper provides a complete asymptotic description of a novel attractive point process, revealing a rich landscape of fluctuation behaviors ranging from order $n$ Bernoulli jumps to order $1$ Gaussian mixtures, depending on the test function's local properties.