|
| 1 | +# # Eigenfunctions of the Koopman operator |
| 2 | + |
| 3 | +#md # [](@__BINDER_ROOT_URL__/generated/Systems and Control/koopman_eigenfunctions.ipynb) |
| 4 | +#md # [](@__NBVIEWER_ROOT_URL__/generated/Systems and Control/koopman_eigenfunctions.ipynb) |
| 5 | +# **Adapted from**: Example 2.6 of [MAI20] |
| 6 | +# |
| 7 | +# [MAI20] Mauroy, Alexandre, Aivar Sootla, and Igor Mezić. |
| 8 | +# *Koopman framework for global stability analysis.* |
| 9 | +# The Koopman Operator in Systems and Control: Concepts, Methodologies, and Applications (2020): 35-58. |
| 10 | + |
| 11 | +using Test #src |
| 12 | +using DynamicPolynomials |
| 13 | +@polyvar x[1:2] |
| 14 | + |
| 15 | +a = 1 |
| 16 | +I = 0.05 |
| 17 | +ε = 0.08 |
| 18 | +γ = 1 |
| 19 | +F0 = [-x[2] - x[1] * (x[1] - 1) * (x[1] - a) + I, ε * (x[1] - γ * x[2])] |
| 20 | + |
| 21 | +# We move equilibrium `(0.0256, 0.0256)` to the origin |
| 22 | + |
| 23 | +x1 = x2 = 0.0256 |
| 24 | +F = [f(x => [x[1] + x1, x[2] + x2]) for f in F0] |
| 25 | + |
| 26 | +# We compute the Jacobian at the equilibrium |
| 27 | + |
| 28 | +J = [j(x => zeros(2)) for j in differentiate(F, x)] |
| 29 | + |
| 30 | +# We see that its eigenvalues indeed have negative real part: |
| 31 | +using LinearAlgebra |
| 32 | +E = eigen(J) |
| 33 | + |
| 34 | +# We set `w` as its dominant eigenvector: |
| 35 | + |
| 36 | +λ = E.values[end] |
| 37 | +w = E.vectors[:, end] |
| 38 | + |
| 39 | +using SumOfSquares |
| 40 | +r = 0.3 |
| 41 | +X = @set x[1]^2 + x[2]^2 ≤ r^2 |
| 42 | + |
| 43 | +import CSDP |
| 44 | +model = SOSModel(CSDP.Optimizer) |
| 45 | +@variable(model, γ) |
| 46 | +@objective(model, Min, γ) |
| 47 | +@variable(model, ϕN, Poly(monomials(x, 2:10))) |
| 48 | +ϕ = w ⋅ x + ϕN |
| 49 | +∇ϕ = differentiate(ϕ, x) |
| 50 | +@constraint(model, -γ ≤ F ⋅ ∇ϕ - λ * ϕ, domain = X) |
| 51 | +@constraint(model, F ⋅ ∇ϕ - λ * ϕ ≤ γ, domain = X) |
| 52 | +optimize!(model) |
| 53 | +solution_summary(model) |
| 54 | + |
| 55 | +ϕ_opt = value(ϕ) |
| 56 | + |
| 57 | +using Plots |
| 58 | +x1s = x2s = range(-0.3, stop = 0.3, length = 40) |
| 59 | +ϕs = ϕ_opt.(x1s', x2s) |
| 60 | +contour(x1s, x2s, ϕs, levels=[-0.2, -0.1, 0, 0.1, 0.2, 0.5, 0.7]) |
| 61 | +θ = range(0, stop = 2π, length = 100) |
| 62 | +plot!(r * cos.(θ), r * sin.(θ), label = "") |
| 63 | +scatter!([0], [0], label = "") |
| 64 | + |
| 65 | +# We can compute the Laplace average as follows: |
| 66 | + |
| 67 | +using DifferentialEquations |
| 68 | +function S(t, x1, x2, solver = DifferentialEquations.Tsit5()) |
| 69 | + tspan = (0.0, t) |
| 70 | + prob = DifferentialEquations.ODEProblem((v, p, t) -> [f(x => v) for f in F], [x1, x2], tspan) |
| 71 | + traj = DifferentialEquations.solve(prob, solver, reltol=1e-4, abstol=1e-4) |
| 72 | + return traj[end] |
| 73 | +end |
| 74 | + |
| 75 | +using QuadGK |
| 76 | +function laplace_average(f, λ, x1, x2, T = 10, args...) |
| 77 | + v, _ = quadgk(0, T, rtol=1e-3) do t |
| 78 | + s = S(t, x1, x2, args...) |
| 79 | + return f(S(t, x1, x2, args...)) * exp(-λ * t) |
| 80 | + end |
| 81 | + return v / T |
| 82 | +end |
| 83 | + |
| 84 | +lap(x1, x2) = laplace_average(v -> ϕ_opt(x => v), λ, x1, x2, 10) |
| 85 | +laplace = lap.(x1s', x2s) |
| 86 | + |
| 87 | +# The error is given by: |
| 88 | + |
| 89 | +norm(laplace - ϕs) |
0 commit comments