How Weakening the PING Attractor Produces Realistic Gamma Oscillations in Silico
Gamma rhythms simulated in silico have long looked too clean to be real, and a study published on Friday in Communications Biology explains why — and how to fix the models.
The Tech TL;DR:
- Standard PING-based gamma models produce oscillations that are stronger and more stable than the sparse, transient gamma seen in living brains, according to the study by Scott Rich, Tarun Shriram and Steven A. Prescott.
- Three biophysical tweaks — increased heterogeneity, synapse-like noise, and a depolarized chloride reversal potential — each weaken the model’s attractor state and produce realistic, bursty gamma.
- The mechanism is reframed as noise-driven damped oscillations around a stable focus rather than a self-sustaining limit cycle, with implications for anyone building spiking-network simulations or neuromorphic systems.
Why simulated gamma rings too steady
Gamma-band oscillations — rhythms in the roughly 30–100 Hz range associated with cognition and attention — have for decades been modeled through the Pyramidal Interneuron Network Gamma (PING) mechanism: excitatory pyramidal cells drive inhibitory interneurons, which fire back and lock the network into a rhythm. In simulation, that loop settles into what dynamical systems theory calls a strongly attracting limit cycle, a self-sustaining orbit the network keeps tracing once it enters.
The problem, as the paper published in Communications Biology lays out, is that real gamma does not behave that way. In vivo, gamma activity is transient, driven by sparse spiking, and far weaker than what PING models produce. The simulated rhythm is, in the authors’ framing, too strong and too stable to match biology.
The study, authored by Rich, Shriram and Prescott, proposes the correction.

Three dials that break the attractor
The paper describes biophysically motivated changes that “degenerately” drive PING-motivated networks toward realistic gamma — meaning multiple distinct mechanisms converge on the same outcome. Per the study, each of the following weakens the attractor:
- Increased heterogeneity across neurons, so cells no longer fire in lockstep.
- Synapse-like noise injected into the connections, mimicking the stochastic release of neurotransmitter.
- A depolarized chloride reversal potential, shifting inhibitory signaling away from its idealized default.
With any of these in place, the network stops producing a continuous, strongly entrained rhythm. Instead, gamma-band power rises and falls spontaneously in transient bursts, with excitatory cells only weakly entrained — the sparse, intermittent signature observed in vivo, the authors report.
A Hopf bifurcation reframes the mechanism
The transient bursts, the authors write, mirror an oscillatory return to a stable focus after a noise-induced perturbation — a damped ringing, not a sustained orbit. Systems that exhibit both a stable focus and a limit cycle are associated with Hopf bifurcations, and the study shows those bifurcations are present in PING-motivated models of reciprocally connected excitatory and inhibitory populations.
The revised account keeps the core of PING — strong reciprocal excitation-inhibition connectivity — but changes what the system does with it. Biophysical phenomena bias the network toward damped oscillations rather than a stable limit cycle. The rhythm is a response to perturbations, not a self-sustaining engine.
What changes for modelers
The study suggests heterogeneity, synaptic noise and chloride dynamics are not nuisance parameters to suppress but the very mechanisms that make the output realistic.
# Parameter sweep inspired by Rich, Shriram & Prescott (2026)
# Weaken the PING attractor; watch gamma-band power become transient
import numpy as np
for heterogeneity in [0.0, 0.05, 0.15]: # spread in cell properties
for noise_sigma in [0.0, 2.0, 5.0]: # synapse-like noise (pA)
for e_cl in [-70.0, -60.0, -50.0]: # chloride reversal (mV)
run_network(
heterogeneity=heterogeneity,
noise_sigma=noise_sigma,
e_cl=e_cl,
metric="gamma_band_transient_power",
)
The open question
The paper proposes a mechanism and demonstrates it in silico; the sources do not report experimental validation in living tissue, and the study does not claim it. Whether the depolarized chloride reversal potential and synaptic noise levels used in the models match measured values in specific cortical circuits remains outside what the published abstract addresses.
Steven A. Prescott brings an unusual combination of training to this kind of work: he completed both an MD and a PhD at McGill University, where his doctoral research examined pain processing in the spinal cord, but chose not to pursue clinical training. Instead he took a postdoctoral fellowship in computational neuroscience at the Salk Institute in San Diego before founding his own lab at the University of Pittsburgh.
That lab later moved to the Hospital for Sick Children and the University of Toronto in 2012, and to the University of Calgary in 2025, where he is a full professor in the Cumming School of Medicine. He holds a Tier I Canada Research Chair in Neural Coding.
The paper, “Physiologically realistic gamma activity produced in silico by weakening the PING attractor state,” is available under its DOI, with licensing terms published by Springer Nature.