pith. sign in

arxiv: 2606.30366 · v1 · pith:D2XDNY2Xnew · submitted 2026-06-29 · 🧬 q-bio.NC

Mean-field theory of rich oscillatory dynamics in low-rank recurrent networks with activity-dependent adaptation

Bowen W. Zheng , Earl K. Miller , Ila R. Fiete This is my paper

Pith reviewed 2026-06-30 03:07 UTC · model grok-4.3

classification 🧬 q-bio.NC
keywords mean-field theoryrecurrent networksneural adaptationchaotic dynamicsHopf bifurcationoscillatory regimeslow-rank connectivityneural dynamics
0
0 comments X

The pith

Mean-field theory shows adaptation driving low-rank chaotic networks through four distinct oscillatory regimes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper develops a dynamical mean-field theory for recurrent neural networks that combine low-rank structure with random connectivity and firing-rate-driven adaptation. It establishes that when random connections are strong enough to produce chaos, rising adaptation strength moves the network through four regimes: a static coherent state, noise-sustained oscillations that shift from regular to irregular, stochastic switching between symmetric states, and a global limit cycle. The theory accounts for these transitions by identifying chaos onset and a Hopf bifurcation of the coherent mode as the two instability mechanisms, both controlled by the frequency dependence that adaptation imposes on single-neuron responses. A reduced three-dimensional model is shown to reproduce the full network's sequence of bifurcations. These patterns match rhythms recorded in brains during wakefulness, sleep, and anesthesia.

Core claim

We develop a dynamical mean-field theory for random recurrent networks with low-rank structure and firing-rate-driven adaptation. When the random connectivity is strong enough to generate chaos, increasing adaptation strength drives the network through four regimes: a static coherent state, noise-sustained oscillations that progress from regular to irregular, stochastic switching between symmetric wells, and a global limit cycle. The theory identifies two instability mechanisms, chaos onset from the random connectivity and a Hopf bifurcation of the coherent mode, and shows how adaptation shapes both through the frequency-dependent single-neuron transfer function. A reduced three-dimensional

What carries the argument

The reduced three-dimensional dynamical model that captures the bifurcation structure of the full network through the frequency-dependent single-neuron transfer function.

If this is right

  • Coherent population-level oscillations can coexist with heterogeneous single-neuron firing rates and network-generated stochasticity.
  • The network produces waxing-and-waning rhythmic episodes, persistent state switching, and slow Up-Down alternations.
  • The two instability mechanisms are both shaped by the frequency dependence that adaptation imposes on neuronal responses.
  • The reduced three-dimensional model reproduces the bifurcation sequence of the full network.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Varying adaptation strength in this framework could model transitions between brain states such as wakefulness and anesthesia.
  • The reduced model offers a route to analyze how low-rank structure interacts with chaos in networks beyond the cases studied here.
  • Recordings from cortical populations during changing arousal levels could test whether observed dynamics shift regimes at the predicted adaptation strengths.

Load-bearing premise

The mean-field approximation remains valid once random connectivity exceeds the chaos threshold, allowing the frequency-dependent transfer function to control the observed regime transitions.

What would settle it

Numerical integration of the full network equations that does not reproduce the predicted sequence of four regimes as adaptation strength is increased above the chaos threshold.

Figures

Figures reproduced from arXiv: 2606.30366 by Bowen W. Zheng, Earl K. Miller, Ila R. Fiete.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5 [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6 [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7 [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8 [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9 [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10 [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11 [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗
read the original abstract

We develop a dynamical mean-field theory for random recurrent networks with low-rank structure and firing-rate-driven adaptation. When the random connectivity is strong enough to generate chaos, increasing adaptation strength drives the network through four regimes: a static coherent state, noise-sustained oscillations that progress from regular to irregular, stochastic switching between symmetric wells, and a global limit cycle. The theory identifies two instability mechanisms, chaos onset from the random connectivity and a Hopf bifurcation of the coherent mode, and shows how adaptation shapes both through the frequency-dependent single-neuron transfer function. A reduced three-dimensional model captures the bifurcation structure of the full network. Above the chaos threshold, coherent population-level oscillations coexist with heterogeneous firing rates and network-generated stochasticity at the single-neuron level. The interaction of adaptation with random and low-rank connectivity produces a rich oscillatory repertoire, including waxing-and-waning rhythmic episodes, persistent state switching, and slow Up-Down alternations, dynamics that have been observed during wakefulness, sleep, and anesthesia.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper develops a dynamical mean-field theory (DMFT) for recurrent networks combining random connectivity, low-rank structure, and firing-rate-driven adaptation. Above the chaos threshold set by random connectivity, increasing adaptation strength produces four regimes (static coherent state, noise-sustained oscillations progressing from regular to irregular, stochastic switching between symmetric wells, and a global limit cycle). The theory attributes the transitions to two mechanisms—chaos onset from random connectivity and a Hopf bifurcation of the coherent mode—both shaped by the frequency-dependent single-neuron transfer function, and supplies a reduced three-dimensional model that reproduces the bifurcation structure of the full network.

Significance. If the DMFT derivation holds, the work supplies an analytic handle on how adaptation interacts with structured and random connectivity to generate biologically relevant oscillatory patterns (waxing-and-waning rhythms, Up-Down states, persistent switching). The reduced 3D model is a concrete strength, as it makes the bifurcation diagram tractable and falsifiable against network simulations.

major comments (2)
  1. [Derivation of the DMFT equations (likely §3)] The central claim that the frequency-dependent transfer function of an isolated neuron governs both the chaos threshold and the Hopf bifurcation rests on the assumption that low-rank deterministic correlations remain negligible in the self-consistent equations for mean and variance once random connectivity exceeds the chaos threshold. The manuscript must derive the additional cross-terms arising from the low-rank component plus the temporal filtering of adaptation and show (analytically or numerically) that they do not alter the effective transfer function used to locate the instabilities.
  2. [Reduced three-dimensional model (likely §4 or §5)] The reduced three-dimensional model is stated to capture the bifurcation structure, but no explicit mapping is given between its parameters and the full-network statistics (e.g., the effective gain and time constants derived from the DMFT). Without this mapping or a quantitative comparison of bifurcation diagrams, it is unclear whether the reduction preserves the two-instability-mechanism picture.
minor comments (2)
  1. [Abstract and §1] The abstract and introduction use “chaos onset from the random connectivity” without citing the precise threshold condition (e.g., the critical variance of the random weights) that is later derived.
  2. [Figure captions] Figure captions should explicitly state which panels show full-network simulations versus the reduced model and which quantities are averaged over realizations.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for their detailed and insightful report. Below we respond point-by-point to the major comments, indicating the revisions we will make to the manuscript.

read point-by-point responses

  1. Referee: [Derivation of the DMFT equations (likely §3)] The central claim that the frequency-dependent transfer function of an isolated neuron governs both the chaos threshold and the Hopf bifurcation rests on the assumption that low-rank deterministic correlations remain negligible in the self-consistent equations for mean and variance once random connectivity exceeds the chaos threshold. The manuscript must derive the additional cross-terms arising from the low-rank component plus the temporal filtering of adaptation and show (analytically or numerically) that they do not alter the effective transfer function used to locate the instabilities.

    Authors: We thank the referee for highlighting this aspect of the derivation. In our DMFT approach, the low-rank structure is treated as a deterministic perturbation whose contributions to the self-consistent equations for the mean and variance are indeed subleading above the chaos threshold set by the random connectivity. However, to make this explicit, we will include in the revised manuscript a detailed derivation of the cross-terms involving the low-rank component and the adaptation filter. We will show both analytically and via numerical verification that these terms do not modify the effective frequency-dependent transfer function used to determine the instabilities. This addition will clarify the separation of scales. revision: yes

  2. Referee: [Reduced three-dimensional model (likely §4 or §5)] The reduced three-dimensional model is stated to capture the bifurcation structure, but no explicit mapping is given between its parameters and the full-network statistics (e.g., the effective gain and time constants derived from the DMFT). Without this mapping or a quantitative comparison of bifurcation diagrams, it is unclear whether the reduction preserves the two-instability-mechanism picture.

    Authors: We agree that providing an explicit mapping would improve the clarity of the reduction. In the revised manuscript, we will add a section detailing how the parameters of the three-dimensional model (effective gain, time constants, etc.) are derived from the DMFT statistics of the full network. Additionally, we will include a quantitative comparison of the bifurcation diagrams obtained from the reduced model and from direct simulations of the network, confirming that the two-instability-mechanism picture is preserved. revision: yes

Circularity Check

0 steps flagged

No circularity: mean-field derivation is self-contained

full rationale

The paper derives a dynamical mean-field theory for low-rank recurrent networks with adaptation, identifies chaos onset and Hopf bifurcation via the frequency-dependent transfer function, and reduces to a 3D model. No steps reduce by construction to inputs, fitted parameters renamed as predictions, or load-bearing self-citations. The central claims rest on explicit derivation of the mean-field equations and bifurcation analysis rather than self-referential definitions or ansatzes smuggled via prior work. This is the expected non-finding for a derivation-focused manuscript.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only text supplies no explicit free parameters, axioms, or invented entities; the mean-field closure and the form of the adaptation variable are implicit modeling choices whose details are unavailable.

pith-pipeline@v0.9.1-grok · 5713 in / 1083 out tokens · 55160 ms · 2026-06-30T03:07:24.515601+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references

  1. [1]

    These are jointly Gaussian with zero mean and covariances Cov(n, m) =γσ mσn =λ eff, Cov(n, h) = λeffκ, Cov(m, h) =σ 2 mκ

    Gaussian conditioning The expectation in (D5) is over the joint distribution of (n, m, h). These are jointly Gaussian with zero mean and covariances Cov(n, m) =γσ mσn =λ eff, Cov(n, h) = λeffκ, Cov(m, h) =σ 2 mκ. Since ˜cand ˆGdepend onh alone, we condition onhusing the Gaussian regression formula: n|h= λeffκ Σ2 h h+ ˜n,(D6) m|h= σ2 mκ Σ2 h h+ ˜m,(D7) whe...

  2. [2]

    Buzs´ aki and A

    G. Buzs´ aki and A. Draguhn, Neuronal oscillations in cor- tical networks, Science304, 1926 (2004)

    1926

  3. [3]

    Fries, A mechanism for cognitive dynamics: neuronal communication through neuronal coherence, Trends Cogn

    P. Fries, A mechanism for cognitive dynamics: neuronal communication through neuronal coherence, Trends Cogn. Sci.9, 474 (2005)

    2005

  4. [4]

    Wang and G

    X.-J. Wang and G. Buzs´ aki, Gamma oscillation by synap- tic inhibition in a hippocampal interneuronal network model, J. Neurosci.16, 6402 (1996)

    1996

  5. [5]

    Wang, Neurophysiological and computational prin- ciples of cortical rhythms in cognition, Physiol

    X.-J. Wang, Neurophysiological and computational prin- ciples of cortical rhythms in cognition, Physiol. Rev.90, 1195 (2010)

    2010

  6. [6]

    H. R. Wilson and J. D. Cowan, Excitatory and inhibitory interactions in localized populations of model neurons, Biophys. J.12, 1 (1972). 21

    1972

  7. [7]

    Brunel and X.-J

    N. Brunel and X.-J. Wang, What determines the fre- quency of fast network oscillations with irregular neural discharges?, J. Neurophysiol.90, 415 (2003)

    2003

  8. [8]

    Destexhe, A

    A. Destexhe, A. Babloyantz, and T. J. Sejnowski, Ionic mechanisms for intrinsic slow oscillations in thalamic re- lay neurons, Biophys. J.65, 1538 (1993)

    1993

  9. [9]

    Compte, M

    A. Compte, M. V. Sanchez-Vives, D. A. McCormick, and X.-J. Wang, Cellular and network mechanisms of slow oscillatory activity (<1 Hz) and wave propagations in a cortical network model, J. Neurophysiol.89, 2707 (2003)

    2003

  10. [10]

    Sompolinsky, A

    H. Sompolinsky, A. Crisanti, and H. J. Sommers, Chaos in random neural networks, Phys. Rev. Lett.61, 259 (1988)

    1988

  11. [11]

    van Vreeswijk and H

    C. van Vreeswijk and H. Sompolinsky, Chaos in neuronal networks with balanced excitatory and inhibitory activ- ity, Science274, 1724 (1996)

    1996

  12. [12]

    Crisanti and H

    A. Crisanti and H. Sompolinsky, Path integral approach to random neural networks, Phys. Rev. E98, 062120 (2018)

    2018

  13. [13]

    Brunel, Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons, J

    N. Brunel, Dynamics of sparsely connected networks of excitatory and inhibitory spiking neurons, J. Comput. Neurosci.8, 183 (2000)

    2000

  14. [14]

    Mastrogiuseppe and S

    F. Mastrogiuseppe and S. Ostojic, Linking connectivity, dynamics, and computations in low-rank recurrent neural networks, Neuron99, 609 (2018)

    2018

  15. [15]

    Kadmon and H

    J. Kadmon and H. Sompolinsky, Transition to chaos in random neuronal networks, Phys. Rev. X5, 041030 (2015)

    2015

  16. [16]

    Schuessler, A

    F. Schuessler, A. Dubreuil, F. Mastrogiuseppe, S. Os- tojic, and O. Barak, Dynamics of random recurrent net- works with correlated low-rank structure, Phys. Rev. Re- search2, 013111 (2020)

    2020

  17. [17]

    I. D. Landau and H. Sompolinsky, Coherent chaos in a recurrent neural network with structured connectivity, PLoS Comput. Biol.14, e1006309 (2018)

    2018

  18. [18]

    I. D. Landau and H. Sompolinsky, Macroscopic fluctua- tions emerge in balanced networks with incomplete recur- rent alignment, Phys. Rev. Research3, 023171 (2021)

    2021

  19. [19]

    Aljadeff, M

    J. Aljadeff, M. Stern, and T. Sharpee, Transition to chaos in random networks with cell-type-specific connectivity, Physical Review Letters114, 088101 (2015)

    2015

  20. [20]

    J. P. Cunningham and B. M. Yu, Dimensionality reduc- tion for large-scale neural recordings, Nat. Neurosci.17, 1500 (2014)

    2014

  21. [21]

    J. A. Gallego, M. G. Perich, L. E. Miller, and S. A. Solla, Neural manifolds for the control of movement, Neuron 94, 978 (2017)

    2017

  22. [22]

    Beiran, A

    M. Beiran, A. Dubreuil, A. Valente, F. Mastrogiuseppe, and S. Ostojic, Shaping dynamics with multiple popu- lations in low-rank recurrent networks, Neural Comput. 33, 1572 (2021)

    2021

  23. [23]

    Dubreuil, A

    A. Dubreuil, A. Valente, M. Beiran, F. Mastrogiuseppe, and S. Ostojic, The role of population structure in com- putations through neural dynamics, Nat. Neurosci.25, 783 (2022)

    2022

  24. [24]

    Sussillo and L

    D. Sussillo and L. F. Abbott, Generating coherent pat- terns of activity from chaotic neural networks, Neuron 63, 544 (2009)

    2009

  25. [25]

    Ostojic, Two types of asynchronous activity in net- works of excitatory and inhibitory spiking neurons, Na- ture Neuroscience17, 594 (2014)

    S. Ostojic, Two types of asynchronous activity in net- works of excitatory and inhibitory spiking neurons, Na- ture Neuroscience17, 594 (2014)

    2014

  26. [26]

    Rivkind and O

    A. Rivkind and O. Barak, Local dynamics in trained re- current neural networks, Phys. Rev. Lett.118, 258101 (2017)

    2017

  27. [27]

    D. G. Clark, L. F. Abbott, and A. Litwin-Kumar, Dimen- sion of activity in random neural networks, Phys. Rev. Lett.131, 118401 (2023)

    2023

  28. [28]

    Benda and A

    J. Benda and A. V. M. Herz, A universal model for spike- frequency adaptation, Neural Comput.15, 2523 (2003)

    2003

  29. [29]

    Fuhrmann, H

    G. Fuhrmann, H. Markram, and M. Tsodyks, Spike fre- quency adaptation and neocortical rhythms, J. Neuro- physiol.88, 761 (2002)

    2002

  30. [30]

    Ermentrout, M

    B. Ermentrout, M. Pascal, and B. Gutkin, The effects of spike frequency adaptation and negative feedback on the synchronization of neural oscillators, Neural Comput. 13, 1285 (2001)

    2001

  31. [31]

    van Vreeswijk and D

    C. van Vreeswijk and D. Hansel, Patterns of synchrony in neural networks with spike adaptation, Neural Comput. 13, 959 (2001)

    2001

  32. [32]

    S. P. Muscinelli, W. Gerstner, and T. Schwalger, How sin- gle neuron properties shape chaotic dynamics and signal transmission in random neural networks, PLoS Comput. Biol.15, e1007122 (2019)

    2019

  33. [33]

    D. G. Clark and L. F. Abbott, Theory of coupled neuronal-synaptic dynamics, Physical Review X14, 021001 (2024)

    2024

  34. [34]

    Gollisch and A

    T. Gollisch and A. V. M. Herz, Input-driven components of spike-frequency adaptation can be unmasked in vivo, J. Neurosci.24, 7435 (2004)

    2004

  35. [35]

    D. A. McCormick, Actions of acetylcholine in the cere- bral cortex and thalamus and implications for function, Progress in brain research98, 303 (1993)

    1993

  36. [36]

    K. M. Stiefel, B. S. Gutkin, and T. J. Sejnowski, Cholin- ergic neuromodulation changes phase response curve shape and type in cortical pyramidal neurons, PloS one 3, e3947 (2008)

    2008

  37. [37]

    T.-A. E. Nghiem, N. Tort-Colet, T. G´ orski, U. Ferrari, S. Moghimyfiroozabad, J. S. Goldman, B. Tele´ nczuk, C. Capone, T. Bal, M. Di Volo,et al., Cholinergic switch between two types of slow waves in cerebral cortex, Cere- bral Cortex30, 3451 (2020)

    2020

  38. [38]

    Rajan and L

    K. Rajan and L. F. Abbott, Eigenvalue spectra of random matrices for neural networks, Phys. Rev. Lett.97, 188104 (2006)

    2006