Shocks and instability in Brownian last-passage percolation

This paper analyzes the structural relationships between shocks, instability, and competition interfaces in Brownian last-passage percolation, demonstrating how the divergence and alignment of shock trees for unstable eternal solutions can be used to reconstruct the instability region.

The Gist

Imagine you are standing on a vast, foggy landscape where the ground is constantly shifting and jiggling. This isn't a normal landscape; it's a mathematical world called Brownian Last-Passage Percolation (BLPP).

In this world, you want to travel from point A to point B. But there's a catch: the "terrain" (the ground) is made of random, wiggly lines (like Brownian motion). To get from A to B, you want to take the path that collects the most "energy" or "points" along the way.

In this paper, the authors, Firas Rassoul-Agha and Mikhail Sweeney, act like cartographers trying to map the hidden rules of this chaotic world. They are looking for three specific things: Shocks, Instability, and Competition Interfaces.

Here is a simple breakdown of what they found, using everyday analogies.

1. The Setting: The Great Race

Imagine thousands of runners starting at different times and places, all trying to run from the past (time $-\infty$) to the future. They all want to run the "best" path (the one with the most points).

• The Path: These paths are called geodesics. Think of them as the most efficient routes a runner can take.
• The Goal: The authors are studying what happens when these runners look back at the infinite past. Do they all agree on the best route? Or do they disagree?

2. The Three Key Concepts

A. Shocks: The Traffic Jams

In a normal, smooth world, if two runners start close together, they might merge onto the same path eventually. But in this jiggly world, sometimes two runners start at the exact same spot and immediately split apart, taking two completely different routes.

• The Analogy: Imagine a highway where, at a specific mile marker, the road suddenly splits into two distinct lanes that never merge back together. This split point is a Shock.
• What the paper says: These shocks aren't random chaos; they form a beautiful, tree-like structure. If you look at all the shocks, they look like a forest where branches grow downward (into the past) and merge together, but they never split upward. It's a "coalescing tree."

B. Instability: The Fork in the Road

Now, imagine a specific speed (or direction) that the runners want to maintain. Usually, if you tell a runner to go at "Speed X," they will all eventually agree on the same path.

• The Problem: Sometimes, for certain "special" speeds, the runners disagree. Two runners starting at the same spot, both trying to go at "Speed X," will take two different paths that never meet again.
• The Analogy: This is Instability. It's like a fork in the road where the sign says "Go North," but half the people go North-East and the other half go North-West, and they never reunite.
• The Map: The authors mapped out exactly where these disagreements happen. They call this the Instability Graph. It looks like a web or a net of vertical lines and horizontal segments.

C. Competition Interfaces: The Border Patrol

When two runners start at the same place and take different paths, there is a "border" between the territory they control.

• The Analogy: Imagine two kingdoms expanding from a single castle. The Competition Interface is the border line that separates the land conquered by the "Left Runner" from the land conquered by the "Right Runner."
• The Discovery: The authors found that these borders are the key to understanding the whole map. If you know where these borders start, you can figure out exactly where the Instability and the Shocks are.

3. The Big Connection: Reconstructing the Map

The most exciting part of the paper is how these three things connect.

• The Shock Trees: The authors showed that the "Shocks" (the traffic jams) form a tree structure.
• The Instability Web: The "Instability" (the disagreements) forms a web.
• The Magic Link: You can actually reconstruct the Instability Web just by looking at the Shock Trees.

Here is the metaphor:
Imagine the Instability Web is a giant, invisible spiderweb stretched across the sky. You can't see the web itself. However, the "Shocks" are like birds that only land on the specific threads of the web.

• If you watch where the birds land (the shocks), you can draw the outline of the web (the instability).
• The authors proved that if you know the locations of these "birds" (shocks), you can draw a "skeleton" of the entire web. You might miss a few tiny, invisible threads, but you can see the whole structure.

4. Why Does This Matter?

This isn't just about math games. This model (BLPP) is a simplified version of real-world physics problems, like:

• Traffic flow: How jams form and move.
• Fluid dynamics: How fluids move through porous rock or how turbulence behaves.
• Growth: How crystals grow or how a forest fire spreads.

The "Hamilton-Jacobi equations" mentioned in the paper are the master equations that describe how these systems evolve. The authors found that instability (where the system is unpredictable) and shocks (where the system breaks or jumps) are deeply linked.

Summary in One Sentence

The authors discovered that in a chaotic, jiggly world, the places where paths split forever (instability) are perfectly mapped out by the places where traffic jams occur (shocks), and you can draw the entire map of the chaos just by tracking where those jams happen.

They turned a messy, unpredictable mathematical problem into a structured, tree-like map that anyone (well, any mathematician!) can follow.

Technical Summary

1. Problem Statement and Context

The paper investigates the structural relationship between shocks (discontinuities in the velocity field) and instability (the existence of multiple eternal solutions with the same asymptotic velocity) within the framework of Brownian Last-Passage Percolation (BLPP).

• The Model: BLPP is a semi-discrete growth model defined on the lattice $\mathbb{Z} \times \mathbb{R}$. It serves as a prototype for inviscid Stochastic Hamilton-Jacobi (SHJ) equations in one space dimension. In this analogy, the passage times correspond to the value function of the SHJ equation, and geodesics correspond to characteristic lines.
• The Core Phenomena:
  • Instability: In stochastic Hamilton-Jacobi equations, the "One Force-One Solution" (1F1S) principle generally holds, meaning solutions starting from different initial conditions synchronize over time. However, for a random, countable, dense set of exceptional velocities $\Theta_\omega$, this principle fails. At these velocities, multiple "eternal solutions" (defined for all time) exist that do not synchronize. The set of space-time points where these solutions differ constitutes the instability region.
  • Shocks: In the inviscid limit, the velocity field becomes discontinuous. These discontinuities are shocks. In the BLPP context, shocks are points from which multiple semi-infinite geodesics (characteristics) emanate immediately and separate.
• The Gap: While the 1F1S principle and instability have been studied in discrete models (like lattice LPP) and continuous models (like the KPZ equation), the specific interplay between the geometry of shocks and the topology of the instability region in a model with continuous space (BLPP) had not been rigorously analyzed. The authors aim to characterize the instability graph and reconstruct it using the locations of shock points and competition interfaces.

2. Methodology

The authors employ a combination of probabilistic tools, geometric analysis of geodesics, and the theory of Busemann functions.

• Busemann Process: The analysis relies heavily on the Busemann process $B^\theta_{\square}(m, s, n, t)$, defined as the limit of differences in last-passage times to infinity in direction $\theta$. The distinction between left ($-$) and right ($+$) limits of the Busemann function as the direction parameter approaches a value is crucial.
  • Instability occurs when $B^{\theta-} \neq B^{\theta+}$.
  • The authors define the instability graph $S_\theta$ based on the points where these limits differ or where the associated functions have points of increase.
• Geodesic Characterization: The paper translates analytical definitions (based on Busemann functions) into geometric properties of semi-infinite geodesics (paths maximizing the weight).
  • Shock Points: Defined as points $(m, s)$ where the leftmost ($\Gamma^{L, \theta}_{(m,s)}$) and rightmost ($\Gamma^{R, \theta}_{(m,s)}$) geodesics split immediately.
  • Competition Interfaces: Paths on the dual lattice separating regions dominated by different geodesic behaviors.
• Key Technical Tools:
  • Monotonicity and Coalescence: Utilizing the fact that geodesics with the same direction coalesce almost surely, except in the instability regime.
  • Non-existence of Bi-infinite Geodesics: Proving that no non-degenerate bi-infinite geodesics exist that are locally extremal (Theorem 5.4), which is essential for establishing the tree structure of shocks.
  • Brownian Scaling: Using the scaling properties of Brownian motion to bound the number of disjoint geodesics (Theorem 11.3), adapting results from the Directed Landscape.

3. Key Contributions and Results

The paper provides a comprehensive structural description of the instability graph and its relationship to shocks.

A. Structure of the Instability Graph ($S_\theta$)

For an exceptional velocity $\theta \in \Theta_\omega$, the instability graph $S_\theta$ is a connected, closed, bi-infinite set in the dual lattice $(\mathbb{Z} + 1/2) \times \mathbb{R}$.

• Topology: It consists of horizontal instability intervals (where the instability is "proper") connected by vertical instability edges.
• Dimension: The set of points from which vertical edges descend has a Hausdorff dimension of 1/2.
• Web Structure: The graph exhibits a "web-like" structure: any two instability points share a common "NE ancestor" (a point reachable by an up-right path on the graph) and a common "SW descendant" (reachable by a down-left path).
• No Double Edges: There are no points in the graph from which vertical edges emanate both upwards and downwards.

B. Structure of Shock Trees

The set of shock points $NU^\theta_1$ forms a tree structure (specifically a forest of trees) on the primal lattice $\mathbb{Z} \times \mathbb{R}$.

• Ancestry: Shock points are partially ordered by ancestry. Branches coalesce in the South-West (time decreasing) direction and die out in the North-East direction.
• Subtrees: The shock points are partitioned based on the sign of the Busemann limit ($\theta-$ vs $\theta+$):
  • $NU^{\theta+}_1 \setminus NU^{\theta-}_1$ forms a subtree.
  • $NU^{\theta-}_1 \setminus NU^{\theta+}_1$ forms a subtree.
  • The intersection $NU^{\theta+}_1 \cap NU^{\theta-}_1$ forms "bushes" (finite-depth subtrees) attached to the main trees.
• Cardinality: There are significantly more $\theta-$ shocks than $\theta+$ shocks. Specifically, the number of shock points increases as the velocity parameter decreases.

C. Reconstruction of the Instability Graph

A major contribution is the demonstration that the instability graph can be reconstructed from the locations of shock points and competition interfaces.

• Shock-to-Instability Mapping:
  • Left Endpoints: A point $(m, s)$ is the left endpoint of a maximal instability interval if and only if it is a $\theta+$ shock but not a $\theta-$ shock.
  • Right Endpoints: A point $(m, t)$ is the right endpoint if it is a $\theta-$ shock but not a $\theta+$ shock (located above the interval).
  • Interior: Points inside an instability interval correspond to $\theta-$ shocks that are not $\theta+$ shocks.
• Competition Interfaces: The starting points of left and right competition interfaces with asymptotic direction $\theta$ exactly identify the vertical instability edges.
  • If a point generates only a right interface, it is a $\theta+$ shock (left endpoint).
  • If it generates only a left interface, it is a $\theta-$ shock (interior/right endpoint).
  • If it generates both (and they match), it is not a shock.

D. Quantitative Results

• Density: Shock points are countable and nowhere dense on any fixed level, whereas the set of instability edges has Hausdorff dimension 1/2.
• Reconstruction Limitation: While the "skeleton" of the instability graph (the horizontal intervals and major vertical edges) can be reconstructed from shock points, the full set of vertical edges (the dimension 1/2 set) cannot be fully recovered from shocks alone because there are only countably many shocks. However, the competition interfaces provide the missing information.

4. Significance and Impact

• Bridging Discrete and Continuous: The paper successfully extends the understanding of instability and shocks from discrete lattice models (where shocks are ill-defined) to the semi-discrete Brownian setting, providing a rigorous prototype for inviscid SHJ equations.
• Geometric Insight: It reveals that the "instability" of the random dynamical system is not a diffuse cloud but a highly structured, fractal-like web (dimension 1/2) that is intimately tied to the tree-like structure of shocks.
• Universality: The methods developed are expected to transfer to other models in the KPZ universality class, including the KPZ equation and the KPZ fixed point. The authors explicitly mention applying these techniques to the KPZ fixed point in forthcoming work.
• New Framework: The paper establishes a new framework for analyzing the interplay between the "pullback attractors" (stability) and "shocks" (instability) in stochastic PDEs, moving beyond the 1F1S principle to characterize the geometry of the failure of synchronization.

In summary, Rassoul-Agha and Sweeney provide a rigorous geometric characterization of how instability manifests in Brownian last-passage percolation, proving that the complex, fractal structure of the instability region is encoded in the simpler, tree-like structure of shock points and competition interfaces.