Genuine and spurious (non-)ergodicity in single particle tracking

This paper critiques the limitations of the standard mean-squared displacement (MSD) criterion for assessing ergodicity in single-particle tracking, demonstrating that it can yield spurious results, and proposes a more robust alternative based on the mean-squared increment (MSI) that accurately characterizes genuine (non-)ergodicity and stationarity in anomalous diffusion systems.

The Gist

Imagine you are a detective trying to solve the mystery of how a tiny particle moves through a crowded room. This particle could be a protein in a cell, a speck of dust in the air, or even a stock price on a screen. Scientists use a technique called Single Particle Tracking (SPT) to follow these particles and figure out the rules of their movement.

For decades, scientists have used a specific "rule of thumb" to decide if a particle's movement is ergodic. In simple terms, ergodicity means that if you watch one particle for a very long time, you will learn the exact same things about its behavior as if you watched a million different particles for a short time. It's the difference between studying one person's entire life to understand human behavior versus studying a thousand people for one day.

The Old Detective's Tool: The "Starting Line" Ruler

The traditional tool scientists used to check for ergodicity was called the MSD (Mean Squared Displacement).

Think of the MSD like a ruler that always starts measuring from the starting line (time $t=0$).

• The Method: You measure how far the particle has traveled from the very beginning, over and over again.
• The Logic: If the average distance traveled by one long journey matches the average distance of many short journeys, the system is "ergodic" (fair and consistent).

The Problem: The authors of this paper argue that this ruler is flawed because it's obsessed with the starting line.

Imagine a runner who starts at the starting line but, after a few minutes, settles into a steady, comfortable jog.

• The Old Ruler (MSD): Says, "Hey, you started at the line! Your total distance depends on where you started. If you started far away, your total distance is huge. If you started close, it's small." It gets confused by the starting point.
• The Reality: Once the runner is jogging, their speed and pattern are consistent, regardless of where they started. The old ruler fails to see this consistency because it's too focused on the beginning.

This leads to two types of "fake" results:

1. Fake Ergodicity: The old ruler says a chaotic, non-repeating process is actually consistent (because the starting line math accidentally cancels out the chaos).
2. Fake Non-Ergodicity: The old ruler says a perfectly consistent, steady process is chaotic (because the starting line creates a mismatch with the long-term average).

The New Detective's Tool: The "Step-by-Step" Ruler

The authors propose a new tool called the MSI (Mean Squared Increment).

Think of the MSI as a ruler that doesn't care about the starting line. Instead, it only measures the steps the particle takes right now.

• The Method: Instead of asking "How far did you go from the start?", it asks "How big was your last step?" and "How big was the step before that?" It looks at the increments (the differences between positions) rather than the total distance.
• The Analogy: Imagine you are judging a dance routine.
  • The Old Ruler (MSD) asks: "How far is the dancer from the stage entrance?" (This depends on where they started).
  • The New Ruler (MSI) asks: "How big are their dance moves?" (This tells you about the style and rhythm, regardless of where they are on stage).

Why This Matters: The "Ghost" in the Machine

The paper shows that using the old ruler leads to "spurious" (fake) conclusions in many famous models of movement:

1. The "Fake" Non-Ergodicity (The Ornstein-Uhlenbeck Process):

   • Scenario: A particle is tethered to a spring. It wiggles around a central spot. It is perfectly stable and predictable.
   • Old Ruler: Says, "This is chaotic! The total distance from the start doesn't match the average step size."
   • New Ruler: Says, "No, look at the steps. They are consistent and stable. This is a perfectly ergodic system."
   • Result: The old tool falsely accused a calm system of being chaotic.

2. The "Fake" Ergodicity (Fractional Brownian Motion):

   • Scenario: A particle moves with "memory," where its past steps influence its future steps in a complex way. It never truly settles down.
   • Old Ruler: Says, "Hey, the math works out! It looks ergodic."
   • New Ruler: Says, "Wait, the steps themselves are changing over time. This system is actually non-ergodic."
   • Result: The old tool falsely gave a pass to a chaotic system.

3. The "Ultra-Weak" Mystery:

   • Sometimes, the old ruler sees the same pattern (scaling) but with a different size (prefactor). It's like seeing two runners with the same stride length but one is running 10% faster. The old ruler gets confused and calls it a mystery. The new ruler looks at the steps directly and realizes, "Ah, the steps are actually consistent; it's just the starting conditions that are different."

The Big Picture

The authors are essentially telling the scientific community: "Stop measuring from the starting line!"

In the real world, especially in biology (like tracking molecules inside a cell) or finance (tracking stock prices), we often don't know exactly when the "system" started or what the initial conditions were. We just have the data we collected.

By switching from measuring total distance from the start (MSD) to measuring step sizes (MSI), scientists can:

• Avoid false alarms about chaos.
• Stop missing real chaos.
• Get a true picture of whether a system is stable and predictable over time.

It's like switching from judging a movie based on the opening scene to judging it based on the quality of the dialogue throughout the whole film. The new method (MSI) gives a much more honest and accurate review of how the particle (or the system) really behaves.

Technical Summary

1. Problem Statement

In Single Particle Tracking (SPT) experiments, characterizing anomalous diffusion requires determining whether a stochastic process is ergodic. Ergodicity implies that the time average of an observable over a single, long trajectory converges to the ensemble average over many trajectories.

Currently, the standard criterion for assessing ergodicity in SPT is the comparison between the Mean Squared Displacement (MSD) and the Time-Averaged Mean Squared Displacement (TAMSD).

• MSD: $\langle [x(t) - x(0)]^2 \rangle$ (Ensemble average of displacement from the initial position).
• TAMSD: $\frac{1}{T-\Delta} \int_0^{T-\Delta} [x(t'+\Delta) - x(t')]^2 dt'$ (Time average of squared increments along a single trajectory).

The Core Issue: The authors argue that the widely accepted criterion—where equality between MSD and TAMSD implies ergodicity, and inequality implies ergodicity breaking—is theoretically flawed.

1. Definition Mismatch: The classical definition of ergodicity (Birkhoff's theorem) applies to stationary processes and compares time averages to ensemble averages of the same observable. The MSD depends explicitly on the initial condition $x(0)$, whereas the TAMSD averages over arbitrary increments $x(t+\Delta) - x(t)$.
2. Spurious Results: This mismatch leads to two types of errors:
   • Spurious Ergodicity: Non-stationary, non-ergodic processes with stationary increments (e.g., Brownian Motion, Fractional Brownian Motion) appear ergodic because MSD $\approx$ TAMSD.
   • Spurious Non-Ergodicity: Stationary, ergodic processes (e.g., Ornstein-Uhlenbeck Process) appear non-ergodic because MSD $\neq$ TAMSD (often by a factor of 2 in the stationary limit).

2. Methodology

The authors propose a new, rigorous criterion based on the Mean-Squared Increment (MSI) rather than the MSD.

• Proposed Observable (MSI):
  $$\text{MSI}(t, \Delta) = \langle [x(t+\Delta) - x(t)]^2 \rangle$$
  This represents the ensemble average of the squared increment starting at an arbitrary time $t$, not just $t=0$.
• New Criterion (MSI-TAMSD):
  Ergodicity is correctly assessed by comparing the MSI with the TAMSD:
  $$\text{MSI}(t, \Delta) \stackrel{?}{=} \text{TAMSD}(\Delta, T)$$
  • If $\lim_{t, T \to \infty} \text{MSI} = \text{TAMSD}$, the process (or its increments) is ergodic.
  • If they differ, genuine ergodicity breaking is present.
• Theoretical Basis: The MSI is equivalent to Kolmogorov's structure function, originally used in turbulence theory. It aligns with the classical definition of ergodicity for stationary increments.
• Validation: The authors analytically derive and numerically simulate the MSD, MSI, and TAMSD for 11 distinct stochastic models covering various diffusion regimes (sub-diffusion, super-diffusion, resetting, etc.). They also utilize the Ergodicity Breaking (EB) parameter to quantify the scatter of TAMSD amplitudes across trajectories.

3. Key Contributions

1. Identification of Spurious (Non-)Ergodicity: The paper demonstrates that the traditional MSD-TAMSD criterion yields incorrect classifications for fundamental models:
   • Brownian Motion (BM) & Fractional Brownian Motion (FBM): These are non-stationary and non-ergodic as processes, but their increments are stationary and ergodic. The MSD-TAMSD criterion falsely suggests ergodicity for the process itself. The MSI-TAMSD criterion correctly identifies that the increments are ergodic.
   • Ornstein-Uhlenbeck Process (OUP): This is a stationary and ergodic process. However, because the MSD retains memory of the initial condition while the TAMSD does not, MSD $\neq$ TAMSD (specifically, TAMSD $\approx 2 \times$ MSD in the stationary limit). The MSD-TAMSD criterion falsely flags this as non-ergodic. The MSI-TAMSD criterion correctly identifies it as ergodic.
2. Resolution of "Ultraweak" Ergodicity Breaking: The authors address systems where MSD and TAMSD share the same scaling exponent but differ by a constant prefactor (e.g., Riemann-Liouville FBM and Lévy walks).
   • The MSD-TAMSD criterion interprets this prefactor difference as ergodicity breaking.
   • The MSI-TAMSD criterion reveals that the increments in these systems are asymptotically stationary and ergodic, despite the process itself being non-ergodic.
3. Unified Framework: The paper provides a comprehensive table (Table I) summarizing the behavior of MSD, MSI, and TAMSD across various models (CTRW, Lévy walks, Resetting processes), offering a practical guide for experimentalists to distinguish genuine ergodicity breaking from artifacts of the measurement method.

4. Key Results

The study analyzes specific models to validate the MSI-TAMSD criterion:

• Fractional Brownian Motion (FBM):
  • Process: Non-ergodic (non-stationary).
  • Increments: Ergodic.
  • Result: MSD = TAMSD (Spurious ergodicity). MSI = TAMSD (Correctly identifies ergodic increments).
• Ornstein-Uhlenbeck Process (OUP) & Resetting FBM:
  • Process: Stationary and Ergodic.
  • Result: MSD $\neq$ TAMSD (Spurious non-ergodicity). MSI = TAMSD (Correctly identifies ergodicity).
• Riemann-Liouville FBM (RL-FBM) & Lévy Walks:
  • Phenomenon: "Ultraweak" ergodicity breaking (MSD and TAMSD scale identically but have different prefactors).
  • Result: The MSI converges to the TAMSD, proving that the increments become asymptotically stationary and ergodic at long times, even if the process itself is not.
• Continuous Time Random Walk (CTRW):
  • Power-law waiting times ($0 < \beta < 1$): Genuine weak ergodicity breaking. Both MSD and MSI differ from TAMSD, and the EB parameter does not vanish. The increments are non-ergodic.
  • Exponential waiting times: Ergodic. MSD = MSI = TAMSD.

5. Significance

• Theoretical Correction: The paper corrects a long-standing misconception in the SPT community regarding the interpretation of MSD-TAMSD discrepancies. It clarifies that comparing MSD and TAMSD is mathematically inconsistent for testing ergodicity because they measure different physical quantities (initial displacement vs. arbitrary increments).
• Experimental Utility: By introducing the MSI (Kolmogorov's structure function) as the standard observable, the authors provide a robust tool for researchers to:
  • Avoid false positives (claiming ergodicity for non-ergodic processes like FBM).
  • Avoid false negatives (claiming non-ergodicity for stationary processes like OUP).
  • Correctly identify "ultraweak" ergodicity breaking and the asymptotic stationarity of increments.
• Broad Applicability: The findings are relevant not only to biophysics (intracellular transport, membrane dynamics) but also to finance (volatility modeling) and geophysics, where distinguishing between genuine non-ergodicity and measurement artifacts is crucial for model selection and parameter regression.

In conclusion, the paper argues that MSI is the physically correct observable for testing ergodicity in SPT, replacing the MSD in the comparison with TAMSD to ensure results align with the classical mathematical definition of ergodicity.