logo

📂Dynamical Systems

Aizawa Attractor

Model

$$ \begin{align*} \dot{x} =& (z - b)x - d y \\ \dot{y} =& d x + (z - b)y \\ \dot{z} =& c + a z - \frac{z^{3}}{3} - (x^{2} + y^{2})(1 + e z) + f z x^{3} \end{align*} $$

Parameters

  • $a = 0.95$
  • $b = 0.7$
  • $c = 3.5$
  • $d = 0.25$
  • $e = 0.1$
  • $f = 0.1$

Explanation

The Aizawa attractor is a dynamical system that includes terms of up to fourth order on the right-hand side and exhibits complex chaos; it also appears as a benchmark in fields such as time series forecasting.

However, while looking into the origin and history of this system, the author found that no paper by Aizawa himself exists, and that even in the literature where the Aizawa attractor does appear, the expression of the equations differs from source to source1 2 3 4 5. This is a rather strange situation for an attractor that has already gained fame and is being cited. In this post, the ordinary differential equations have been compiled and recorded in the form that seems to be the most common among them.

Attractor

alt text

Even among strange attractors, it looks quite strange. It basically takes the shape of a sphere, yet appears to have a structure that rotates about the $z$-axis while rising and falling; on closer inspection, one can see that it is a torus6.

Bifurcation

alt text

At least as of the time of writing this post, the author has never seen any research result that mentions the bifurcation related to the Aizawa attractor. The bifurcation diagram above records the maximum value of $z$ while varying $b$ from $0.6$ to $0.9$. As can be seen, the route to chaos that one would normally expect to see does not appear, and it has an extremely bizarre form. I do not have time to study the Aizawa attractor right now, but someday I would like to dig deeply into this system.

Code

The following is Julia code that can simulate the Aizawa attractor.

using DataFrames, DifferentialEquations, OrdinaryDiffEqLowOrderRK

function factory_aizawa(b::Number; ic = [0.1, 0.1, 0.1], saveat = 0:1e-2:100)
      a,  c,   d,   e,  f = (
    .95, .6, 3.5, .25, .1)
    function sys(du, u, p, t)
        x, y, z = u
        b = p[1]

        du[1] = (z - b)*x - d*y
        du[2] = d*x + (z - b)*y
        du[3] = c + a*z - (z^3)/3 - (x^2 + y^2)*(1 + e*z) + f*z*x^3
        return du
    end
    sol = solve(ODEProblem(sys, ic, (0, last(saveat)), [b]), RK4(), dt = saveat.step.hi, adaptive=false, maxiters = 1e+7)
    matrix = Matrix([sol.t'; sol[:, :]; stack([sys(zeros(3), u, [b], 0) for u in sol.u])]')
    return matrix[sol.t .≥ first(saveat), :][1:end-1, :]
end
factory_aizawa(T::Type, args...; kargs...) =
DataFrame(factory_aizawa(args...; kargs...), ["t", "x", "y", "z", "dx", "dy", "dz"])

sol = factory_aizawa(DataFrame, 0.7)
plot(sol.x, sol.y, sol.z, alpha = .5, color = :black, xlabel = L"x", ylabel = L"y", zlabel = L"z", size = [400, 400])

pm, pM = 0.6, 0.9; p0, p1 = 0.72, 0.74;
p_ = range(pm, pM, length = 1001)
bfcn = callbfcn("G:/BF/aizawa/bfcnA.jld2")
@showprogress @threads for k in eachindex(p_)
    sol = factory_aizawa(DataFrame, p_[k], saveat = 0:1e-3:3000)
    z_ = sol.z[sol.t .≥ 2000]
    bfcn[p_[k]] = z_[arglmax(z_)]
end

  1. Ghanbari, B., Gómez-Aguilar, J.F. Two efficient numerical schemes for simulating dynamical systems and capturing chaotic behaviors with Mittag–Leffler memory. Engineering with Computers 38, 2139–2167 (2022). https://doi.org/10.1007/s00366-020-01170-0 ↩︎

  2. Zhai, ZM., Stern, B.D. & Lai, YC. Bridging known and unknown dynamics by transformer-based machine-learning inference from sparse observations. Nat Commun 16, 8053 (2025). https://doi.org/10.1038/s41467-025-63019-8 ↩︎

  3. Zhang, Y., Li, W., & Carvalho, R. (2026). Fast and principled equation discovery from chaos to climate. arXiv preprint arXiv:2604.11929. https://arxiv.org/pdf/2604.11929 ↩︎

  4. https://www.algosome.com/articles/aizawa-attractor-chaos.html ↩︎

  5. https://sequelaencollection.home.blog/3d-chaotic-attractors/ ↩︎

  6. https://doi.org/10.1007/978-3-0348-6256-1_19 ↩︎