Inhomogeneous central limit theorems for the voter model occupation times

This paper extends functional central limit theorems for voter model occupation times on lattices to the case of spatially inhomogeneous product initial distributions by leveraging the duality with coalescing random walks and Donsker's invariance principle.

The Gist

Here is an explanation of the paper "Inhomogeneous central limit theorems for the voter model occupation times" using simple language, analogies, and metaphors.

The Big Picture: A Town of Shifting Opinions

Imagine a giant, infinite city grid (like a 3D chessboard) where every single house has a resident. Each resident holds one of two opinions: Red (1) or Blue (0). This is the Voter Model.

The rule of this city is simple:

• Every day, a resident looks at one of their neighbors.
• If the neighbor has a different opinion, the resident might change their mind to match them.
• This happens randomly and continuously.

Over time, the opinions spread like a rumor or a virus. Sometimes a whole neighborhood turns Red; sometimes Blue.

The Question: How Much Time Does a House Spend Being Red?

The author, Xiaofeng Xue, is interested in a specific question: If we pick one specific house (let's call it "House Zero"), how much time does it spend holding the Red opinion over a long period?

In math terms, this is called the Occupation Time.

The Twist: The City Isn't Uniform

In previous studies, scientists assumed the city was "homogeneous." This means that at the very beginning, every house had the same chance (say, 50%) of being Red, regardless of where it was located. It was a perfectly mixed bag.

This paper asks a more realistic question: What if the city is inhomogeneous?

• Maybe the downtown area starts with 90% Red supporters.
• Maybe the suburbs start with 10% Red supporters.
• Maybe the opinion density changes smoothly as you move from the city center to the countryside.

The author wants to know: If we start with this messy, uneven distribution of opinions, what happens to the "Red time" of House Zero as we zoom out and look at the big picture?

The Magic Scale: The "N" Factor

To answer this, the author uses a trick called scaling. Imagine taking a photo of the city and zooming out.

• As you zoom out, the individual houses become tiny dots.
• The "time" we watch them speeds up.
• The "density" of opinions smooths out into a continuous curve (like a heat map).

The paper proves that even with this messy, uneven starting point, the total time House Zero spends being Red follows a very predictable pattern called a Central Limit Theorem (CLT).

What is a CLT?
Think of it like rolling dice. If you roll one die, the result is random. If you roll 1,000 dice and add them up, the result isn't random anymore; it forms a perfect "Bell Curve" (a Gaussian distribution).

• The Paper's Finding: Even though the starting opinions are uneven, if you watch House Zero long enough and scale things correctly, the fluctuations in its "Red time" will eventually look like a Bell Curve (or a specific type of random motion called Brownian Motion).

The Three Dimensions: Different Rules for Different Cities

The paper finds that the "shape" of this randomness depends entirely on how many dimensions the city has. This is the most fascinating part:

1. 4D and Higher (The "High-Dimensional" City):

   • In a city with 4 or more dimensions, people are so spread out that they rarely bump into each other.
   • The Result: The randomness behaves like a standard Brownian Motion (like a drunk person walking in a straight line, but with a twist). The "drunkness" (volatility) changes over time depending on the local density of opinions, but the path is smooth and predictable.

2. 3D (The "Real World" City):

   • In our 3D world, people bump into each other more often.
   • The Result: The randomness is more complex. It's not a simple Brownian motion. It's a "Gaussian process without independent increments."
   • Analogy: Imagine a drunk person walking in 3D. In 4D, their steps are independent. In 3D, their current step depends heavily on where they were a moment ago because the "crowd" (the opinions) is denser and more interconnected. The path is "sticky" and correlated.

3. 2D (The "Flatland" City):

   • The paper mentions this briefly, noting it behaves differently again (involving logarithms), but the main focus is on 3D and 4D+.

The Secret Weapons: How They Proved It

The author didn't just guess; they used two powerful mathematical "superpowers" to prove this:

1. The "Shadow" Game (Duality):

   • The Voter Model is hard to track directly. But it has a "shadow twin" called the Coalescing Random Walk.
   • Analogy: Instead of watching the opinions change, imagine two invisible ghosts starting at House Zero and a neighbor. They wander randomly. If they bump into each other, they merge into one ghost.
   • The math shows that the probability of House Zero being Red is exactly the same as the probability that these ghosts haven't merged yet. This turns a complex opinion problem into a simpler "ghost walking" problem.

2. The "Smoothie" Effect (Donsker's Invariance Principle):

   • This is a famous math theorem that says: "If you take a random walk and zoom out enough, it looks like a smooth, continuous curve (Brownian Motion)."
   • The author uses this to show that even though the starting opinions were uneven (like a lumpy smoothie), once you zoom out and let time pass, the "lumps" smooth out into a perfect curve.

The Takeaway

This paper is a bridge between chaos and order.

• Chaos: The starting opinions are messy, uneven, and change based on location.
• Order: Despite the mess, the long-term behavior of a single house's opinion follows a strict, beautiful mathematical law (a Gaussian distribution).

The author shows us that even in a world where everyone starts with different biases, the collective behavior of the system eventually settles into a predictable rhythm. It's a testament to the power of probability: No matter how uneven the starting line is, the finish line (the statistical limit) looks remarkably similar for everyone.

Technical Summary

Here is a detailed technical summary of the paper "Inhomogeneous central limit theorems for the voter model occupation times" by Xiaofeng Xue.

1. Problem Statement

The paper investigates the asymptotic behavior of occupation times for the voter model on the $d$-dimensional integer lattice $\mathbb{Z}^d$. Specifically, it addresses the functional Central Limit Theorem (CLT) for the centered occupation time process of the origin, denoted as $\xi^N_t$, under a spatially inhomogeneous initial distribution.

• The Model: The voter model is a continuous-time Markov process where each site $x \in \mathbb{Z}^d$ holds an opinion $\eta_t(x) \in \{0, 1\}$. Sites update their opinion by adopting the opinion of a randomly chosen neighbor at rate 1.
• The Initial Condition: Unlike previous studies (e.g., reference [8]) that assumed a homogeneous initial distribution (where the probability of a site holding opinion '1' is a constant $p$), this paper considers an initial distribution $\nu_{\rho, N}$ which is a product measure with spatially varying probabilities:
  $$\nu_{\rho, N}(\eta(x) = 1) = \rho\left(\frac{x}{\sqrt{N}}\right)$$
  Here, $\rho: \mathbb{R}^d \to (0, 1)$ is a uniformly continuous function. This setup models a system where the initial density of opinions varies smoothly across space on a macroscopic scale.
• The Object of Study: The centered occupation time process at the origin:
  $$\xi^N_t = \int_0^t \left( \eta_s(0) - \mathbb{E}_{\nu_{\rho, N}}[\eta_s(0)] \right) ds$$
  The goal is to determine the weak limit of the rescaled process $\frac{1}{h_d(N)} \xi^N_{tN}$ as $N \to \infty$, where $h_d(N)$ is a dimension-dependent scaling factor.

2. Methodology

The author employs a combination of probabilistic tools, extending techniques used for the homogeneous case to handle spatial inhomogeneity.

• Duality: The core of the proof relies on the duality relationship between the voter model and coalescing random walks. The state of the voter model at time $t$ can be determined by tracing back the ancestry of particles to time 0.
• Martingale Decomposition: The occupation time process is decomposed using Dynkin's martingale formula. The process is split into a martingale term and a remainder term. The convergence of the remainder term to zero is established, leaving the martingale term as the primary driver of the limit.
• Poisson Flow Method: To analyze the quadratic variation of the martingale, the paper utilizes the Poisson flow method (introduced in reference [1]). This involves representing the voter model dynamics via Poisson processes governing opinion updates.
• Donsker's Invariance Principle: A critical technical step involves estimating the term $\mathbb{E}_{\nu_{\rho, N}}[(\eta_s(x) - \eta_s(y))^2]$ for neighboring sites $x \sim y$. In the homogeneous case, this is a constant. In the inhomogeneous case, the author uses Donsker's invariance principle for simple random walks to show that this term converges to a function involving the heat kernel solution $\varrho(s, u)$, where $\varrho$ solves the heat equation with initial condition $\rho$.
• Tightness: The tightness of the sequence of processes is established using moment bounds (specifically fourth moments for $d \ge 4$ and second moments for $d=3$), relying on the fact that the initial distribution is a product measure.

3. Key Contributions

The primary contribution is the extension of functional CLTs from homogeneous to spatially inhomogeneous initial conditions for the voter model.

1. Generalization of Initial Conditions: The paper proves that the functional CLT holds even when the initial density of opinions is not constant but varies according to a smooth function $\rho(x/\sqrt{N})$.
2. Identification of Limit Processes:
   • For $d \ge 4$, the limit is shown to be a stochastic integral with respect to a standard Brownian motion, where the integrand is a time-dependent function derived from the solution to the heat equation.
   • For $d = 3$, the limit is a Gaussian process without independent increments, characterized by a specific covariance structure involving the heat kernel and the initial density profile.
3. Handling Inhomogeneity in Quadratic Variation: The paper provides rigorous estimates for the quadratic variation of the martingale in the inhomogeneous setting, demonstrating how the spatial variation of $\rho$ propagates through the coalescing random walk duality to affect the limiting variance.

4. Main Results

The paper establishes two main theorems based on the dimension $d$:

Case $d \ge 4$ (Theorem 2.1)

Let $h_d(N)$ be the scaling factor defined as $\sqrt{N}$ for $d \ge 5$ and $\sqrt{N \log N}$ for $d=4$.
The rescaled process converges weakly in $C[0, T]$ to:
$$\left\{ \int_0^t A_{s, d} \, dB_s : 0 \le t \le T \right\}$$
where $B_t$ is a standard Brownian motion and $A_{s, d}$ is a deterministic function:

• For $d=4$: $A_{s, 4} = \sqrt{\frac{\gamma_4 \varrho(s, 0)(1-\varrho(s, 0))}{\pi^2}}$
• For $d \ge 5$: $A_{s, d} = \sqrt{4d\gamma_d \left(\int_0^\infty \theta p_{\theta, d}(0,0) d\theta\right) \varrho(s, 0)(1-\varrho(s, 0))}$

Here, $\varrho(s, u) = \mathbb{E}[\rho(W^d_{2s} + u)]$ is the solution to the heat equation, and $\gamma_d$ is the probability that a simple random walk never returns to the origin.
Significance: The limit is a time-inhomogeneous diffusion. The volatility of the limit process depends on the local "temperature" (variance of the opinion) $\varrho(s, 0)(1-\varrho(s, 0))$ at the origin.

Case $d = 3$ (Theorem 2.2)

The scaling factor is $N^{3/4}$. The rescaled process converges weakly to:
$$\left\{ \sqrt{12\gamma_3} \zeta_t : 0 \le t \le T \right\}$$
where $\zeta_t$ is a centered Gaussian process with continuous sample paths. Its covariance function is:
$$\text{Cov}(\zeta_{t_1}, \zeta_{t_2}) = \int_0^{t_1} \left( \int_{\mathbb{R}^3} b_{t_1}(s, u) b_{t_2}(s, u) \varrho(s, u)(1-\varrho(s, u)) \, du \right) ds$$
where $b_t(s, u)$ is an integral kernel related to the heat kernel.
Significance: Unlike $d \ge 4$, the limit is not a simple stochastic integral with independent increments. The spatial inhomogeneity creates long-range correlations in the occupation time fluctuations.

5. Significance and Implications

• Theoretical Advancement: This work bridges the gap between homogeneous interacting particle systems and those with spatially varying initial conditions. It demonstrates that the macroscopic behavior of the voter model is robust to spatial inhomogeneity, provided the inhomogeneity is smooth.
• Connection to Heat Equation: The results explicitly link the fluctuations of the voter model to the solution of the heat equation ($\varrho$). This highlights a deep connection between the microscopic dynamics of opinion spreading and macroscopic diffusion processes.
• Analogy to Simple Exclusion Process (SEP): The paper draws parallels to the Simple Exclusion Process, where similar inhomogeneous CLTs have been studied. The voter model results in $d \ge 4$ serve as an analogue to the SEP results, suggesting universality in the behavior of occupation times for different interacting particle systems under inhomogeneous initial states.
• Methodological Robustness: The techniques developed, particularly the estimation of the quadratic variation term under inhomogeneity using Donsker's principle, provide a template for analyzing other interacting particle systems with non-constant initial measures.

In summary, the paper successfully generalizes the functional CLT for voter model occupation times to spatially inhomogeneous settings, revealing that the limiting process is a Gaussian process whose volatility and correlation structure are governed by the heat equation solution derived from the initial density profile.