Pith. sign in

REVIEW 2 major objections 5 minor 53 references

Square-root price impact is necessary for self-sustained manipulation cycles in learning-agent markets; linear impact kills the Hopf bifurcation and leaves retail markets stable.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-11 08:09 UTC pith:7ZXDJYGZ

load-bearing objection Solid computational result: evolutionary agents find multi-cycle predation, and square-root impact is necessary for the Hopf; the regularization of the singular derivative is the only real soft spot. the 2 major comments →

arxiv 2607.05141 v1 pith:7ZXDJYGZ submitted 2026-07-06 q-fin.CP econ.THnlin.AOq-fin.TR

Square-Root Price Impact Is Necessary for Endogenous Manipulation Cycles in Learning-Agent Markets

Yang Zhou , Jianwen Chen , Ruipeng Wei This is my paper
classification q-fin.CP econ.THnlin.AOq-fin.TR
keywords price impactHopf bifurcationagent-based marketpredatory tradingmean-field reductionsquare-root lawlimit cycleevolutionary optimization
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

A single learning institutional trader, optimized by evolutionary search among 20,000 herding retail agents, spontaneously invents a multi-cycle predatory strategy that returns roughly +38% on average over 2000 days. Mean-field reduction turns the market into a nonlinear oscillator that undergoes a continuous Hopf bifurcation once institutional capital crosses a threshold, plus a discontinuous fold transition in herding strength. The limit cycle survives even when retail herding is switched off entirely: position-tracking feedback together with square-root price impact alone is enough to keep the oscillator self-sustained. Linear impact removes the Hopf transition completely and leaves the retail market unconditionally stable. The paper therefore claims that the familiar square-root impact law is not just an empirical regularity but a structural prerequisite for endogenous manipulation cycles, which appear as the optimal-control solution of that nonlinear dynamical system. A structural Maxwell-demon analogy casts the agent as an information-processing controller that measurably lowers the entropy rate of the price process while extracting profit.

Core claim

Manipulation cycles are endogenous limit cycles of a nonlinear dynamical system whose Hopf bifurcation requires square-root price impact. Position-tracking feedback coupled with that impact produces a self-sustained oscillator even at zero retail herding; linear impact eliminates the bifurcation and renders the retail market unconditionally stable. The cycles are therefore the optimal-control solution of the market-impact structure, not an artifact of herding rules or network architecture.

What carries the argument

The mean-field reduction to a two-dimensional autonomous system (price deviation x, institutional holding fraction q) whose Jacobian trace supplies an analytical Hopf boundary Cc(λ). The singular derivative of square-root impact near zero demand, regularized at a noise floor, supplies the finite effective gain that enables the bifurcation; linear impact lacks that singularity and cannot cross the stability threshold.

Load-bearing premise

The paper regularizes the infinite derivative of square-root impact near zero demand by evaluating the gain at the rms fluctuation level or a noise floor; if that regularization is invalid or small-flow impact is not square-root, both the stability analysis and the necessity claim change character.

What would settle it

Replace square-root impact with linear impact (or any non-singular small-flow form) inside the same mean-field ODE or agent-based market and check whether a continuous Hopf bifurcation still appears as institutional capital is increased; if it does, the necessity claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Markets whose empirical impact remains square-root at institutional sizes are structurally capable of supporting self-sustained predatory limit cycles once capital exceeds a calculable threshold.
  • Retail herding is neither necessary nor sufficient for the cycles; the oscillator is driven by institutional position feedback plus nonlinear impact alone.
  • Simple linear-impact models of market impact will systematically miss the possibility of endogenous manipulation cycles.
  • Architecture-independent controllers (LSTM or shallow MLP) discover the same four-phase bang-bang pattern, so the strategy is a property of the market microstructure rather than of recurrent memory.

Where Pith is reading between the lines

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

  • If the Hopf threshold scales with free float and impact coefficient as the analytical boundary suggests, regulators could in principle monitor institutional capital relative to that threshold as an early-warning indicator of cycle risk.
  • The same position-tracking-plus-square-root-impact oscillator may appear in other adaptive-agent settings (inventory control, market-making) that never involve retail herding.
  • Empirical tests that measure the small-flow derivative of impact (rather than the large-flow square-root regime) would directly probe whether real markets sit above or below the paper’s bifurcation boundary.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper studies a minimal agent-based market with one CMA-ES-optimized institutional controller (LSTM or MLP) interacting with 20,000 herding retail traders under square-root price impact. The agent discovers a multi-cycle predatory strategy (8–11 cycles over 2000 days; best return +51%, mean +37.7%). Mean-field reduction yields a 2D nonlinear oscillator that undergoes a continuous Hopf bifurcation in institutional capital C (amplitude A ∝ (C−Cc)^α with α≈0.48) and a discontinuous fold transition in herding-scale space. The cycle persists at β=0, driven by position-tracking feedback plus square-root impact; linear impact is claimed to eliminate the Hopf entirely. A Maxwell’s-demon analogy quantifies entropy-rate reduction of the price process.

Significance. If the necessity claim holds, the work supplies a clean dynamical-systems explanation for endogenous manipulation cycles: they are the optimal-control solution of a nonlinear oscillator whose restoring force is the singular derivative of square-root impact. Strengths include multi-seed ABM evidence, architecture independence (LSTM vs 162-parameter MLP), mechanism ablations (Table I), baseline comparisons (Table S3), and an analytical Hopf boundary (Eq. 7) whose numerical amplitude scaling is consistent with the textbook α=1/2. The β=0 self-sustained oscillator and the linear-impact contrast are falsifiable predictions that would matter for both market-microstructure theory and regulatory design.

major comments (2)
  1. §III.B–C and Appendix S3 (Eqs. 5–7, S7–S10): the Hopf boundary Cc and the necessity claim for square-root impact rest on regularizing the singular derivative ∂I/∂r̄ → ∞ by evaluating at rms fluctuation √v0 or noise floor ε (κε = λ/√(ε V0)). The ABM itself applies square-root impact at every scale and therefore cannot independently validate that regularization. A controlled small-flow diagnostic (or an explicit comparison against the linear-to-square-root crossover of Bucci et al. that the paper cites) is needed before the claim that “linear impact eliminates the Hopf bifurcation entirely” can be regarded as settled.
  2. §III.C (Eq. 6–7) and the capital-sweep discussion: the analytical Cc is derived under a smooth tanh feedback surrogate whose gains gq, gx and target qt are extracted from the trained controller. The paper reports a ~20 % gap between analytical and numerical Cc and notes that CMA-ES still finds profitable strategies below Cc. The manuscript should clarify more sharply that Cc marks the onset of sustained limit cycles, not of profitability, and should quantify how sensitive the boundary is to the particular extraction of the feedback gains.
minor comments (5)
  1. Fig. 1(a): the SDE–ABM correlation r = 0.62 is only moderate; a short discussion of residual discrepancy (gradual unwind, discrete price limits) would help the reader gauge the mean-field fidelity.
  2. Table S1 / main-text cycle-period comparison: the ODE under-predicts the distribution phase (57 vs 130 days). The text attributes this to gradual unwinding; a one-sentence quantification of that effect would strengthen the comparison.
  3. Maxwell’s-demon section and Table III: the structural analogy is useful, but the Sagawa–Ueda-style bound is presented as a “motivated consistency check.” Softening the language in the abstract and conclusion to match that caveat would avoid over-claiming thermodynamic equivalence.
  4. Notation: κε, CSR, HS and geff appear in several places with slightly different regularizations (half-gain vs full); a single consistent definition early in §III.B would improve readability.
  5. Data-availability statement promises code and controllers “upon publication”; for a computational paper this is acceptable, but an anonymized repository link at review stage would aid reproducibility.

Circularity Check

0 steps flagged

No significant circularity: Hopf analysis, square-root necessity, and α≈1/2 are properties of a reduced dynamical system checked against external theory and ABM, not inputs renamed as predictions.

full rationale

The derivation chain is self-contained and non-circular. The ABM (CMA-ES + LSTM/MLP) discovers multi-cycle strategies by optimizing terminal return; that is an empirical optimization result, not a definition of the later claims. Mean-field reduction builds an autonomous ODE whose coefficients come from market parameters, ABM statistics, or controller gains extracted from the trained policy—standard model reduction, not a self-definitional loop. The continuous Hopf boundary (tr J = 0 → analytical C_c), the order-parameter scaling A ∝ (C−C_c)^α with α≈0.48 compared to the textbook Hopf value 1/2 (Strogatz), the β=0 self-sustained oscillator, and the linear-impact contrast (no Hopf; retail market stable under square-root, unstable under linear) are mathematical properties of that ODE under two different impact laws. Fitting α on the ODE bifurcation diagram and comparing to the external critical-exponent prediction is ordinary consistency checking, not a fitted input called a prediction of the same quantity. Phase-duration and SDE–ABM comparisons (r=0.62, Table II/S1) are independent validation of the reduced model against the simulator. The Sagawa–Ueda-style bound is explicitly a loose consistency check with a 3–10× margin, not a load-bearing derivation. Maxwell’s-demon framing is structural analogy only. No self-citation uniqueness theorem, no ansatz smuggled from the authors’ prior work, and no renaming of a known empirical pattern as a first-principles result. Concerns about regularization of the singular square-root derivative (√v_0 or ε) affect correctness risk of the linearization, not circularity of the argument.

Axiom & Free-Parameter Ledger

8 free parameters · 6 axioms · 1 invented entities

The central claims rest on a standard square-root impact law, a reduced-form herding demand, a mean-field reduction that regularizes a singular derivative, and a large set of market and controller parameters that are either taken from the ABM configuration or fitted to ABM statistics. No new physical entities are postulated; the Maxwell-demon framing is an analogy. The free-parameter count is high, as is typical for calibrated agent-based markets, and the regularization step is the most paper-specific modeling choice.

free parameters (8)
  • price-impact coefficient λ = 0.008
    Set to 0.008 at the calibrated operating point; enters both the ABM and the analytical Hopf boundary Cc(λ).
  • herding strength β = 6.0
    Default β=6; swept in {2,4,6,8,12} but the functional form of retail demand is fixed.
  • herding scale HS = 10^{-3}
    Calibrated to 10^{-3} so that effective retail demand is ~140× smaller than institutional capacity; controls the fold transition.
  • mean-reversion rate μ = 0.01
    Set to 0.01; appears in the damping Γ and the Hopf boundary.
  • institutional capital fraction C = 25%
    Calibrated operating point C=25%; Hopf threshold Cc is measured relative to this scale.
  • noise-floor / regularization ε (or √v0) = ε ∈ [3e4, 3e5] shares
    Used to regularize the singular derivative of square-root impact at D=0; physical range estimated from daily demand variance; α and Cc are mildly sensitive to it.
  • retail amplitude AR and base volume V0 = AR≈2.3e6
    AR≈2.3e6 estimated from ABM statistics; V0 sets the impact scale; both enter CSR and the Hopf condition.
  • feedback gains gq, gx and target holding qt = extracted from controller
    Extracted from the trained LSTM to close the mean-field ODE; not free in the ABM but free relative to the pure theory.
axioms (6)
  • domain assumption Daily price impact follows the square-root law ΔP/P = λ sgn(D) √(|D|/V0).
    Taken from the empirical and theoretical market-microstructure literature (Almgren, Toth, Donier et al.); load-bearing for the necessity claim.
  • domain assumption Aggregate retail excess demand is DR = AR tanh(β r5d / 2) (plus stealth, exhaustion, and drawdown modifiers).
    Reduced-form herding rule standard in Lux–Marchesi-type ABMs; used throughout the ABM and mean-field.
  • ad hoc to paper The singular derivative of square-root impact may be regularized by evaluating at the rms return fluctuation √v0 (or noise floor ε), yielding a finite effective gain geff.
    Required to obtain a well-defined Jacobian and Hopf boundary (Eqs. 5–7, Appendix S3); the most paper-specific modeling step.
  • ad hoc to paper A smooth tanh feedback law is a sufficient surrogate for the LSTM’s near-bang-bang policy when computing the Hopf boundary.
    Invoked in Section III.C to replace Pontryagin bang-bang control by a differentiable controller whose Jacobian can be written in closed form.
  • domain assumption CMA-ES optimization of terminal portfolio return discovers a near-optimal control for the finite-horizon market.
    Standard black-box evolutionary assumption; supported by architecture and reward ablations but not proved.
  • domain assumption Mean-field reduction of 20 000 discrete agents plus one controller to a 2-D autonomous ODE preserves the bifurcation structure of interest.
    Validated by qualitative agreement (r=0.62, phase durations within factor ~2) but remains an approximation.
invented entities (1)
  • Institutional agent as Maxwell’s demon (information-processing controller that reduces price entropy rate) no independent evidence
    purpose: Structural analogy linking profit extraction to entropy-rate reduction and to Sagawa–Ueda-style information bounds.
    The analogy is explicitly labeled structural, not thermodynamic; entropy-rate reduction is measured but the demon itself is interpretive scaffolding rather than a new dynamical object.

pith-pipeline@v1.1.0-grok45 · 21567 in / 4115 out tokens · 49445 ms · 2026-07-11T08:09:39.902207+00:00 · methodology

0 comments
read the original abstract

We study a minimal agent-based market in which a single evolutionary-optimized institutional agent interacts with 20{,}000 herding retail traders. The agent spontaneously discovers a multi-cycle predatory strategy, producing 8--11 complete cycles over 2000 trading days with total portfolio return of $+51\%$ (best of 20 seeds; mean $+37.7\%$). Mean-field reduction maps the system onto a nonlinear oscillator that undergoes two distinct bifurcations: a continuous Hopf transition as institutional capital exceeds a critical threshold $C_c$, with oscillation amplitude $A \propto (C-C_c)^\alpha$ where $\alpha$ is consistent with the standard prediction of $1/2$; and a discontinuous fold transition in the herding-scale parameter space. The limit cycle persists even at $\beta = 0$: position-tracking feedback coupled with square-root price impact creates a self-sustained nonlinear oscillator requiring no retail herding. Square-root impact is shown to be necessary: linear impact eliminates the Hopf bifurcation entirely and renders the retail market unconditionally stable. Manipulation cycles thus emerge as the optimal-control solution of a nonlinear dynamical system, and a structural analogy to Maxwell's demon frames the agent as an information-processing controller that reduces the entropy rate of the price process.

Figures

Figures reproduced from arXiv: 2607.05141 by Jianwen Chen, Ruipeng Wei, Yang Zhou.

Figure 1
Figure 1. Figure 1: FIG. 1. Emergent manipulation cycle and mean-field the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Phase-space portrait ( [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Bifurcation analysis of the mean-field ODE. (a) Os [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Entropy rate analysis (20 seeds). The institution [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

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

53 extracted references · 1 canonical work pages

  1. [1]

    The agent acts as aMaxwell’s demon: it identifies the behavioral predictability of re- tail herding and extracts profit by reducing the informa- tion entropy of the price process

    LSTM agent among herding retail traders, calibrated ∗ zhouyang@westlake.edu.cn in the spirit of recent black-box estimation methods for agent-based models [22]. The agent acts as aMaxwell’s demon: it identifies the behavioral predictability of re- tail herding and extracts profit by reducing the informa- tion entropy of the price process. The closest meth...

    Pith/arXiv arXiv 2026

  2. [2]

    double square-root

    with 7 inputs, 8 hidden units, and 2 outputs (530 parameters), optimized by CMA-ES with terminal port- folio return as reward overT= 2000 trading days (full training and architecture details in Supplemental Mate- rial). The four-phase cycle is architecture-independent: an MLP controller with 162 parameters (only price and position as inputs) reproduces th...

    2000

  3. [3]

    ODE derivation The mean-field ODE (the price equation of the main text) is derived by replacing the stochastic price equation with its deterministic skeleton. Writingx= (P−P 0)/P0 (price deviation) andq(institutional holding fraction), the 2D autonomous system is: ˙x=I(D net) +I wash +µ P0 −P P ,(S1) ˙q= (bt −s t)Q max Nshares ,(S2) where the control (b t...

  4. [4]

    We lin- earize ˙xaroundx= 0 to obtain the effective damping

    Effective damping of the retail-only market For the retail-only system (u= 0, no institutional trad- ing), the market equilibrium isx ∗ = 0,q ∗ = 0. We lin- earize ˙xaroundx= 0 to obtain the effective damping. Near equilibrium, retail demand simplifies toD R ≈ AR tanh(β r5d/2), sinceS= 1,E= 1, Φ ov = 1, and δ= 0. The 5-day returnr 5d is an exponentially w...

  5. [5]

    observed cycle parameters TABLE S1

    Predicted vs. observed cycle parameters TABLE S1. Predicted (mean-field ODE) vs. observed cycle parameters. Parameter Predicted Observed Accumulation duration 28 days 26 days Cycle period 150–200 days 222 days Peak holdingq max 14.5% 14.6% κε (inst. impact) 0.004 0.004 CSR (herding) 0.011 0.011 Retail stability Γ>0∀βNo retail cycles From the ODE with net ...

  6. [6]

    The SDE uses the LSTM’s control input with analytically derived parameters and all ABM mechanisms (square-root impact, overvaluation dampening, drawdown selling, price limits)

    Theory–simulation comparison Figure 1 of the main text presents the theory– simulation comparison. The SDE uses the LSTM’s control input with analytically derived parameters and all ABM mechanisms (square-root impact, overvaluation dampening, drawdown selling, price limits). The SDE re- produces the cyclic oscillations with correlationr= 0.62 and captures...

  7. [7]

    molecular velocities

    Phase-space portrait The phase-space portrait in (q, x) appears as Fig. 2 of the main text. Each trading cycle traces a closed orbit that starts near the origin (low holdings, price near fun- damental), moves to highqand positivexduring accu- mulation/push, then returns as the institution distributes and price mean-reverts. Appendix S4: Maxwell’s Demon An...

  8. [8]

    Information-thermodynamic bound on profit The Maxwell’s demon analogy suggests a quantitative bound using the information-thermodynamic framework of Sagawa and Ueda [42, 43], who established that for a feedback-controlled system, the extractable work is bounded by the mutual information between the mea- surement and the outcome [44]. In our market, the an...

    2000

  9. [9]

    Cont, Quantitative Finance1, 223 (2001)

    R. Cont, Quantitative Finance1, 223 (2001)

    2001

  10. [10]

    Bouchaud, J

    J.-P. Bouchaud, J. Bonart, J. Donier, and M. Gould, Trades, Quotes and Prices: Financial Markets Under the Microscope(Cambridge University Press, 2018)

    2018

  11. [11]

    Sornette,Why Stock Markets Crash: Critical Events in Complex Financial Systems(Princeton University Press, 2009)

    D. Sornette,Why Stock Markets Crash: Critical Events in Complex Financial Systems(Princeton University Press, 2009)

    2009

  12. [12]

    Lux and M

    T. Lux and M. Marchesi, Nature397, 498 (1999)

    1999

  13. [13]

    V. M. Egu´ ıluz and M. G. Zimmermann, Physical Review Letters85, 5659 (2000)

    2000

  14. [14]

    Di Francesco and C

    T. Di Francesco and C. Hommes, Journal of Economic Dynamics and Control175, 105092 (2025)

    2025

  15. [15]

    Galanis, I

    G. Galanis, I. Kollias, I. Leventides, and J. Lusten- houwer, Journal of Economic Dynamics and Control177, 105125 (2025)

    2025

  16. [16]

    Cividino, R

    D. Cividino, R. Westphal, and D. Sornette, Physical Re- view Research5, 013009 (2023)

    2023

  17. [17]

    Malevergne, D

    Y. Malevergne, D. Sornette, and R. Wei, Quantitative Finance 10.1080/14697688.2025.2479633 (2025)

    doi:10.1080/14697688.2025.2479633 2025

  18. [18]

    W. Li, S. Shi, and Q. Zhao, Discrete and Continu- ous Dynamical Systems - Series S 10.3934/dcdss.2025014 (2025)

    work page doi:10.3934/dcdss.2025014 2025

  19. [19]

    Ninomiya, Journal of Economic Structures11, 1 (2022)

    K. Ninomiya, Journal of Economic Structures11, 1 (2022)

    2022

  20. [20]

    Yang, X.-Y

    H. Yang, X.-Y. Liu, Q. Zhong, and A. Walid, arXiv preprint arXiv:2011.09507 (2020)

    Pith/arXiv arXiv 2011

  21. [21]

    C. Yu, A. Velu, E. Vinitsky, J. Gao, Y. Wang, A. Bayen, and Y. Wu, Advances in Neural Information Processing Systems35, 24611 (2022)

    2022

  22. [22]

    Arifovic, X.-Z

    J. Arifovic, X.-Z. He, and L. Wei, Journal of Economic Dynamics and Control139, 104438 (2022)

    2022

  23. [23]

    X. Zhou, S. Lin, and X.-Z. He, Journal of Economic Dy- namics and Control172, 104991 (2025)

    2025

  24. [24]

    Allen and D

    F. Allen and D. Gale, The Review of Financial Studies 5, 503 (1992)

    1992

  25. [25]

    Goldstein and A

    I. Goldstein and A. Guembel, The Review of Economic Studies75, 133 (2008)

    2008

  26. [26]

    R. K. Aggarwal and G. Wu, The Journal of Business79, 1915 (2006)

    1915

  27. [27]

    M. K. Brunnermeier and L. H. Pedersen, The Journal of Finance60, 1825 (2005)

    2005

  28. [28]

    Takayama, Journal of Economic Dynamics and Con- trol125, 104086 (2021)

    S. Takayama, Journal of Economic Dynamics and Con- trol125, 104086 (2021)

    2021

  29. [29]

    Hansen and A

    N. Hansen and A. Ostermeier, Evolutionary Computa- tion9, 159 (2001)

    2001

  30. [30]

    J. Dyer, P. Cannon, J. D. Farmer, and S. M. Schmon, Journal of Economic Dynamics and Control161, 104827 (2024)

    2024

  31. [31]

    X. Wang, C. Hoang, Y. Vorobeychik, and M. P. Wellman, Games12, 46 (2021)

    2021

  32. [32]

    Bikhchandani, D

    S. Bikhchandani, D. Hirshleifer, and I. Welch, Journal of Political Economy100, 992 (1992)

    1992

  33. [33]

    Gualdi, J.-P

    S. Gualdi, J.-P. Bouchaud, G. Cencetti, M. Tarzia, and F. Zamponi, Physical Review Letters114, 088701 (2015)

    2015

  34. [34]

    Hochreiter and J

    S. Hochreiter and J. Schmidhuber, Neural Computation 9, 1735 (1997)

    1997

  35. [35]

    Almgren, C

    R. Almgren, C. Thum, E. Hauptmann, and H. Li, Risk 18, 57 (2005)

    2005

  36. [36]

    T´ oth, Y

    B. T´ oth, Y. Lemp´ eri` ere, C. Deremble, J. de Lataillade, J. Kockelkoren, and J.-P. Bouchaud, Physical Review X 1, 021006 (2011)

    2011

  37. [37]

    A. S. Kyle, Econometrica53, 1315 (1985)

    1985

  38. [38]

    Donier, J

    J. Donier, J. Bonart, I. Mastromatteo, and J.-P. Bouchaud, Quantitative Finance15, 1109 (2015)

    2015

  39. [39]

    Maitrier, G

    G. Maitrier, G. Loeper, K. Kanazawa, and J.-P. Bouchaud, Quantitative Finance26, 491 (2026)

    2026

  40. [40]

    Bouchaud, J

    J.-P. Bouchaud, J. D. Farmer, and F. Lillo, arXiv preprint arXiv:0809.0822 (2008)

    Pith/arXiv arXiv 2008

  41. [41]

    S. H. Strogatz,Nonlinear Dynamics and Chaos: With Applications to Physics, Biology, Chemistry, and Engi- neering, 2nd ed. (CRC Press, 2018)

    2018

  42. [42]

    L. S. Pontryagin, International Journal of Control7, 297 (1968)

    1968

  43. [43]

    E. Moro, J. Vicente, L. G. Moyano, A. Gerig, J. D. Farmer, G. Vaglica, F. Lillo, and R. N. Mantegna, Phys- ical Review E80, 066102 (2009)

    2009

  44. [44]

    Bucci, M

    F. Bucci, M. Benzaquen, F. Lillo, and J.-P. Bouchaud, Physical Review Letters122, 108302 (2019)

    2019

  45. [45]

    Y. A. Kuznetsov,Elements of Applied Bifurcation The- ory, 3rd ed. (Springer, 2004)

    2004

  46. [46]

    H. S. Leff, inAIP Conference Proceedings, Vol. 643 (AIP,

  47. [47]

    E. F. Fama, The Journal of Finance25, 383 (1970)

    1970

  48. [48]

    C. E. Shannon, Bell System Technical Journal27, 379 (1948)

    1948

  49. [49]

    Lempel and J

    A. Lempel and J. Ziv, IEEE Transactions on Information Theory22, 75 (1976)

    1976

  50. [50]

    Sagawa and M

    T. Sagawa and M. Ueda, Physical Review Letters104, 090602 (2010)

    2010

  51. [51]

    Sagawa and M

    T. Sagawa and M. Ueda, Physical Review E85, 021104 (2012)

    2012

  52. [52]

    J. M. R. Parrondo, J. M. Horowitz, and T. Sagawa, Na- ture Physics11, 131 (2015)

    2015

  53. [53]

    Landauer, IBM Journal of Research and Development 5, 183 (1961)

    R. Landauer, IBM Journal of Research and Development 5, 183 (1961)

    1961