VRJP recurrence and fractional-moment decay for the $H^{2|2}$ model's effective field on the hierarchical lattice

This paper proves that the vertex-reinforced jump process on the hierarchical lattice is recurrent for spectral dimensions $d < 2$ by establishing fractional-moment decay for the effective field of the associated $H^{2|2}$ model, thereby identifying the recurrent phase in the model's phase diagram while leaving the weak-reinforcement critical regime as the remaining open problem.

The Gist

Imagine a vast, infinite city where every single house is connected to every other house by a road. This is the "Hierarchical Lattice" in our story. It's not a normal city; it has a special, nested structure. Think of it like a set of Russian nesting dolls: small groups of houses are inside larger groups, which are inside even larger groups, all the way up to the whole city.

In this city, there is a traveler called the VRJP (Vertex-Reinforced Jump Process). This traveler has a very specific personality: they love familiarity.

The Traveler's Rule

Every time the traveler jumps from House A to House B, they leave a "stamp" on House B. The more stamps a house has, the more likely the traveler is to visit it again.

• Strong Reinforcement: If the traveler is very "sticky" (strong reinforcement), they get addicted to the houses they've already visited. They keep circling back to the same few spots.
• Weak Reinforcement: If they are less sticky, they wander more freely, behaving more like a random tourist who just picks a direction without caring about the past.

The big question the authors asked is: Will this traveler get stuck in a loop, visiting the same houses forever (Recurrent), or will they eventually wander off to the edge of the city and never come back (Transient)?

The Shape of the City Matters

The city isn't just a random mess; it has a specific "shape" or dimension, defined by how the houses are grouped. The authors found that the answer depends entirely on this shape:

1. The "Flat" City (Dimension < 2): If the city is "flat" enough, the traveler always gets stuck in a loop. No matter how they start, the "familiarity" rule eventually traps them. They will visit every house infinitely many times.
2. The "Spiky" City (Dimension > 2): If the city is "spiky" or high-dimensional, the traveler can escape. Even with the familiarity rule, the sheer size and structure of the city allow them to wander off and never return.
3. The "Critical" City (Dimension = 2): This is the tricky middle ground. Here, the outcome depends on how "sticky" the traveler is.
   • If they are very sticky (strong reinforcement), they get trapped and stay forever.
   • If they are not sticky enough, they might escape (though the paper didn't solve this specific weak-sticky case).

The Secret Weapon: The "Effective Field"

To prove this, the authors didn't just watch the traveler. They looked at the "weather" of the city, which they call the Effective Field.

Imagine the city has a magical force field that changes based on where the traveler has been.

• If the traveler is likely to get stuck, this force field creates a "valley" that pulls them back in.
• If the traveler is likely to escape, the force field creates a "slope" that pushes them away.

The authors proved that in the "stuck" scenarios (the recurrent ones), this force field decays geometrically.

• Analogy: Imagine shouting in a canyon. If the canyon is shaped right (the recurrent case), your echo fades away very quickly as you move further from the source. The "memory" of the shout doesn't travel far.
• The authors showed that this "echo" (the mathematical value of the field) gets weaker and weaker the further you go from the starting point, following a strict, predictable pattern.

How They Solved It: The "Zoom-Out" Trick

The math behind this is usually incredibly hard because the traveler can jump from any house to any other house instantly. Counting all the possible paths is like trying to count every grain of sand on a beach.

The authors used a clever trick called Coarse-Graining:

1. The Zoom-Out: Instead of looking at individual houses, they started grouping them into blocks (like looking at a map where neighborhoods are just single dots).
2. The Exact Identity: They discovered a special mathematical rule that says: "If you zoom out and treat a whole neighborhood as a single dot, the rules of the game stay exactly the same."
3. The Recursion: By zooming out step-by-step (from houses to blocks, to super-blocks, to the whole city), they turned a messy, infinite problem into a simple, repeating pattern. They could then calculate exactly how the "echo" (the field) fades as you move up the scales.

The Bottom Line

This paper is like a map for a very specific type of traveler in a very specific type of city.

• If the city is small/flat: The traveler is doomed to wander forever in circles.
• If the city is huge/spiky: The traveler can escape.
• If the city is just right: The traveler is trapped only if they are very clingy.

The authors proved this by showing that the "memory" of the traveler's path fades away rapidly in the trapped scenarios, using a method that simplifies the complex web of connections into a neat, step-by-step ladder. They have successfully identified the "safe zone" where the traveler never leaves, leaving only one small, difficult scenario (the weakly clingy traveler in the critical city) for future explorers to solve.

Technical Summary

Technical Summary: VRJP Recurrence and Fractional-Moment Decay for the H2|2 Model on the Hierarchical Lattice

Problem Statement
This paper investigates the recurrence and transience properties of the Vertex-Reinforced Jump Process (VRJP) on the hierarchical lattice, a structure defined as a weighted infinite complete graph with edge weights decaying according to a hierarchical scale. The study focuses on the associated effective $H^{2|2}$ field (the horospherical $u$-field arising from the supersymmetric representation of the model). The central question is to determine the phase diagram of the VRJP with respect to the spectral dimension $d = 2 \log 2 / \log \rho$ and the conductance parameter $W$. Specifically, the authors aim to resolve the behavior in the recurrent regime ($d < 2$) and the critical regime ($d = 2$), complementing existing results on long-range order in the transient regime ($d > 2$).

Methodology
The proof strategy combines the fractional-moment method, originally developed for Anderson localization, with an exact hierarchical coarse-graining identity specific to the $H^{2|2}$ model.

1. Supersymmetric Representation: The VRJP is analyzed via its connection to a random Schrödinger operator $H_\beta = 2\beta - P_W$, where $\beta$ is a random potential. The effective field $u_i$ is defined as the ratio of Green's function entries, $u_i = G(i_0, i)/G(i_0, i_0)$, which serves as the random environment for the time-changed VRJP.
2. Fractional-Moment Estimates: The core technical effort involves establishing bounds on the fractional moments of the Green's function, specifically $E(e^{s u_i})$ for $0 < s < 1/2$. The authors utilize the property that the inverse diagonal Green's function follows a Gamma distribution to decouple local variables.
3. Exact Coarse-Graining: A key innovation is the application of an exact coarse-graining identity (Lemma 7, Corollary 8). This identity allows a set of vertices that appear identical from the outside to be merged into a single effective vertex without altering the distribution of the effective field. This transforms the path expansion of the Green's function into a recursion over block scales.
4. Recursive Control: By iterating this coarse-graining, the authors control the combinatorial explosion of paths on the complete hierarchical graph. They derive recursive inequalities (Proposition 9) that bound the fractional moment at a given scale by a sum of moments at coarser scales.

Key Contributions and Results

• Recurrence for $d < 2$: The authors prove that for any conductance parameter $W > 0$, the VRJP is recurrent when the spectral dimension $d < 2$. This is established by showing that the effective conductance between a vertex and the boundary grows sufficiently fast, preventing the process from escaping to infinity.
• Recurrence at Criticality ($d = 2$): For the critical spectral dimension $d = 2$, the paper proves recurrence in the strong-reinforcement regime (sufficiently small $W$). This identifies a recurrent phase at the critical point, contrasting with the weak-reinforcement regime which remains open.
• Geometric Decay of Fractional Moments: In the recurrent regimes ($d < 2$ and $d=2$ with small $W$), the paper establishes geometric decay of the fractional moments of the normalized Green's function across hierarchical scales.
  • For $d < 2$, the decay rate is optimal, matching the natural lower bound provided by the direct edge weight: $E(e^{s u_i}) \leq C \rho^{-sn}$.
  • For $d = 2$ with small $W$, the decay is geometric with a rate determined by a constant $c(W, s)$ that approaches 2 as $W \to 0$.
• Two-Point Estimates: The methodology is extended to derive bounds for the field pinned at an arbitrary vertex $j$, $E(e^{s u_i^{[j]}})$, showing decay dependent on the hierarchical distance $d_X(i, j)$.

Significance and Claims
The paper claims to identify the recurrent side of the hierarchical VRJP phase diagram. By proving geometric decay of fractional moments, the authors provide rigorous evidence that $d=2$ acts as the critical spectral dimension on the hierarchical lattice. The results support the interpretation that the phase behavior away from the critical point is governed by the spectral dimension, while the critical case requires a more delicate analysis of the reinforcement strength.

The work complements previous findings on long-range order for $d > 2$ (cited as [4, 5, 6]), thereby providing a more complete picture of the phase transition. The authors explicitly state that the weak-reinforcement regime at the critical dimension $d=2$ remains an open problem. The proof technique, specifically the combination of fractional-moment bounds with exact hierarchical coarse-graining, is presented as a robust tool for handling the combinatorial growth of long-range edges in hierarchical models.