Dissipative Avalanche Regimes Driven by Memory-Biased Random Walks on Networks

This paper investigates a network model where a memory-biased random walker drives stress accumulation and avalanche dynamics, revealing that while memory shapes stress localization, the emergence of broad, power-law-like cascades versus runaway events is primarily governed by the interplay of dissipation rules, stress balance, and network topology.

The Gist

Imagine a giant, bustling city where a single delivery driver (the "walker") is constantly moving from house to house. This driver has a unique habit: they don't just wander randomly. They have a memory. If they've visited a house a lot before, they are more likely to return there again.

Every time the driver drops off a package, it adds a tiny bit of "stress" (like a heavy box) to that house. Most houses can handle a few boxes, but if a house gets too many, it overloads. When it overloads, it dumps its boxes onto its neighbors. If those neighbors get too heavy, they dump their boxes too, creating a chain reaction. In physics, we call this a "sandpile avalanche."

The paper asks a simple question: What makes these avalanches huge and unpredictable (like a self-organizing critical system), and what makes them small and manageable?

Here is the breakdown of their findings using everyday analogies:

1. The "Memory" Factor: Does the Driver's Habit Matter?

You might think that because the driver keeps returning to the same popular houses (the "memory"), this would cause massive, city-wide traffic jams (avalanches).

The Surprise: The driver's memory does matter, but not in the way you'd expect.

• What it does: It creates "hotspots." The driver keeps dropping boxes on the same few popular houses, making them the most likely to overflow.
• What it doesn't do: It doesn't actually control how big the resulting chain reaction is. Whether the driver remembers the past or just wanders randomly, the size of the avalanche is mostly determined by how the boxes are redistributed, not the order in which they arrive.

2. The "Brittle" Rule: The Glass House Problem

The researchers first tried a simple rule: When a house overloads, it dumps exactly the same amount of weight onto every neighbor, no matter how many neighbors it has.

• The Analogy: Imagine a house with 4 neighbors. It dumps 100 lbs on each. Total dumped: 400 lbs. But the house only had 100 lbs of stress to begin with! It created more weight than it lost.
• The Result: This is like a glass house. It works fine until you hit a very specific, narrow balance point. If you add just a tiny bit more weight, the whole system explodes into a runaway disaster (a "runaway" event) that never stops. If you add a tiny bit less, nothing happens. It's too fragile to be useful.

3. The "Dissipative" Rule: The Shock Absorber

To fix the glass house, the researchers tried a new rule: When a house overloads, it keeps a tiny bit of the weight for itself (dissipation) and only passes the rest on.

• The Analogy: Imagine the house has a shock absorber. When it dumps its load, it keeps 0.5% of the weight in a "trash can" and only passes 99.5% to the neighbors.
• The Result: This tiny bit of loss changes everything. Even though the driver is still dropping boxes on the same hotspots, the chain reaction now stops. The avalanches stay large and interesting (following a "power law," which means you get a mix of small, medium, and large events), but they never spiral out of control. The system becomes stable and robust.

4. The "Shuffled" Test: Does Timing Matter?

To prove that the driver's memory wasn't the secret sauce, the researchers did a clever experiment. They took the exact same list of houses the driver visited, but they shuffled the order (randomized the timeline).

• The Finding: The size of the avalanches stayed almost exactly the same.
• The Lesson: It doesn't matter when the boxes arrive, only where they land. The "memory" just decides which houses get hit, but the rules of the game (the shock absorbers) decide how big the explosion is.

5. The "Hub" Problem: The Busy Airport

Finally, they tested this on a different type of city: one with a few massive "hubs" (like a giant airport) and many small villages.

• The Issue: If a giant hub overloads, it has hundreds of neighbors. If it dumps weight on all of them, it creates a massive explosion.
• The Fix: They had to change the rules so the hub only dumps a total amount of weight, regardless of how many neighbors it has. Even with this fix, the results on these "hub-heavy" networks looked more like a simple exponential decay (mostly small events) rather than the complex, critical behavior seen in the other networks.

The Big Takeaway

The paper concludes that we shouldn't blame "memory" for creating complex, critical systems.

• Memory just creates the "hotspots" where stress builds up.
• The Rules (specifically, having a tiny bit of "loss" or dissipation) are what make the system stable and interesting.
• The Network Shape (whether it's a small-world neighborhood or a hub-heavy city) dictates the final outcome.

In short: You can have a driver with a great memory, but if the houses don't have "shock absorbers" to absorb a tiny bit of the stress, the whole city will either do nothing or collapse into chaos. The secret to a healthy, dynamic system isn't the driver's memory; it's the ability of the system to let go of a little bit of energy along the way.

Technical Summary

1. Problem Statement

The study investigates the interplay between memory-biased random walks (where a walker preferentially returns to previously visited nodes) and threshold-driven stress relaxation (sandpile-like dynamics) on complex networks.

• Context: While Self-Organized Criticality (SOC) is often invoked to explain broad avalanche distributions in complex systems, modern analysis suggests that broad tails alone do not guarantee asymptotic criticality. Conservation, dissipation, finite-size scaling, and network topology must be rigorously assessed.
• Core Question: Do memory-driven driving mechanisms generate true SOC, or are the observed broad cascade distributions artifacts of specific transfer rules, dissipation levels, or network topologies? The authors aim to disentangle the effects of the temporal ordering of memory from the spatial distribution of stress injection.

2. Methodology

The authors propose a coupled model on two types of networks: Watts–Strogatz (WS) (small-world, homogeneous degree) and Barabási–Albert (BA) (scale-free, heterogeneous degree).

A. The Driving Mechanism: Memory-Biased Random Walk

A single walker moves on the network $G=(V, E)$ with $N$ nodes. At each step:

• With probability $1-q$, the walker moves to a uniformly chosen neighbor (diffusion).
• With probability $q$, the walker resets to a previously visited node $k$ with probability proportional to its visitation count $v_k(t)$.
• Parameter: $q \in [0, 1]$ controls the strength of memory bias.

B. Stress Dynamics and Redistribution

Each arrival deposits 1 unit of stress ($s_i$) on the target node. If $s_i > T$ (threshold), the node "topples," redistributing stress to neighbors. The study compares three redistribution rules:

1. Fixed Per-Neighbor Transfer: A toppling node clears its stress and adds a fixed amount $\alpha$ to every neighbor.
2. Degree-Normalized Transfer: Stress is distributed such that the total released is fixed at $\alpha$, regardless of the node's degree.
3. Subtractive Dissipative Transfer: The node loses exactly $T$ units, but only a fraction $\beta T$ (where $0 < \beta \le 1$) is redistributed to neighbors ($\beta < 1$ implies dissipation).

C. Simulation and Diagnostics

• Networks: WS ($k=4$, rewiring $p$) and BA ($m=2$). Threshold $T=6$.
• Metrics:
  • Fraction of system-scale events ($S \ge 0.1N$).
  • Tail fits using Discrete Maximum Likelihood Estimation (DMLE) compared via AIC against Power Law, Exponential, and Lognormal distributions.
  • Bootstrap Kolmogorov–Smirnov (KS) Test: To test the null hypothesis of a pure power law.
  • Branching Ratio Proxy: Mean number of new unstable nodes created per toppling.
• Control Experiment: Shuffled-Order Controls. The temporal sequence of arrivals is randomized while preserving the empirical node-visit frequencies. This isolates the effect of temporal memory ordering from the spatial visitation distribution.

3. Key Results

A. Fixed Transfer on WS Networks (Brittle Transition)

• The fixed per-neighbor rule exhibits a sharp transition near the stress-balance condition $\alpha k \approx T$.
• Subcritical ($\alpha k < T$): Cascades remain short; no broad distributions.
• Supercritical ($\alpha k > T$): The system becomes unstable, generating "runaway-capped" events (system-spanning avalanches) as $N$ increases.
• Conclusion: The fixed rule does not support a stable, broad finite-cascade regime; it is "brittle" and highly sensitive to small parameter changes.

B. Subtractive Dissipative Rule (Stabilization)

• Introducing small dissipation ($\beta = 0.995$ or $0.998$) stabilizes the system.
• Regime: Broad, non-runaway cascades persist up to $N=4096$.
• Criticality Check:
  • AIC favors power-law tails.
  • However, the bootstrap KS test rejects the pure power-law hypothesis ($p_{boot} < 0.01$).
  • The system-scale event fraction decreases with $N$.
  • The branching ratio proxy stabilizes below unity ($\approx 0.81–0.83$).
• Interpretation: The system operates in a broad finite dissipative regime, not at a true SOC critical point. The distributions are heavy-tailed at accessible scales but do not converge to a critical fixed point as $N \to \infty$.

C. Role of Temporal Memory

• Shuffled-Order Controls: When the temporal order of arrivals is randomized (preserving visit frequencies), the macroscopic avalanche statistics (system-scale fraction, branching ratio) remain nearly identical for $\beta < 1$ across all memory strengths ($q=0$ to $0.6$).
• Conclusion: Memory biases the spatial distribution of stress (creating "hotspots" on frequently visited nodes), but the temporal ordering of these visits has a negligible effect on avalanche macrostatistics in the dissipative regime.

D. Behavior on Barabási–Albert (BA) Networks

• Fixed Rule: Highly hub-sensitive; high-degree nodes inject massive stress, causing artificial supercriticality.
• Degree-Normalized Rule: Suppresses runaways but yields distributions better described by exponentials rather than power laws.
• Conclusion: On heterogeneous networks, standard transfer rules can produce misleading SOC signatures; normalization stabilizes the system but eliminates scale-free behavior.

4. Key Contributions

1. Identification of a Dissipative Regime: The study demonstrates that a subtractive dissipative rule ($\beta \lesssim 1$) can stabilize broad avalanche distributions on WS networks, preventing the runaway instability seen in conservative fixed-transfer models.
2. Refutation of Memory-Driven SOC: The authors show that memory-biased driving does not inherently generate SOC. The broad distributions observed are a result of the specific balance between stress injection, dissipation, and topology, rather than the temporal memory mechanism itself.
3. Decoupling Spatial and Temporal Effects: Through shuffled-order controls, the paper proves that in dissipative regimes, the frequency of node visits (spatial structure) dominates avalanche statistics, while the sequence of visits (temporal memory) is secondary.
4. Topology Dependence: Highlights that transfer rules must be adapted to network topology (e.g., degree-normalization for BA networks) to avoid artifacts, and that "broad tails" do not automatically imply criticality.

5. Significance and Implications

• Theoretical Impact: The paper challenges the simplistic application of SOC language to network cascades. It argues for a more precise classification of regimes (e.g., "broad finite dissipative" vs. "asymptotic critical").
• Methodological Rigor: It emphasizes the necessity of using multiple diagnostics (AIC, KS tests, finite-size scaling, branching ratios) rather than relying solely on visual inspection of power-law tails.
• Practical Application: The findings are relevant for modeling real-world systems where non-conservative loading and heterogeneous structures are unavoidable, such as:
  • Neural Networks: Understanding seizure propagation or neural avalanches.
  • Infrastructure Systems: Analyzing cascading failures in power grids or communication networks.
• Design Principle: The identification of a specific dissipative redistribution design ($\beta \approx 0.995$) offers a mechanism to achieve robust, broad cascades without the instability of runaway events, which is crucial for designing resilient or controllable networked systems.

Final Conclusion: The model is best understood as a framework for exploring network cascade regimes driven by stress balance and dissipation, rather than as a mechanism for self-organized criticality. Memory shapes the stress landscape, but the redistribution rules and topology dictate the criticality.