Entrance laws for coalescing and annihilating Brownian motions

This paper characterizes the extreme points of entrance laws for systems of instantaneously annihilating or coalescing Brownian motions on the line as Pfaffian point processes and explicitly identifies their kernels.

The Gist

Imagine a long, endless highway stretching from left to right. On this highway, we have a swarm of invisible cars (particles) driving randomly, like people wandering aimlessly in a fog. This is what mathematicians call Brownian motion.

The rules of the road are simple but dramatic:

1. The Crash: When two cars bump into each other, they don't just stop; they have an instant reaction.
2. The Choice: When they crash, a coin is flipped.
   • With some probability, they annihilate (both disappear in a puff of smoke).
   • With the remaining probability, they coalesce (they merge into a single, new car).

The paper by Roger Tribe and Oleg Zaboronski asks a very specific, tricky question: "What does this system look like if we rewind time all the way to the very beginning?"

The Problem: The "Infinite" Start

Usually, we start a simulation with a few cars at specific spots and watch them crash. But what if we want to describe the system as if it always existed, or started with a car at every single point on the highway?

If you try to put a car at every single point on an infinite line, you get a mathematical nightmare. The cars would crash instantly, and the system would change its nature immediately. The authors are trying to figure out: What are all the possible "starting states" (entrance laws) that make sense for this chaotic system?

The Solution: The "Pfaffian" Recipe

The authors discovered that no matter how you start this system (as long as it's a valid "extreme" starting point), the pattern of where the cars end up at any future time follows a very specific, elegant mathematical recipe called a Pfaffian point process.

Think of a Pfaffian not as a scary math word, but as a special "fingerprint" or "blueprint."

• If you know the blueprint, you can predict the probability of finding cars at any set of locations.
• The authors found that for every possible valid "starting scenario," there is a unique blueprint.

The Two Types of Starters

The paper classifies these starting scenarios into two main "flavors" based on the probability of the cars merging vs. disappearing:

1. The "Pure Mergers" (Coalescing only):
Imagine a scenario where cars always merge when they hit. The authors found that the "extreme" starting points for this are like having a "ghost" car at every location, but the pattern is determined by a simple function that assigns a value (like +1 or -1) to every spot on the road. It's like a checkerboard pattern of potential cars.

2. The "Mixed Bag" (Merging and Disappearing):
This is the more complex case where cars sometimes vanish. Here, the "extreme" starting points are determined by closed sets (think of them as specific "safe zones" or "forbidden zones" on the highway).

• If a gap between two points contains no forbidden zones, the cars behave one way.
• If a gap does contain a forbidden zone, the behavior changes.
• The "extreme" laws correspond to the simplest possible arrangements of these zones (like a single solid block of "no-go" area).

The "Dust Settling" Analogy

The authors explain a fascinating intuition about why some starting points aren't "extreme" (meaning they aren't unique, but just a mix of others).

Imagine you start with two cars sitting exactly on top of each other at time zero.

• They crash instantly.
• They might both vanish (0 cars left).
• They might merge into one (1 car left).
• The math shows that this "double car" start is just a mixture of a "zero car" start and a "one car" start. It's not a unique, fundamental state; it's just a 50/50 gamble between two simpler states.

The paper proves that only the simplest, most "pure" starting configurations (like the ones described by the specific functions and sets mentioned above) are the true "extreme" building blocks. Any other weird starting condition is just a cocktail of these pure ones.

Why Does This Matter?

You might ask, "Who cares about invisible cars crashing on a math highway?"

This isn't just about cars. These models describe:

• Polymer chains in chemistry (how long molecules tangle and break).
• Voter models in sociology (how opinions spread and merge or cancel out in a crowd).
• Fluid dynamics (how interfaces between different fluids move).

By finding the "extreme entrance laws," the authors have provided a complete dictionary for describing the initial conditions of these complex systems. They showed that despite the chaos of random motion and instant reactions, the underlying structure is surprisingly orderly, governed by these beautiful "Pfaffian" blueprints.

In a nutshell: The paper takes a chaotic system of random particles that crash and merge, and proves that every possible way to "start" this system is just a combination of a few very specific, mathematically beautiful patterns. It turns a messy fog of particles into a clear, predictable map.

Technical Summary

1. Problem Statement

The paper investigates systems of particles moving on the real line $\mathbb{R}$ as independent Brownian motions between collisions. Upon collision, pairs of particles react instantaneously:

• Annihilation: They disappear with probability $\theta$.
• Coalescence: They merge into a single particle with probability $1-\theta$.

The parameter $\theta \in [0, 1]$ interpolates between the purely coalescing case ($\theta=0$) and the purely annihilating case ($\theta=1$). The state space is $M_0$, the space of simple point measures (at most one particle per point) on $\mathbb{R}$.

The core mathematical problem is the classification of entrance laws for this Markov process. An entrance law is a family of probability measures $(Q_t)_{t>0}$ on $M_0$ satisfying the semigroup property:
$$\int_{M_0} T_t F(\mu) Q_s(d\mu) = \int_{M_0} F(\mu) Q_{t+s}(d\mu)$$
where $T_t$ is the transition semigroup. While specific entrance laws (like the "Arratia flow" starting with a particle at every point) are known, a complete characterization of the extreme points of the convex set of all possible entrance laws was lacking, particularly for the mixed case $\theta \in (0, 1)$.

2. Methodology

The authors employ a combination of Markov duality, Pfaffian point process theory, and functional analysis (specifically weak-$\ast$ topology in $L^\infty$).

• Pfaffian Structure: The paper relies on the established fact that for any deterministic initial condition, the particle system at time $t$ forms a Pfaffian point process. The $n$-point intensities $\rho_t(x_1, \dots, x_n)$ are given by the Pfaffian of a kernel matrix $K_t$:
  $$\rho_t(x_1, \dots, x_n) = \text{pf}(K_t(x_i, x_j))$$
  The kernel $K_t$ is derived from a scalar kernel $K_t(x,y)$ which solves a heat equation with specific boundary conditions determined by $\theta$.

• Duality Function: The analysis utilizes a duality function $s_\mu(x,y) = (-\theta)^{\mu(x,y)}$, where $\mu(x,y)$ is the number of particles in the interval $(x,y)$. This function links the particle system to a dual process, allowing the authors to characterize the laws of the system via the behavior of these spin functions.

• Weak-$\ast$ Closure Analysis: The authors define a set of "finite spin functions" $S_\theta$ corresponding to finite initial particle configurations. They analyze the weak-$\ast$ closure of this set in $L^\infty(V_2)$ (where $V_2 = \{(x,y) : x < y\}$). They prove that the extreme entrance laws correspond to the extreme points of this closure.

• Choquet-type Representation: By showing that any entrance law can be represented as a mixture of laws generated by functions in the closure of $S_\theta$, and then identifying the extreme points of that closure, they classify the extreme entrance laws.

3. Key Contributions and Results

A. Identification of Extreme Entrance Laws (Theorem 1)

The paper provides a complete classification of the extreme points of the set of entrance laws. These are parameterized by a specific set of functions $C_\theta \subseteq L^\infty(V_2)$:

1. Purely Coalescing Case ($\theta = 0$):
   The extreme entrance laws are indexed by functions of the form $f(x)f(y)$, where $f: \mathbb{R} \to [-1, 1]$ is a measurable function.
   $$C_0 = \{ f(x)f(y) \mid f \text{ measurable}, f: \mathbb{R} \to [-1, 1] \}$$
   *(Note: The paper notation uses $\theta=1$ for coalescing in the definition of $C_1$, but standard literature often swaps these. Based on the text provided: $C_1$ corresponds to the coalescing case $\theta=0$ in the abstract's context of "purely coalescing", but the text defines $C_1$ for the case where the duality is $(-1)^{\mu}$. Let us strictly follow the paper's definitions: The paper defines $C_1$ for the case where the duality is $(-1)^{\mu}$, which corresponds to $\theta=0$ (coalescing) in the standard Arratia flow context, but the text says "For the coalescing case... $\theta=0$". Wait, the text says: "For the coalescing case... $\theta=0$". Then it defines $C_1 = \{f(x)f(y)\}$. This implies the paper uses $\theta=1$ for coalescing? Let's re-read carefully.
   Correction based on text: The text states: "interpolates between the purely coalescing case and the purely annihilating case... $\theta \in [0,1]$".
   Then it says: "For the coalescing case... $\theta=0$".
   Then Theorem 1 says: $C_1 = \{f(x)f(y)\}$.
   Then it says: "for $\theta \in [0, 1)$... $C_\theta = \{I(S \cap (x,y) = \emptyset)\}$.
   Crucial Interpretation: The paper defines $C_1$ for the case where the duality function is $(-1)^{\mu}$, which occurs when $\theta=0$ (coalescing) because $(-0)^k$ is 0 unless $k=0$, but the text says "when $\theta=0$ the expression becomes the indicator".
   Actually, looking at Eq (7): $s_\mu(x,y) = (-\theta)^{\mu(x,y)}$.
   If $\theta=0$ (coalescing), $s_\mu = 0^{\mu} = I(\mu=0)$.
   If $\theta=1$ (annihilating), $s_\mu = (-1)^\mu$.
   Theorem 1 defines $C_1$ for the case where the duality is $f(x)f(y)$. This corresponds to the annihilating case ($\theta=1$) where the spin function factorizes.
   Theorem 1 defines $C_\theta$ for $\theta \in [0, 1)$ as indicators of closed sets. This corresponds to the coalescing case ($\theta=0$) and mixed cases.
   Let's re-verify the text's specific mapping:
   Text: "For the coalescing case... $\theta=0$".
   Text: "For the mixed case $\theta \in (0,1)$".
   Text: "Theorem 1... $C_1 = \{f(x)f(y)\}$".
   Text: "for $\theta \in [0, 1)$... $C_\theta = \{I(S \cap (x,y) = \emptyset)\}$".
   Conclusion: The paper's notation for the sets $C_\theta$ in Theorem 1 seems to map the index to the parameter $\theta$ in the duality function $(-\theta)$.

   • If $\theta=1$ (pure annihilation), the extreme laws are $f(x)f(y)$.
   • If $\theta \in [0, 1)$ (including pure coalescence $\theta=0$), the extreme laws are indicators of closed sets $I(S \cap (x,y) = \emptyset)$.
     Wait, the text says: "The extreme elements... are $(Q_f : f \in C_\theta)$ where... $C_1 = \{f(x)f(y)\}$... and for $\theta \in [0, 1)$... $C_\theta = \{I(S \cap (x,y) = \emptyset)\}$."
     This implies:
   • Pure Annihilation ($\theta=1$): Extreme laws are generated by product functions $f(x)f(y)$.
   • Coalescing/Mixed ($\theta < 1$): Extreme laws are generated by indicators of closed sets (representing "gaps" where no particles exist).

2. Mixed and Coalescing Cases ($\theta \in [0, 1)$):
   The extreme entrance laws are indexed by closed sets $S \subseteq \mathbb{R}$:
   $$C_\theta = \{ I(S \cap (x, y) = \emptyset) \mid S \subseteq \mathbb{R} \text{ is closed} \}$$
   These correspond to initial configurations where particles are distributed such that the "holes" (intervals with no particles) are determined by the complement of a closed set $S$.

B. Mixture Representation

The authors prove that any entrance law $Q_t$ can be represented as a mixture (integral) of these extreme laws:
$$Q_t(d\nu) = \int_{\tilde{C}_\theta} Q^f_t(d\nu) \Theta(df)$$
where $\Theta$ is a probability measure on the space of functions $\tilde{C}_\theta$ (the weak-$\ast$ closure of the spin functions).

C. Non-Extremality of "Heavy" Initial Conditions

The paper demonstrates that entrance laws corresponding to initial conditions with "weights" (multiple particles at a single point) are not extreme. For example, if $k$ particles start at the same point, the resulting law is a convex combination of laws starting with 0 or 1 particle at that point. This is verified using the Pfaffian structure and the specific identity:
$$(-\theta)^k = p_k + (1-p_k)(-\theta)$$
where $p_k$ is the probability that $k$ colliding particles result in 0 survivors.

D. Proof of Lemma 2 (Closure of Spin Functions)

A significant technical contribution is the proof that the weak-$\ast$ closure of the set of spin functions for finite particle systems is exactly the set $\tilde{C}_\theta$.

• For $\theta=1$, the closure consists of products $f(x)f(y)$.
• For $\theta < 1$, the closure consists of functions defined by closed sets $S$ and weights $w$ on isolated points, where the function is $I(S^c \cap (x,y) = \emptyset)(-\theta)^{\sum w(z)}$.

4. Significance

• Complete Classification: This work provides the first complete description of the set of entrance laws for coalescing-annihilating Brownian motions for all $\theta \in [0, 1]$. Previous work (e.g., by Arratia or Durrett) focused on specific cases or used duality without fully characterizing the extreme points of the entrance law space.
• Unification of Cases: It unifies the treatment of purely coalescing, purely annihilating, and mixed systems under a single framework of Pfaffian point processes and duality.
• Structural Insight: The result highlights a fundamental difference between the coalescing/mixed regime ($\theta < 1$) and the purely annihilating regime ($\theta = 1$). In the former, the extreme laws are determined by the geometry of "empty intervals" (closed sets), while in the latter, they are determined by independent spin fields (product functions).
• Methodological Advancement: The use of Fredholm Pfaffians and the explicit construction of the weak-$\ast$ closure of spin functions offers a robust toolkit for analyzing other interacting particle systems with instantaneous reactions.
• Applications: The findings are relevant to models in polymer physics, voter models, and coalescent theory, where understanding the "entrance" behavior (how a system emerges from a singular or infinite initial state) is crucial for defining the process rigorously.

In summary, Tribe and Zaboronski resolve the classification of entrance laws by linking the probabilistic behavior of the particle system to the analytic properties of Pfaffian kernels and the topological structure of duality functions, revealing that the extreme laws are determined by either closed sets (for $\theta < 1$) or product measures (for $\theta = 1$).