🌀 nonlinear sciences

Phenomenological renormalization group in neuronal models near criticality

This study validates the reliability of the phenomenological renormalization group (PRG) method for detecting genuine criticality in neuronal data by demonstrating that it yields consistent results only within a narrow vicinity of the critical point and by introducing a data-driven adaptive binning procedure to mitigate the substantial influence of time-binning choices.

Original authors: Kaio F. R. Nascimento, Daniel M. Castro, Gustavo G. Cambrainha, Mauro Copelli

Published 2026-07-02

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Kaio F. R. Nascimento, Daniel M. Castro, Gustavo G. Cambrainha, Mauro Copelli

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ 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 your brain is a massive, bustling city where billions of neurons are like individual citizens. Sometimes, these citizens act in perfect, chaotic harmony, creating a "critical" state where the city is most efficient at processing information. Scientists have been trying to prove this critical state exists by looking for specific patterns in how these neurons fire, using a mathematical tool called the Phenomenological Renormalization Group (PRG).

Think of PRG as a special pair of "zoom-out" glasses. When you look at the city through these glasses, if the city is truly in that special critical state, the patterns you see should look the same whether you are zooming in on a single street or zooming out to see the whole country. This is called "scale invariance."

However, the authors of this paper asked a crucial question: Are these glasses reliable? Could they be tricking us into seeing patterns that aren't really there?

Here is what they discovered, explained through simple analogies:

1. The Problem: The Wrong "Camera Shutter Speed"

To study the neurons, scientists have to take snapshots of their activity over time. They chop time into little chunks called "bins."

The Mistake: Imagine trying to photograph a hummingbird. If you use a shutter speed that is too fast, you only see empty air (silence). If you use a shutter speed that is too slow, you just see a blurry, solid blob (constant noise).
The Paper's Finding: The researchers found that if you use a fixed "shutter speed" (a fixed time bin) for the brain, you get fake results.
- If the brain is quiet (subcritical), a fast shutter makes it look like there are strange, sharp patterns because you are mostly seeing silence.
- If the brain is hyper-active (supercritical), a slow shutter makes it look like there are patterns because you are mostly seeing noise.
- The Result: In both cases, the "glasses" (PRG) falsely claimed the brain was in a special critical state, even when it wasn't. It was like a camera tricking you into thinking a blurry photo was a masterpiece.

2. The Solution: An "Adaptive" Camera

The researchers realized that to see the truth, the camera needs to adjust its shutter speed based on how fast the birds are actually flying.

The Fix: They created a new method where the size of the time bin automatically changes based on the average time between neuron spikes.
- If neurons are firing slowly, the bin gets bigger.
- If neurons are firing rapidly, the bin gets smaller.
The Result: With this "adaptive camera," the fake patterns disappeared. The PRG glasses only showed the special, critical patterns when the brain was actually near the critical point. When the brain was too quiet or too noisy, the patterns looked "boring" (Gaussian), which is exactly what you expect from a non-critical system.

3. The "Goldilocks" Zone

The study used two different computer models of the brain (one like a simple grid of cells, the other like a complex network with exciters and inhibitors) where they knew exactly where the "critical point" was.

They found that the special signatures of criticality only appear in a very narrow window right next to the critical point.
It's like standing on a mountain peak. If you take one step left or right, the view changes completely. The PRG method is very sensitive; it only works if you are standing almost exactly on that peak.

4. What This Means for Real Science

The paper concludes that previous studies claiming to find criticality in real brain data (from rats, humans, etc.) are likely correct, but only if they used the right way to process the data.

If researchers used a fixed time bin, their results might have been accidental artifacts (like the blurry photo).
By using the new "adaptive" method, the researchers confirmed that the brain does seem to operate near this critical point, but it's a delicate balance. The method is a powerful tool, but it requires a "sanity check" to ensure the time bins aren't distorting the picture.

In short: The brain's critical state is real, but detecting it is like trying to photograph a hummingbird. You can't use a fixed camera setting; you have to adjust your focus to the bird's speed, or you'll just see a blur and think you've found a pattern that isn't there.

Technical Summary: The Phenomenological Renormalization Group in Neuronal Models Near Criticality

Problem Statement
The phenomenological renormalization group (PRG) has become a prominent tool for identifying scale-invariant phenomena and critical behavior in neuronal data, supporting the "critical brain hypothesis." However, a significant gap remains in understanding the reliability of PRG signatures: it is unclear how close a system must be to a critical point for these signatures to emerge, or if they can be artifacts of data preprocessing. Specifically, the dependence of PRG results on time-binning choices has not been systematically investigated across different dynamical phases (subcritical, critical, and supercritical). Previous applications to canonical models (e.g., contact process, Ising model) lacked key biological features like inhibition, limiting their comparability with real neuronal data.

Methodology
The authors investigate the robustness of PRG by applying it to two neuronal models with exactly known critical points, both belonging to the mean-field directed percolation (MF-DP) universality class:

Excitable Cellular Automaton: A spiking model on a random network where the control parameter is the branching ratio ( $\sigma$ ).
Stochastic Integrate-and-Fire (I&F) Network: A network with excitatory and inhibitory neurons where the control parameter is the inhibition-to-excitation coupling ratio ( $g$ ).

The study employs two PRG coarse-graining approaches:

Momentum-space: Uses the eigenvalue spectrum of the covariance matrix to project data into progressively coarser representations. The deviation from Gaussianity in the distribution of coarse-grained variables ( $\psi$ ) is measured using Kurtosis ( $\kappa$ ) and Kullback-Leibler (KL) divergence.
Real-space: Iteratively pairs the most correlated neurons to form clusters, analyzing the scaling of intra-cluster mean variance ( $M^2 \propto C_{size}^\alpha$ ).

Crucially, the authors systematically vary the time-binning size ( $\Delta t$ ) and introduce a data-driven adaptive binning strategy. Instead of a fixed $\Delta t$ , the bin size is set proportional to the average interspike interval ( $\langle ISI \rangle$ ) of the network ( $\Delta t = f \cdot \langle ISI \rangle$ ). This ensures the active bin density ( $\rho_{bin}$ ) remains stable across dynamical phases. Surrogate data, generated by shuffling interspike intervals to destroy correlations, serves as a control.

Key Results

Timescale Mismatch Creates Spurious Signatures: When using fixed time bins, the PRG method yields misleading results.
- In the subcritical regime, small bins result in sparse data (mostly zeros), causing the coarse-grained distribution to peak sharply at zero and artificially inflate kurtosis, mimicking criticality.
- In the supercritical regime, large bins lead to saturation (mostly ones), similarly producing a peaked distribution and false scaling signatures.
- No single fixed bin size can accommodate both phases without generating artifacts.
Adaptive Binning Reveals Genuine Criticality: By using the adaptive binning strategy ( $\Delta t \propto \langle ISI \rangle$ ), the authors demonstrate that PRG signatures of criticality (non-Gaussian distributions and non-trivial scaling exponents) appear only within a narrow vicinity of the critical point.
- The KL divergence between the coarse-grained distribution and a Gaussian peaks sharply at the critical point and drops to near-zero in subcritical and supercritical phases.
- The real-space scaling exponent $\alpha$ transitions from the uncorrelated baseline ( $\alpha=1$ ) to a non-trivial regime ( $1 < \alpha < 2$ ) only near criticality.
- These results are robust across a range of scaling factors ( $f$ ), provided the binning maintains a stable active density.
Model Validation: The inclusion of inhibition in the stochastic I&F model yields even clearer critical signatures, reinforcing the relevance of these findings for biological systems. The results hold even under heavy subsampling (simulating limited experimental observability).

Significance and Claims
The paper provides the first systematic computational validation that PRG signatures of scale invariance are reliable indicators of proximity to a critical point, provided that data preprocessing (specifically time-binning) is handled correctly.

Reinforcement of Experimental Interpretations: The finding that scaling behavior is restricted to a narrow window around the critical point reinforces the interpretations drawn from PRG analyses of experimental neuronal data, suggesting that observed signatures likely reflect genuine critical dynamics rather than methodological artifacts.
Methodological Correction: The study highlights that inappropriate time-binning can introduce false signatures of scaling. The proposed adaptive binning procedure offers a practical solution to circumvent this issue, ensuring that PRG results reflect the intrinsic dynamics of the system rather than the arbitrary choice of $\Delta t$ .
Scope: The authors conclude that while PRG is a powerful probe, its application requires careful attention to the timescales of the underlying dynamics to distinguish genuine criticality from trivial behavior.

1. The Problem: The Wrong "Camera Shutter Speed"

2. The Solution: An "Adaptive" Camera

3. The "Goldilocks" Zone

4. What This Means for Real Science

More like this