🔢 mathematics

Parabolic Anderson Model in the Hyperbolic Space. Part II: Quenched Asymptotics

This paper establishes the exact quenched asymptotic growth of the solution to the parabolic Anderson model in hyperbolic space with a Gaussian potential, demonstrating that it follows a distinct $e^{L^* t^{5/3}}$ scaling law driven by a unique non-Euclidean localization mechanism.

Original authors: Xi Geng, Sheng Wang, Weijun Xu

Published 2026-02-03

📖 5 min read🧠 Deep dive

CC0 1.0

Original authors: Xi Geng, Sheng Wang, Weijun Xu

Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are watching a very strange, invisible cloud of "energy" (called a Gaussian field) spread out over a vast, curved landscape. This landscape isn't flat like a sheet of paper (Euclidean space); it's shaped like a hyperbolic saddle or a Pringles chip that curves away from you in every direction. This is Hyperbolic Space.

Now, imagine a tiny, invisible particle (a Brownian motion) wandering randomly across this landscape. At the same time, there is a "Parabolic Anderson Model" (PAM) happening. You can think of the PAM as a game where the particle tries to grow as large as possible. The particle grows faster if it stands in areas where the energy cloud is very high (peaks) and grows slower (or shrinks) if it stands in low areas (valleys).

The big question the authors asked is: How fast does this particle grow over a very long time, and where does it go to do it?

The Two Different Worlds: Flat vs. Curved

To understand why this paper is special, you have to compare two worlds:

The Flat World (Euclidean): In our normal, flat world, the energy cloud has peaks, but they are relatively rare and hard to find. The particle has to wander a long way to find a good spot. The authors note that in this flat world, the particle's growth is relatively slow. It's like finding a treasure chest in a huge, flat desert; you have to walk a long time, and the chest isn't that much bigger than the sand around it.
The Curved World (Hyperbolic): In this hyperbolic world, the landscape expands exponentially. It's like a fractal tree that keeps branching out faster and faster. Because the space is so huge, there are massive numbers of energy peaks. It's like being in a forest where every tree is a treasure chest, but they are scattered in a way that gets denser the further you go.

The Big Discovery: A New Strategy

The authors discovered that in this curved world, the particle doesn't just wander aimlessly. It adopts a very specific, highly optimized strategy to maximize its growth. They call this the "Quenched Asymptotics" (which is a fancy way of saying: "What happens if we look at one specific, fixed map of the energy cloud and see how the particle behaves on it?").

Here is the particle's winning strategy, broken down simply:

The Sprint: The particle doesn't stay put. It realizes that the best energy peaks are far away, near the "edge" of its reachable world. So, it decides to sprint toward a specific distant peak.
The Timing: It doesn't sprint the whole time. The math shows the optimal plan is to sprint for exactly 20% of the total time (1/5th).
The Destination: It aims for a peak located at a specific distance that grows with time (specifically, proportional to time to the power of 4/3).
The Stay: Once it reaches this peak at the 20% mark, it stops moving. It stays right there, soaking up the maximum energy for the remaining 80% of the time.

The Result: A Super-Fast Growth Spurt

Because of this specific strategy, the particle grows incredibly fast.

In the flat world, the growth is slow and follows a pattern involving square roots of logarithms (very slow).
In this curved world, the growth follows a $t^{5/3}$ pattern.

Think of it like this: If you were saving money, the flat world might give you interest that grows slowly. The hyperbolic world, with this specific strategy, gives you interest that explodes exponentially faster. The authors calculated the exact "speed limit" of this explosion (a constant called $L^*$ ), which depends on the shape of the landscape and the strength of the energy cloud.

Why is this different from the flat world?

In the flat world, the particle can reach a good spot almost instantly, so the "cost" of traveling there is negligible. It just finds a spot and stays.

In the hyperbolic world, the landscape is so vast that traveling is expensive. The particle has to spend a significant amount of time (20%) just getting to the good spot. The final growth rate is a "compromise" or a balance between:

The Reward: How much energy it gets from staying at the peak.
The Cost: How much energy it "loses" (or how much probability it burns) just by traveling there.

The authors proved that the $t^{5/3}$ growth rate is the perfect balance point where the reward of the peak exactly offsets the cost of the journey.

The "Localization" Secret

The paper also reveals a fascinating phenomenon called localization. In the flat world, the particle might spread out a bit. But in the hyperbolic world, the particle is extremely picky. It essentially says, "I will ignore 99.9% of the universe. I will go to this one specific spot near the edge, and I will never leave."

The authors used advanced math to prove that no other strategy (like visiting multiple peaks or wandering around) could beat this specific "Sprint-then-Stay" plan. They showed that even though there are billions of peaks in this curved space, the particle only cares about the one that is perfectly reachable within that 20% time window.

Summary

In short, this paper solves a puzzle about how a random walker grows in a curved, expanding universe. They found that the walker doesn't wander randomly; it executes a precise, calculated plan: Run fast to a specific distant peak, arrive at exactly 20% of the time, and stay there forever. This strategy allows it to grow at a rate of $t^{5/3}$ , which is much faster than anything possible in our flat, everyday world.

Technical Summary: Parabolic Anderson Model in the Hyperbolic Space (Part II: Quenched Asymptotics)

Problem Statement
This article investigates the long-time quenched asymptotic behavior of the solution $u(t, x)$ to the Parabolic Anderson Model (PAM) on the standard hyperbolic space $\mathbb{H}^d$ (with constant curvature $\kappa = -1$ ). The equation is given by:
$\partial_t u = \Delta u + \xi \cdot u, \quad u(0, \cdot) \equiv 1$
where $\xi$ is a real-valued, centered, time-independent, stationary Gaussian field with a regular covariance function $C(d(x,y))$ that vanishes beyond a finite correlation length $r_0$ . The study focuses on the almost sure growth rate of the solution, specifically determining the limit:
$\lim_{t \to \infty} \frac{1}{t^{5/3}} \log u(t, x)$
This work serves as the second part of a series, following [GX25], which established the annealed (moment) asymptotics for the same model.

Main Results
The primary contribution is the establishment of the exact quenched growth rate, which differs fundamentally from the Euclidean case.

Exact Growth Rate: There exists a deterministic constant $L^* > 0$ such that, with probability one, for every $x \in \mathbb{H}^d$ :
$\lim_{t \to \infty} \frac{1}{t^{5/3}} \log u(t, x) = L^*$
The growth rate is of order $t^{5/3}$ , contrasting with the $t \sqrt{\log t}$ rate found in Euclidean settings.
Explicit Constant: The constant $L^*$ is determined by an explicit optimization problem involving the variance $\sigma^2$ of the field and the dimension $d$ :
$L^* = \sup_{\theta \in [0,1], K>0} \left\{ (1-\theta)\sqrt{2(d-1)\sigma^2 K} - \frac{K^2}{4\theta} \right\}$
The optimal parameters are $\theta^* = 1/5$ and $K^* = \frac{2^{5/3}}{5^{4/3}}((d-1)\sigma^2)^{1/3}$ , yielding:
$L^* = \frac{3 \cdot 2^{4/3}}{5^{5/3}} ((d-1)\sigma^2)^{2/3}$

Methodology and Proof Strategy
The proof relies on the Feynman-Kac representation $u(t, o) = \mathbb{E}_o[\exp(\int_0^t \xi(W_s) ds)]$ , where $W$ is a hyperbolic Brownian motion starting at the origin $o$ . The analysis involves a delicate competition between the "reward" (the maximum values of the Gaussian field $\xi$ ) and the "cost" (the probability of the Brownian motion reaching and staying in those regions).

1. Lower Bound Construction (The Optimal Scenario)
The authors identify a specific Brownian trajectory that achieves the growth rate $e^{L^* t^{5/3}}$ . The optimal scenario involves:

Travel Phase: The Brownian motion travels from the origin to a point $z_t$ near the boundary of a ball $B(o, K^* t^{4/3})$ within a time fraction $\theta^* t = t/5$ . The path is essentially a geodesic.
Staying Phase: The motion stays within a small neighborhood of $z_t$ (radius $\sim t^{-2/9}$ ) for the remaining time $(1-\theta^*)t = 4t/5$ .
Mechanism: In this region, the field $\xi$ achieves its maximal growth $\approx \mu_0 \sqrt{R}$ (where $R \sim t^{4/3}$ ). The cost of traveling this distance in time $t/5$ is balanced against the reward of staying in the high-potential region. The $t^{5/3}$ exponent arises from the interplay between the linear time scaling of the reward ( $t \cdot \sqrt{t^{4/3}}$ ) and the quadratic cost of the large deviation for the Brownian motion.

2. Upper Bound Analysis (Ruling Out Other Scenarios)
Proving the upper bound is more challenging due to the exponential volume growth of hyperbolic space, which creates exponentially many potential "peak" regions. The authors employ a multi-step localization strategy:

Localization to a Ball: First, they show the Brownian motion is unlikely to travel beyond a ball of radius $K_0 t^{4/3}$ .
Cluster Decomposition: The high-value regions of $\xi$ are decomposed into "islands" and "clusters." The authors prove that these clusters are locally sparse; specifically, a ball of radius $\eta t^{4/3}$ cannot contain too many well-separated high peaks.
Discrete Trajectories: The continuous Brownian path is mapped to a discrete sequence of clusters visited.
Fine Localization: Using a two-scale clustering argument and large deviation estimates, the authors demonstrate that any "reasonable" trajectory must spend most of its time outside the peak regions. They show that the cost of visiting multiple clusters or deviating from the optimal geodesic path strictly penalizes the growth, ensuring no trajectory can exceed the rate $L^* t^{5/3}$ .

Key Contributions and Distinctions

Non-Euclidean Localization Mechanism: Unlike the Euclidean case, where the Brownian motion reaches the optimal peak almost instantaneously (time fraction $\to 0$ ) and the cost is negligible, the hyperbolic case requires a non-trivial time fraction ( $\theta^* = 1/5$ ) to reach the peak. The cost of travel is comparable to the reward, leading to a unique optimization balance.
Uniqueness of the Optimal Scenario: The optimal scenario in the hyperbolic setting is essentially unique (fixed time fraction and fixed distance scaling), whereas the Euclidean case allows for a flexible range of parameters.
Stronger Localization: The proof reveals a stronger localization mechanism driven by the geometry of the hyperbolic space and the specific growth of the Gaussian field ( $\sqrt{R}$ vs $\sqrt{\log R}$ ).

Significance and Claims
The authors claim that this work establishes the first exact quenched asymptotics for the PAM in a non-Euclidean setting with a regular Gaussian potential. The results highlight that the geometry of the underlying space fundamentally alters the intermittency mechanism. While the annealed (moment) asymptotics in [GX25] were identical to the Euclidean case (determined by the Euclidean Laplacian), the quenched (almost sure) behavior is distinct and governed by the hyperbolic geometry.

The paper notes that the choice of constant negative curvature is for technical convenience, but the exponential volume growth is the essential feature. The methods developed are expected to be robust and extendable to more general geometric settings. The authors explicitly state that the case of time-dependent potentials or more singular Gaussian potentials (like white noise) remains open and would require different techniques.