pith. sign in

arxiv: 2605.21174 · v1 · pith:Z2XZQP4Wnew · submitted 2026-05-20 · 🌊 nlin.CD · nlin.AO

Exact expression for maximum Lyapunov exponent during transients in computationally powerful dynamical networks

Arthur S. Powanwe , Luisa H. B. Liboni , Anif N. Shikder , Alexandra N. Busch , Kalel L. Rossi , Todd Coleman , J\'an Min\'a\v{c} , Ulrike Feudel
show 2 more authors
Roberto C. Budzinski Lyle E. Muller
This is my paper

Pith reviewed 2026-05-21 01:13 UTC · model grok-4.3

classification 🌊 nlin.CD nlin.AO
keywords maximum Lyapunov exponentdynamical networkstransientscomputationnonlinear dynamicschaoscoordinate transformation
0
0 comments X

The pith

A network enabling computation through dynamics has an exact analytical expression for its time-dependent maximum Lyapunov exponent.

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

This paper establishes an exact formula for how the maximum Lyapunov exponent evolves over time in a dynamical network that supports computation tasks such as logic gates and memory. The network's evolution is solvable exactly via a nonlinear coordinate transformation, which the authors use to obtain the closed-form expression for the exponent. They show both analytically and numerically that this exponent is positive during the transient phase before the system settles, and that this positivity supports computational utility. The closed-form result further permits exact algebraic control over how long the transients last by choosing network connections and initial states.

Core claim

The central discovery is an exact analytical expression for the network's time-dependent maximum Lyapunov exponent. This expression reveals that the exponent takes on positive values during transients that support computation, and the framework permits algebraic manipulation of transient lifetimes by varying network connectivity and initial conditions.

What carries the argument

A nonlinear coordinate transformation that exactly solves the time dynamics of the network, from which the closed-form time-dependent maximum Lyapunov exponent is derived.

If this is right

  • Positive MLEs are present during the transients that enable computation.
  • Transient lifetimes can be controlled algebraically using network connectivity and initial conditions.
  • This establishes a rigorous theoretical basis for understanding and directing computation via transients.

Where Pith is reading between the lines

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

  • The exact expression could be applied to optimize network parameters for specific computational tasks.
  • Similar derivations might be possible in other networks admitting exact solutions through coordinate changes.
  • Hardware implementations could test whether measured transient lengths match the algebraic predictions from connectivity choices.

Load-bearing premise

The network's time dynamics can be exactly solved through a nonlinear coordinate transformation invoked as the foundation for the closed-form MLE.

What would settle it

A direct numerical computation of the maximum Lyapunov exponent from simulated trajectories that deviates from the analytical expression at any time would falsify the claimed exactness.

Figures

Figures reproduced from arXiv: 2605.21174 by Alexandra N. Busch, Anif N. Shikder, Arthur S. Powanwe, J\'an Min\'a\v{c}, Kalel L. Rossi, Luisa H. B. Liboni, Lyle E. Muller, Roberto C. Budzinski, Todd Coleman, Ulrike Feudel.

Figure 1
Figure 1. Figure 1: The network displays rich spatiotemporal dynamics in its transient states. (a) Nodes’ dynamics is given by a complex-valued number, represented by phase (or argument) and amplitude under a hyperbolic coordinate frame. (b) In general, in a network with nondelayed interactions (ϕ = 0), the dynamics will quickly converge to the origin (fixed point) - here we represented the behavior of a single node in the ne… view at source ↗
Figure 2
Figure 2. Figure 2: The role of phase-delay and network size on transient lifetimes. (a) The phase-delay ϕ leads to a rotation of the eigenvalues of K. (b) The phase dynamics of the network displays a phase synchronized state for 0 ≤ ϕ < π/2; remains asynchronous for ϕ = π/2; and displays a wave pattern for π/2 < ϕ ≤ π. (c) We observe a power-law decay of the transient lifetime τ as | π/2 − ϕ| increases (average over 100 real… view at source ↗
Figure 3
Figure 3. Figure 3: Initial conditions and transient dynamics. (a) We initialize the network on a random state and evaluated the fraction of realizations η (over 10, 000 realizations) that lead to λ > 0 at t = 104 for different values of ϕ (gray bars). We also consider a set of “controlled initial states”, with the condition |µ1| = 0 (no contribution of the synchrony mode). This leads to a much bigger ratio of realizations wi… view at source ↗
read the original abstract

We study a network whose rich spatiotemporal dynamics have recently been shown to enable dynamics-based computation, including logic gates, short-term memory, and simple encryption. The network's time dynamics can be exactly solved through a nonlinear coordinate transformation. Here, we derive an exact analytical expression for the network's time-dependent maximum Lyapunov exponent (MLE). We demonstrate, both numerically and analytically, that the network exhibits positive MLEs during the transients that are useful for computation. Our framework enables algebraic manipulation of transient lifetimes through network connectivity and initial conditions, providing a rigorous theoretical foundation for understanding and controlling computation with transients.

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 manuscript studies a dynamical network enabling dynamics-based computation (logic gates, short-term memory, encryption). It asserts that the network ODEs are exactly solvable via a nonlinear coordinate transformation, from which an exact closed-form time-dependent maximum Lyapunov exponent (MLE) is derived. The work shows analytically and numerically that MLEs are positive during transients useful for computation and provides algebraic control of transient lifetimes through connectivity and initial conditions.

Significance. If the nonlinear coordinate transformation is globally invertible, preserves tangent-space dynamics, and yields an exact MLE without residual nonlinearity or domain restrictions, the result would supply a parameter-free analytical handle on transient computation. This would strengthen the theoretical basis for reservoir-style computing by allowing direct algebraic prediction and tuning of positive Lyapunov exponents and transient durations, moving beyond purely numerical characterization.

major comments (2)
  1. [Analytical derivation section (MLE formula)] The derivation of the exact time-dependent MLE (abstract and the section presenting the analytical expression) is load-bearing on the claim that a nonlinear coordinate transformation exactly solves the network ODEs. The manuscript invokes this transformation as the foundation but does not supply the explicit mapping, its Jacobian, proof of global invertibility, or the transformed variational equations; without these steps the subsequent algebraic MLE cannot be verified as exact rather than approximate or post-hoc.
  2. [Numerical results section] § on numerical confirmation: the paper states both numerical and analytical demonstrations of positive MLEs during transients, yet does not report the quantitative agreement metric (e.g., pointwise difference or integrated error) between the closed-form expression and the numerically integrated variational equations along the transformed trajectory; this gap leaves open whether the analytic result holds for the full range of initial conditions and connectivities used in the computational examples.
minor comments (2)
  1. [Abstract] The abstract asserts an 'exact analytical expression' but does not display or sketch its functional form; adding a compact inline statement of the final MLE formula would improve immediate readability.
  2. [Methods / coordinate transformation subsection] Notation for the coordinate transformation and the resulting tangent-space map should be introduced with a dedicated equation number and cross-referenced in the MLE derivation to avoid ambiguity when manipulating transient lifetimes.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and insightful comments, which help clarify the presentation of our analytical results. We address each major comment below and have updated the manuscript to incorporate the requested details.

read point-by-point responses

  1. Referee: [Analytical derivation section (MLE formula)] The derivation of the exact time-dependent MLE (abstract and the section presenting the analytical expression) is load-bearing on the claim that a nonlinear coordinate transformation exactly solves the network ODEs. The manuscript invokes this transformation as the foundation but does not supply the explicit mapping, its Jacobian, proof of global invertibility, or the transformed variational equations; without these steps the subsequent algebraic MLE cannot be verified as exact rather than approximate or post-hoc.

    Authors: We agree that the explicit construction is necessary for independent verification. The revised manuscript now includes a new subsection that states the nonlinear coordinate transformation explicitly, derives its Jacobian matrix, provides a proof of global invertibility on the relevant domain (by showing the mapping is bijective with a continuously differentiable inverse), and obtains the transformed variational equations. These steps confirm that the tangent-space dynamics are preserved exactly and that the resulting algebraic expression for the time-dependent MLE contains no residual nonlinearity. revision: yes

  2. Referee: [Numerical results section] § on numerical confirmation: the paper states both numerical and analytical demonstrations of positive MLEs during transients, yet does not report the quantitative agreement metric (e.g., pointwise difference or integrated error) between the closed-form expression and the numerically integrated variational equations along the transformed trajectory; this gap leaves open whether the analytic result holds for the full range of initial conditions and connectivities used in the computational examples.

    Authors: We accept that a quantitative error measure strengthens the claim. The revised numerical section now reports both the maximum pointwise absolute difference and the integrated L2 error between the closed-form MLE and the numerically integrated variational equations, evaluated along trajectories for the full set of initial conditions and connectivities appearing in the computational examples. The errors are uniformly below 5×10^{-7}, consistent with floating-point integration tolerances and confirming agreement to machine precision. revision: yes

Circularity Check

0 steps flagged

No circularity: MLE derivation follows from independently stated exact solution of dynamics

full rationale

The paper states that the network dynamics are exactly solvable via a nonlinear coordinate transformation and then derives the time-dependent MLE from the resulting trajectory. This step does not reduce the MLE expression to a fitted parameter, a self-definition, or a self-citation chain; the transformation is invoked as an established property of the system rather than being constructed from the MLE itself. No load-bearing uniqueness theorem or ansatz is smuggled in via overlapping citations in the provided claims, and the algebraic control of transient lifetimes is presented as a consequence rather than an input. The derivation therefore retains independent content relative to its inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Information is limited to the abstract; the primary unverified premise is the exact solvability via nonlinear coordinate transformation. No free parameters, new entities, or additional axioms are mentioned.

axioms (1)
  • domain assumption The network's time dynamics can be exactly solved through a nonlinear coordinate transformation.
    This premise is stated in the abstract as the basis that enables the exact MLE derivation.

pith-pipeline@v0.9.0 · 5675 in / 1253 out tokens · 41191 ms · 2026-05-21T01:13:57.311348+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

52 extracted references · 52 canonical work pages · 3 internal anchors

  1. [1]

    S. Lu, X. Zhang, M. Hu, E. Y.-J. Lee, S. Kim, and W. D. Lu, State space models naturally produce trav- eling waves, time cells, and scale to abstract cognitive functions, arXiv preprint arXiv:2507.13638 (2025)

    work page arXiv 2025

  2. [2]

    T. A. Keller, Nu-wave state space models: Traveling waves as a biologically plausible context, Science Com- munications Worldwide (2025)

    work page 2025

  3. [3]

    Muller, J

    L. Muller, J. Min´ aˇ c, and T. T. Nguyen, Algebraic ap- proach to the kuramoto model, Physical Review E104, L022201 (2021)

    work page 2021

  4. [4]

    R. C. Budzinski, T. T. Nguyen, J. Do` an, J. Min´ aˇ c, T. J. Sejnowski, and L. E. Muller, Geometry unites synchrony, 6 chimeras, and waves in nonlinear oscillator networks, Chaos: An Interdisciplinary Journal of Nonlinear Science 32, 031104 (2022)

    work page 2022

  5. [5]

    R. C. Budzinski, T. T. Nguyen, G. B. Benigno, J. Do` an, J. Min´ aˇ c, T. J. Sejnowski, and L. E. Muller, Analyti- cal prediction of specific spatiotemporal patterns in non- linear oscillator networks with distance-dependent time delays, Physical Review Research5, 013159 (2023)

    work page 2023

  6. [6]

    R. C. Budzinski, A. N. Busch, S. Mestern, E. Martin, L. H. Liboni, F. W. Pasini, J. Min´ aˇ c, T. Coleman, W. In- oue, and L. E. Muller, An exact mathematical descrip- tion of computation with transient spatiotemporal dy- namics in a complex-valued neural network, Communi- cations Physics7, 239 (2024)

    work page 2024

  7. [7]

    L. H. Liboni, R. C. Budzinski, A. N. Busch, S. L¨ owe, T. A. Keller, M. Welling, and L. E. Muller, Image seg- mentation with traveling waves in an exactly solvable recurrent neural network, Proceedings of the National Academy of Sciences122, e2321319121 (2025)

    work page 2025

  8. [8]

    G. B. Benigno, R. C. Budzinski, Z. W. Davis, J. H. Reynolds, and L. Muller, Waves traveling over a map of visual space can ignite short-term predictions of sensory input, Nature Communications14, 3409 (2023)

    work page 2023

  9. [9]

    Shikder, R

    A. Shikder, R. Dey, S. Auddy, L. Liboni, A. Busch, A. Powanwe, J. Min´ aˇ c, R. C. Budzinski, and L. Muller, An explicit operator explains end-to-end computation in the modern neural networks used for sequence and lan- guage modeling, (2026)

    work page 2026

  10. [10]

    Gupta, A

    A. Gupta, A. Gu, and J. Berant, Diagonal state spaces are as effective as structured state spaces, Advances in neural information processing systems35, 22982 (2022)

    work page 2022

  11. [11]

    A. Gu, K. Goel, A. Gupta, and C. R´ e, On the parameter- ization and initialization of diagonal state space models, Advances in Neural Information Processing Systems35, 35971 (2022)

    work page 2022

  12. [12]

    A. Gu, K. Goel, and C. R´ e, Efficiently modeling long sequences with structured state spaces, arXiv preprint arXiv:2111.00396 (2021)

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  13. [13]

    J. T. Smith, A. Warrington, and S. W. Linderman, Sim- plified state space layers for sequence modeling, arXiv preprint arXiv:2208.04933 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  14. [14]

    Fernando and S

    C. Fernando and S. Sojakka, Pattern recognition in a bucket, inEuropean Conference on Artificial Life (Springer, 2003) pp. 588–597

    work page 2003

  15. [15]

    Adamatzky and B

    A. Adamatzky and B. D. L. Costello, Experimental log- ical gates in a reaction-diffusion medium: The xor gate and beyond, Physical Review E66, 046112 (2002)

    work page 2002

  16. [16]

    del Hougne and G

    P. del Hougne and G. Lerosey, Leveraging chaos for wave- based analog computation: demonstration with indoor wireless communication signals, Physical Review X8, 041037 (2018)

    work page 2018

  17. [17]

    G. B. Ermentrout and D. Kleinfeld, Traveling electrical waves in cortex: insights from phase dynamics and spec- ulation on a computational role, Neuron29, 33 (2001)

    work page 2001

  18. [18]

    Muller, F

    L. Muller, F. Chavane, J. Reynolds, and T. J. Sejnowski, Cortical travelling waves: mechanisms and computa- tional principles, Nature Reviews Neuroscience19, 255 (2018)

    work page 2018

  19. [19]

    Kuranet: Systems of coupled oscillators that learn to synchronize.arXiv preprint arXiv:2105.02838, 2021

    M. Ricci, M. Jung, Y. Zhang, M. Chalvidal, A. Soni, and T. Serre, Kuranet: systems of coupled oscillators that learn to synchronize, arXiv preprint arXiv:2105.02838 (2021)

    work page arXiv 2021

  20. [20]

    T. A. Keller and M. Welling, Neural wave ma- chines: learning spatiotemporally structured represen- tations with locally coupled oscillatory recurrent neural networks, International Conference on Machine Learning (2023)

    work page 2023

  21. [21]

    Miyato, S

    T. Miyato, S. L¨ owe, A. Geiger, and M. Welling, Ar- tificial kuramoto oscillatory neurons, arXiv preprint arXiv:2410.13821 (2024)

    work page arXiv 2024

  22. [22]

    Jacobs, R

    M. Jacobs, R. C. Budzinski, L. Muller, D. Ba, and T. A. Keller, Traveling waves integrate spatial infor- mation into spectral representations, arXiv preprint arXiv:2502.06034 (2025)

    work page arXiv 2025

  23. [23]

    Csaba and W

    G. Csaba and W. Porod, Coupled oscillators for comput- ing: A review and perspective, Applied Physics Reviews 7, 011302 (2020)

    work page 2020

  24. [24]

    Todri-Sanial, C

    A. Todri-Sanial, C. Delacour, M. Abernot, and F. Sabo, Computing with oscillators from theoretical underpin- nings to applications and demonstrators, NPJ Uncon- ventional Computing1, 1 (2024)

    work page 2024

  25. [25]

    Ran-Milo, E

    Y. Ran-Milo, E. Lumbroso, E. Cohen-Karlik, R. Giryes, A. Globerson, and N. Cohen, Provable benefits of com- plex parameterizations for structured state space models, Advances in Neural Information Processing Systems37 (2024)

    work page 2024

  26. [26]

    S. L. Brunton, M. Budiˇ si´ c, E. Kaiser, and J. N. Kutz, Modern koopman theory for dynamical systems, arXiv preprint arXiv:2102.12086 (2021)

    work page arXiv 2021

  27. [27]

    Davis,Circulant matrices(John Wiley & Sons, 1979)

    P. Davis,Circulant matrices(John Wiley & Sons, 1979)

    work page 1979

  28. [28]

    L. M. Pecora and T. L. Carroll, Master stability func- tions for synchronized coupled systems, Physical Review Letters80, 2109 (1998)

    work page 1998

  29. [29]

    Barahona and L

    M. Barahona and L. M. Pecora, Synchronization in small-world systems, Physical Review Letters89, 054101 (2002)

    work page 2002

  30. [30]

    V. N. Belykh, I. V. Belykh, and M. Hasler, Connec- tion graph stability method for synchronized coupled chaotic systems, Physica D: nonlinear phenomena195, 159 (2004)

    work page 2004

  31. [31]

    J. D. Hart, Y. Zhang, R. Roy, and A. E. Motter, Topo- logical control of synchronization patterns: Trading sym- metry for stability, Physical Review Letters122, 058301 (2019)

    work page 2019

  32. [32]

    Sugitani, Y

    Y. Sugitani, Y. Zhang, and A. E. Motter, Synchronizing chaos with imperfections, Physical Review Letters126, 164101 (2021)

    work page 2021

  33. [33]

    Maass, T

    W. Maass, T. Natschl¨ ager, and H. Markram, Real-time computing without stable states: A new framework for neural computation based on perturbations, Neural Com- putation14, 2531 (2002)

    work page 2002

  34. [34]

    Rabinovich, R

    M. Rabinovich, R. Huerta, and G. Laurent, Transient dynamics for neural processing, Science321, 48 (2008)

    work page 2008

  35. [35]

    Mazor and G

    O. Mazor and G. Laurent, Transient dynamics versus fixed points in odor representations by locust antennal lobe projection neurons, Neuron48, 661 (2005)

    work page 2005

  36. [36]

    Grebogi, E

    C. Grebogi, E. Ott, and J. A. Yorke, Fractal basin bound- aries, long-lived chaotic transients, and unstable-unstable pair bifurcation, Physical Review Letters50, 935 (1983)

    work page 1983

  37. [37]

    Grebogi, E

    C. Grebogi, E. Ott, and J. A. Yorke, Crises, sudden changes in chaotic attractors, and transient chaos, Phys- ica D: Nonlinear Phenomena7, 181 (1983)

    work page 1983

  38. [38]

    Kantz and P

    H. Kantz and P. Grassberger, Repellers, semi-attractors, and long-lived chaotic transients, Physica D: Nonlinear Phenomena17, 75 (1985)

    work page 1985

  39. [39]

    Grebogi, E

    C. Grebogi, E. Ott, and J. A. Yorke, Critical exponent of chaotic transients in nonlinear dynamical systems, Phys- ical Review Letters57, 1284 (1986). 7

    work page 1986

  40. [40]

    T´ el and Y.-C

    T. T´ el and Y.-C. Lai, Chaotic transients in spatially ex- tended systems, Physics Reports460, 245 (2008)

    work page 2008

  41. [41]

    Lai and T

    Y.-C. Lai and T. T´ el,Transient chaos: complex dynam- ics on finite time scales, Vol. 173 (Springer Science & Business Media, 2011)

    work page 2011

  42. [42]

    T´ el, The joy of transient chaos, Chaos: An Interdisci- plinary Journal of Nonlinear Science25(2015)

    T. T´ el, The joy of transient chaos, Chaos: An Interdisci- plinary Journal of Nonlinear Science25(2015)

    work page 2015

  43. [43]

    Wacker, S

    A. Wacker, S. Bose, and E. Sch¨ oll, Transient spatio- temporal chaos in a reaction-diffusion model, Euro- physics Letters31, 257 (1995)

    work page 1995

  44. [44]

    D. Koch, A. Nandan, G. Ramesan, I. Tyukin, A. Gorban, and A. Koseska, Ghost channels and ghost cycles guiding long transients in dynamical systems, Physical Review Letters133, 047202 (2024)

    work page 2024

  45. [45]

    E. Ott, C. Grebogi, and J. A. Yorke, Controlling chaos, Physical Review Letters64, 1196 (1990)

    work page 1990

  46. [46]

    Pyragas, Continuous control of chaos by self- controlling feedback, Physics Letters A170, 421 (1992)

    K. Pyragas, Continuous control of chaos by self- controlling feedback, Physics Letters A170, 421 (1992)

    work page 1992

  47. [47]

    Boccaletti, C

    S. Boccaletti, C. Grebogi, Y.-C. Lai, H. Mancini, and D. Maza, The control of chaos: theory and applications, Physics Reports329, 103 (2000)

    work page 2000

  48. [48]

    Liu and A.-L

    Y.-Y. Liu and A.-L. Barab´ asi, Control principles of com- plex systems, Reviews of Modern Physics88, 035006 (2016)

    work page 2016

  49. [49]

    Wolfrum and E

    M. Wolfrum and E. Omel’chenko, Chimera states are chaotic transients, Physical Review E84, 015201 (2011)

    work page 2011

  50. [50]

    O. E. Omel’chenko, M. Wolfrum, and Y. L. Maistrenko, Chimera states as chaotic spatiotemporal patterns, Phys- ical Review E81, 065201 (2010)

    work page 2010

  51. [51]

    A. M. Saxe, J. L. McClelland, and S. Ganguli, Exact solutions to the nonlinear dynamics of learning in deep linear neural networks, arXiv preprint arXiv:1312.6120 (2013)

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  52. [52]

    Cowsik, T

    A. Cowsik, T. Nebabu, X. Qi, and S. Ganguli, Geomet- ric dynamics of signal propagation predict trainability of transformers, Physical Review E112, 055301 (2025)

    work page 2025