Information-fluctuation inequalities for collective response

This paper derives a universal upper bound on relative fluctuations in many-particle systems driven by hidden stochastic effects, establishing an information-fluctuation inequality that links macroscopic correlations to generalized mutual information and demonstrating its validity through applications to non-interacting Brownian gases.

The Gist

Imagine you are at a crowded concert. Usually, if everyone is just standing there, their movements are random. If you look at the crowd as a whole, the "wiggles" and "jiggles" of the group tend to cancel each other out. The bigger the crowd, the smoother the overall picture looks. In physics, we call this self-averaging: as a system gets huge, the random noise of individuals disappears, leaving a predictable average.

But what if there's a hidden conductor?

The Hidden Conductor

This paper explores a scenario where a group of people (or particles) aren't talking to each other, but they are all listening to the same hidden radio signal. Let's call this signal the "Hidden Variable."

Imagine 1,000 people in a room.

• Normal Scenario: Each person is flipping their own coin. If you count the total number of "heads," the result will be very close to 500. If you add another 1,000 people, the percentage of heads stays exactly 50%. The randomness averages out.
• The "Hidden Conductor" Scenario: Imagine there is a hidden radio playing in the room. When the radio plays a loud "Go!" signal, everyone flips their coin to "Heads." When it plays a "Stop!" signal, everyone flips to "Tails." The people aren't talking to each other; they are just reacting to the same invisible force.

In this second scenario, even if you have a million people, the crowd doesn't smooth out. Instead, the whole group swings wildly together. Sometimes the whole room is 100% Heads; sometimes it's 100% Tails. The "relative fluctuation" (the size of the wobble compared to the total size) does not disappear as the crowd gets bigger.

The Big Discovery: The "Information" Limit

The author, Kristian Stølevik Olsen, asks a crucial question: How wild can this collective swinging get?

Is there a limit to how much the group can wobble, even with a hidden conductor?

The paper says yes, and the limit is determined by Information.

Think of it like a detective game:

• The System: The crowd of people.
• The Hidden Variable: The radio signal.
• The Clue: If you look at the crowd and see them all jumping, you instantly know the radio is playing "Go!"

The paper proves that the amount of "wobble" (fluctuation) the crowd can exhibit is directly tied to how much you can learn about the hidden radio signal just by watching the crowd.

• Low Information: If the crowd looks random even when the radio is playing, the radio isn't really controlling them. The wobble is small.
• High Information: If the crowd's behavior perfectly reveals the radio signal (e.g., they jump only when the signal is high), the wobble can be huge.

The paper provides a mathematical "speed limit" for this chaos. It says:

The size of the collective wobble cannot exceed a specific value determined by how much the crowd's behavior tells you about the hidden signal.

They call this a "Generalized Mutual Information." In simple terms, it's a measure of how "tuned in" the group is to the hidden force.

Two Real-World Examples

The author tests this theory with two scenarios:

1. The Brownian Gas (The Drifting Particles)
Imagine a gas of particles floating in a box. Usually, they bounce around randomly. Now, imagine the whole box is being shaken by a random, invisible hand (the hidden variable).

• The Result: The particles don't just bounce randomly; they all drift together with the shaking hand.
• The Application: This helps predict how fast chemical reactions happen. If a reaction needs particles to meet, and they are all drifting together because of the "shaking hand," the reaction rate will fluctuate wildly. The paper gives a formula to predict the maximum size of these fluctuations.

2. The Random Trap (The Sudden Energy Cost)
Imagine a bunch of particles diffusing freely. Suddenly, at a random time, a "trap" (a potential energy wall) snaps shut, trapping them.

• The Result: The energy required to snap the trap shut depends on where the particles were at that exact moment. Since the trap snaps at a random time, the energy cost fluctuates.
• The Application: This helps us understand the "cost" of resetting systems (like erasing data or resetting a biological clock). The paper shows that the energy cost will fluctuate, but the size of that fluctuation is strictly limited by how much the timing of the trap tells us about the particles' positions.

The Takeaway

In a world where things are influenced by hidden, shared forces (like climate change affecting all crops, or a market crash affecting all stocks), we cannot assume that "big numbers" will smooth out the chaos.

This paper gives us a new rulebook: The chaos of the group is limited by the "secret handshake" between the group and the hidden force. If the group doesn't know the secret (low information), the chaos is small. If the group is perfectly synced to the secret (high information), the chaos can be massive, but it still has a hard ceiling defined by information theory.

It's a beautiful bridge between Information Theory (how much we know) and Physics (how much things wiggle).

Technical Summary

1. Problem Statement

The paper addresses a fundamental question in statistical physics: How do macroscopic fluctuations behave in many-particle systems where individual particles are causally independent but share a common, fluctuating external driver (a "hidden variable")?

• Context: In conventional interacting many-body systems, macroscopic observables are typically "self-averaging," meaning relative fluctuations decay as $n^{-1/2}$ (where $n$ is the system size) as the system approaches the thermodynamic limit.
• The Anomaly: In systems with conditionally independent degrees of freedom (e.g., non-interacting particles subject to a shared stochastic force, diffusing diffusivity, or collective resetting), the shared driver induces strong correlations.
• The Gap: While it is known that these systems exhibit "super-extensive" scaling (fluctuations do not vanish in the thermodynamic limit), there was no universal, non-perturbative bound quantifying the maximum magnitude of these remnant fluctuations. The paper seeks to derive a universal upper bound for these fluctuations in terms of information-theoretic quantities.

2. Methodology

The author employs a combination of statistical mechanics, probability theory, and information geometry to derive the bounds.

• Model Setup:
  • Consider a system of $n$ particles with states $x_1, \dots, x_n$.
  • The particles are conditionally independent given a hidden random variable $\Phi$ (representing the global disorder).
  • The observable of interest is additive: $O_n = \sum_{j=1}^n O(x_j)$.
• Variance Decomposition:
  • Using the Law of Total Covariance, the variance of the macroscopic observable is decomposed into an extensive part (intrinsic particle noise) and a super-extensive part (correlations induced by $\Phi$):
    $$\text{Var}(O_n) = n\text{Var}(O) + n(n-1)\text{Var}_\Phi(\langle O|\Phi \rangle)$$
  • This leads to a coefficient of variation ($c_V$) that does not vanish as $n \to \infty$.
• Information-Theoretic Tool:
  • The paper utilizes the $\chi^2$-distance (chi-squared divergence) between probability densities.
  • A key inequality relates the change in the expectation value of an observable to the $\chi^2$-distance between two distributions:
    $$\frac{|\langle O \rangle_p - \langle O \rangle_q|^2}{\text{Var}_q(O)} \leq \chi^2(p \parallel q)$$
  • By setting $p = p(x|\phi)$ (conditional) and $q = p(x)$ (marginal), the author bounds the variance of the conditional mean $\text{Var}_\Phi(\langle O|\Phi \rangle)$.
• Derivation of the Bound:
  • The term $\langle \chi^2(p(\cdot|\phi) \parallel p(\cdot)) \rangle_\phi$ is identified as a generalized mutual information, denoted $I_{\chi^2}(X; \Phi)$.
  • This quantity replaces the standard Kullback-Leibler (KL) divergence-based mutual information to provide a tighter, non-perturbative bound valid for large deviations.

3. Key Contributions

The primary contribution is the derivation of a universal information-fluctuation inequality that limits the relative fluctuations of macroscopic observables in conditionally independent systems.

The Main Result (Eq. 9 & 10):
For a system of size $n$, the ratio of the coefficient of variation of the macroscopic observable ($c_V(O_n)$) to that of the microscopic component ($c_V(O)$) is bounded by:
$$\frac{c_V(O_n)}{c_V(O)} \leq \sqrt{\frac{1}{n} + \left(1 - \frac{1}{n}\right) I_{\chi^2}(X; \Phi)}$$

In the thermodynamic limit ($n \to \infty$):
$$\lim_{n \to \infty} \frac{c_V(O_n)}{c_V(O)} \leq \sqrt{I_{\chi^2}(X; \Phi)}$$

Significance of the Result:

• Universality: The bound depends only on the statistical coupling between the system states and the hidden variable, not on the specific form of the observable $O$.
• Non-Perturbative: Unlike fluctuation-response relations derived from linear response theory, this bound holds for strong disorder and large deviations.
• Physical Interpretation: It establishes that the persistence of macroscopic fluctuations in the thermodynamic limit is strictly determined by the generalized mutual information between the hidden driver and the system state. If $I_{\chi^2} = 0$ (no coupling), self-averaging is restored.

4. Results and Applications

The author validates the theoretical bounds using two distinct physical scenarios involving Brownian particles:

Case I: Collective Driving of a Brownian Gas (Reaction Rates)

• System: $n$ Brownian particles in a harmonic trap subject to a common Ornstein-Uhlenbeck stochastic force.
• Observable: Instantaneous reaction rate (fraction of particles in a specific domain).
• Result: The generalized mutual information is calculated exactly as $I_{\chi^2} = \frac{B\tau}{\kappa k_B T}$.
• Finding: The bound correctly predicts that reaction rate fluctuations scale linearly with noise strength at low noise but scale as $\sqrt{B}$ at high noise strengths, demonstrating a non-perturbative regime where fluctuations grow slower than the source disorder.

Case II: Energetic Cost of Random Activation (Stochastic Resetting)

• System: Particles diffuse freely for a random time $\tau$ (exponentially distributed), after which a harmonic potential is suddenly activated.
• Observable: The thermodynamic work (energetic cost) required to activate the trap.
• Result: For exponentially distributed waiting times, the generalized mutual information is derived to be exactly $I_{\chi^2} = 1/2$.
• Finding: The derived bound $\frac{c_V(E_n)}{c_V(w_1)} \leq \sqrt{\frac{n+1}{2n}}$ is shown to hold for all $n$, including for complex disordered potentials (cosine landscapes), confirming the universality of the bound beyond simple harmonic cases.

5. Significance and Implications

• Breaking Self-Averaging: The paper provides a rigorous mathematical framework explaining why and how much self-averaging breaks down in systems driven by global disorder, a phenomenon previously observed but not universally bounded.
• Connection to Information Theory: It bridges statistical physics and information theory, showing that the "cost" of collective response (fluctuations) is directly proportional to the information shared between the environment and the system.
• Practical Utility: The bounds offer a way to constrain the variability of macroscopic quantities (like reaction rates or work extraction) in biological and chemical systems subject to environmental noise, without needing to simulate the full many-body dynamics.
• Generalization: The use of $\chi^2$-distance allows for bounds that are valid even when the system is far from equilibrium or when the disorder is strong, extending the reach of fluctuation-response relations beyond the linear regime.

In summary, Olsen demonstrates that hidden stochastic drivers generate a "memory" of the disorder in the macroscopic fluctuations, and the magnitude of this memory is strictly capped by a generalized mutual information measure. This provides a fundamental limit on the predictability and stability of collective phenomena in disordered environments.