Extreme Values of Infinite-Measure Processes

Imagine you are running a massive experiment to find the "extremes" of a system. You want to know: What is the longest a person waited in line? What is the highest temperature recorded? What is the fastest a car drove?

In the standard world of statistics (the "textbook" version), if you take enough samples, these extremes usually settle into a predictable pattern. It's like rolling a million dice; eventually, you know exactly how likely it is to roll a 6. This is called Extreme Value Theory, and it usually gives you three standard answers (like the Gumbel or Fréchet distributions).

But this paper says: "Hold on. What if the rules of the game are broken?"

The authors, Talia Baravi and Eli Barkai, are studying systems that are weirdly chaotic. In these systems, the "average" behavior doesn't exist. The time it takes for a particle to return to its starting point is so long that the average is infinite. It's like a game where you keep rolling a die, but sometimes you get stuck in a loop that lasts a million years before you roll again.

Because of this, the usual "average" statistics break down. Instead of a normal bell curve, these systems are governed by an "Infinite Invariant Density."

The Core Concept: The "Ghost" Distribution

Think of a normal probability distribution (like a bell curve) as a pie. The whole pie represents 100% of the probability. You can slice it up, and the slices add up to a whole pie.

In the systems this paper studies, the "pie" is infinite. It's a pie that keeps growing forever. You can't normalize it (make it equal 1). It's a "ghost distribution" that describes how the system behaves over a very long time, even though it never settles down.

The paper asks: If we look at the "extremes" (the biggest or smallest values) in a system with this infinite pie, what happens?

The Magic Ingredient: The "Joint Limit"

Here is the twist. You can't just look at one particle for a long time, and you can't just look at a huge crowd at one specific moment. You have to do both at the same time.

Imagine you are watching a crowd of people (N) trying to escape a maze.

Time (t): You watch them for a very long time.
Crowd (N): You have a very large number of people.

The paper discovers a "sweet spot" (a specific ratio between the crowd size and the time watched). If you balance these two correctly, a new, beautiful pattern emerges. This pattern isn't one of the three standard textbook answers. It's a fourth, unique class of extreme statistics.

The Three Real-World Examples (The Metaphors)

To prove this, the authors tested three different "weird" systems:

1. The Drifting Drunkard (Overdamped Diffusion)

The Setup: Imagine a drunk person walking on a beach. Usually, they wander back and forth. But here, the beach has a "flat" area far away where there is no force pulling them back.
The Extreme: We look for the person who wandered the closest to the dangerous cliff (the origin).
The Result: Even though most people drift far away, the "closest" person is controlled by the "ghost distribution" near the cliff. The paper shows that if you have enough people and watch long enough, you can predict exactly where the closest person will be, based on the shape of the "infinite pie."

2. The Sticky Fly (Weakly Chaotic Maps)

The Setup: Imagine a fly buzzing around a room. Most of the time, it gets stuck in a corner (a "marginal fixed point") and buzzes there for a long time. Occasionally, it flies fast across the room.
The Extreme: We look for the fly that flew the farthest from the sticky corner.
The Result: Because the fly gets stuck so often, the "average" time it spends in the corner is infinite. The paper shows that the "farthest flyer" follows a specific rule based on how sticky the corner is. The more sticky the corner, the more the extreme values behave differently than normal.

3. The Laser Trap (Sub-recoil Laser Cooling)

The Setup: Imagine atoms being cooled by lasers. The lasers act like a trap that gets stronger the slower the atoms move. The slower an atom gets, the longer it stays trapped.
The Extreme: We look for the fastest atom in a group.
The Result: Since the slow atoms get stuck forever (infinite time), the "fastest" atom is actually a rare event that escaped the trap. The paper predicts exactly how fast this fastest atom will be, based on the "infinite density" of the trapped atoms.

Why Does This Matter?

This is like discovering a new law of physics for "rare events."

In the real world: Many systems (from financial crashes to climate extremes to how atoms behave in lasers) don't follow the "normal" rules. They have "heavy tails" or "infinite averages."
The Takeaway: If you try to use standard statistics to predict the worst-case scenario in these systems, you will be wrong. You need this new "Infinite Ergodic" math.
The Tool: The authors provide a formula. If you measure the "return exponent" (how long things get stuck) and the "infinite density" (the shape of the ghost pie), you can predict the extremes.

The Simple Summary

Imagine you are trying to predict the longest wait time at a bus stop.

Normal World: Buses come every 10 minutes. The longest wait is predictable.
This Paper's World: Buses come every 10 minutes, but sometimes the bus gets stuck in a time loop for a million years. The "average" wait time is infinite.

If you stand there for a long time with a huge group of friends, the paper tells you exactly how to calculate the probability of the longest wait among your group. It turns out, the answer depends on the "shape" of the time loop and the ratio of your group size to how long you've been waiting.

In short: When the rules of the universe get weird and "infinite," the extremes follow a new, hidden pattern. The authors found the map to that pattern.

Here is a detailed technical summary of the paper "Extreme Values of Infinite-Measure Processes" by Talia Baravi and Eli Barkai.

1. Problem Statement

Standard Extreme Value Theory (EVT) typically assumes that the underlying random variables are independent and identically distributed (i.i.d.) with a normalizable probability density function (PDF). Under these conditions, the statistics of the maximum or minimum of a large sample $N$ converge to one of three universal classes (Gumbel, Fréchet, or Weibull) as $N \to \infty$ .

However, many physical systems governed by infinite ergodic theory do not possess a normalizable invariant density. Instead, their long-time behavior is described by a non-normalizable infinite invariant density $I(x)$ . In these systems:

The mean return time to a reference region diverges.
Time-averaged observables do not converge to a single deterministic value but remain broadly distributed.
The parent distribution $p(x, t)$ decays algebraically over time, meaning the "typical" behavior changes with observation time $t$ .

The central problem addressed is: How do extreme value statistics behave in systems where the underlying measure is infinite, and how do the statistics of extremes scale when both the sample size $N$ and the observation time $t$ become large?

2. Methodology

The authors develop a theoretical framework that combines infinite ergodic theory with extreme value statistics by taking a joint limit where both $N \to \infty$ and $t \to \infty$ .

Key Theoretical Steps:

Infinite Invariant Density Definition: They define the infinite invariant density $I(x)$ via the scaling limit:
$I(x) = \lim_{t \to \infty} N t^{1-\alpha} p(x, t)$
where $\alpha \in (0, 1)$ is the return (or persistence) exponent, determined by the tail of the return-time distribution ( $P(\tau > t) \sim t^{-\alpha}$ ). $N$ is a model-specific normalization constant.
Scaling Regimes: The authors distinguish two cases based on the behavior of $I(x)$ :
- Case 1 (Singularity at $x \to 0$ ): $I(x) \sim x^{-\beta}$ as $x \to 0$ (e.g., weakly chaotic maps). Here, extremes are determined by the maximum of the variables.
- Case 2 (Singularity at $x \to \infty$ ): $I(x) \sim x^{-\beta}$ as $x \to \infty$ (e.g., diffusion in flat potentials). Here, extremes are determined by the minimum of the variables.
Joint Limit Scaling: The authors identify a critical scaling parameter $\rho$ that must remain fixed for non-trivial statistics to emerge:
$\rho \equiv N t^{\alpha-1} = \text{const}$
If $N$ grows too slowly relative to $t$ , the system behaves like a single trajectory (trivial limit). If $N$ grows too fast, the extreme is dominated by the bulk of the distribution. The fixed $\rho$ regime balances the decay of the single-particle probability with the number of samples.
Derivation of Limit Laws:
- For the Maximum (Case 1): The cumulative distribution function (CDF) of the maximum $X_{\max}$ converges to:
  $Q_{\max}(m) = \exp\left[ -\rho J(m) \right]$
  where $J(m) = \int_m^\infty I(u) du$ is the integrated infinite density (tail integral).
- For the Minimum (Case 2): The CDF of the minimum $X_{\min}$ converges to:
  $Q_{\min}(m) = 1 - \exp\left[ -\rho J(m) \right]$
  where $J(m) = \int_0^m I(u) du$ is the integrated density near the origin.

3. Key Contributions

New Universality Class: The paper establishes that extreme values in infinite-measure systems do not follow the classical Fréchet/Gumbel/Weibull laws. Instead, they are governed by the integrated infinite invariant density and the return exponent $\alpha$ .
Unified Framework: The theory connects three distinct fields: Extreme Value Statistics, Infinite Ergodic Theory, and First-Passage/Persistence Theory.
Analytical Solutions: The authors derive closed-form expressions for the limiting PDFs of extremes in terms of the system's specific infinite density $I(x)$ .
Experimental/Model Validation: The theory is rigorously tested against three distinct physical models, demonstrating its universality across deterministic and stochastic systems.

4. Results and Model Illustrations

The authors validate their theory using three representative models:

A. Overdamped Diffusion in an Asymptotically Flat Potential (Case 2)

System: A Brownian particle in a potential $V(x)$ where $V(x) \to 0$ as $x \to \infty$ (e.g., Lennard-Jones potential).
Dynamics: The return exponent is $\alpha = 1/2$ (standard diffusion). The infinite density is the non-normalizable Boltzmann weight: $I(x) \propto e^{-V(x)/k_B T}$ .
Result: The minimum position of $N$ $N$ particles is controlled by the rare particles trapped in the potential well.
- At low temperatures, the PDF of the minimum shows a peak near the potential minimum.
- As $T$ increases, thermal fluctuations flatten the distribution, approaching free diffusion.
- Numerical simulations confirm the collapse of data onto the predicted curve $1 - \exp[-\rho J(m)]$.

B. Weakly Chaotic Intermittent Maps (Case 1)

System: Pomeau–Manneville (PM) and Thaler maps, characterized by a marginally unstable fixed point at $x=0$ .
Dynamics: Trajectories spend long "laminar" phases near $x=0$ before chaotic excursions. The return exponent $\alpha = 1/(z-1)$ depends on the map's nonlinearity $z$ .
Infinite Density: $I(x) \sim x^{-1/\alpha}$ near the origin.
Result: The maximum of $N$ $N$ trajectories is determined by the rare excursions away from the fixed point.
- The distribution of the maximum depends on the integrated tail of $I(x)$ .
- Simulations show that increasing $\alpha$ (weaker trapping) shifts the maximum distribution toward larger values.

C. Sub-Recoil Laser Cooling (Case 1)

System: Atoms cooled below the recoil limit via velocity-selective dark states.
Dynamics: Modeled as a renewal process where the trapping time $\tau(v) \sim v^{-\gamma}$ diverges as velocity $v \to 0$ . Here $\alpha = 1/\gamma$ .
Infinite Density: $I(v) \propto v^{-\gamma}$ for small velocities.
Result: The maximum speed of $N$ $N$ atoms is governed by the tail of the infinite density.
- The PDF of the maximum speed exhibits an essential singularity at $v=0$ (highly unlikely to have a very small maximum speed).
- The peak of the distribution shifts to lower speeds as time increases (cooling effect) but to higher speeds as $N$ increases (more samples).

5. Significance and Implications

Inferring System Structure: The paper demonstrates that measuring extreme values (e.g., the maximum speed in a cooling cloud or the minimum position in a potential) provides a direct experimental probe of the infinite invariant density structure, which is otherwise difficult to measure directly in non-stationary systems.
Beyond Standard Statistics: It highlights that in systems with diverging time scales (glassy dynamics, anomalous diffusion, intermittent chaos), standard statistical tools fail. A new "thermodynamic" scaling for extremes is required, controlled by the parameter $\rho = N t^{\alpha-1}$ .
Physical Predictions: The theory predicts specific shapes for extreme value distributions that depend on the microscopic details of the potential or map (via $I(x)$ ), unlike the universal laws of classical EVT. This allows for the characterization of complex systems through their rare events.
Future Directions: The authors note open questions regarding order statistics (k-th extreme), unbounded processes, and the transition between sparse sampling (small $N$ ) and dense sampling regimes.

In summary, this work provides a rigorous mathematical foundation for understanding rare events in non-ergodic, infinite-measure systems, revealing that the "extremes" in such systems are not just statistical outliers but are fundamentally encoded in the system's infinite invariant measure.