Lyapunov Function Search

Adapted from: SOSTOOLS' SOSDEMO2 (See Section 4.2 of SOSTOOLS User's Manual)

In [1]:
using DynamicPolynomials
@polyvar x[1:3]
Out[1]:
(DynamicPolynomials.PolyVar{true}[x₁, x₂, x₃],)

We define below the vector field $\text{d}x/\text{d}t = f$

In [2]:
f = [-x[1]^3 - x[1] * x[3]^2,
     -x[2] - x[1]^2 * x[2],
     -x[3] - 3x[3] / (x[3]^2 + 1) + 3x[1]^2 * x[3]]
Out[2]:
3-element Vector{MultivariatePolynomials.RationalPoly{DynamicPolynomials.Polynomial{true, Int64}, DynamicPolynomials.Polynomial{true, Int64}}}:
 (-x₁³ - x₁x₃²) / (1)
 (-x₁²x₂ - x₂) / (1)
 (3x₁²x₃³ + 3x₁²x₃ - x₃³ - 4x₃) / (x₃² + 1)

We need to pick an SDP solver, see here for a list of the available choices. We use SOSModel instead of Model to be able to use the >= syntax for Sum-of-Squares constraints.

In [3]:
using SumOfSquares
using CSDP
solver = optimizer_with_attributes(CSDP.Optimizer, MOI.Silent() => true)
model = SOSModel(solver);

We are searching for a Lyapunov function $V(x)$ with monomials $x_1^2$, $x_2^2$ and $x_3^2$. We first define the monomials to be used for the Lyapunov function:

In [4]:
monos = x.^2
Out[4]:
3-element Vector{DynamicPolynomials.Monomial{true}}:
 x₁²
 x₂²
 x₃²

We now define the Lyapunov function as a polynomial decision variable with these monomials:

In [5]:
@variable(model, V, Poly(monos))
Out[5]:
$$ (_[1])x_{1}^{2} + (_[2])x_{2}^{2} + (_[3])x_{3}^{2} $$

We need to make sure that the Lyapunov function is strictly positive. We can do this with a constraint $V(x) \ge \epsilon (x_1^2 + x_2^2 + x_3^2)$, let's pick $\epsilon = 1$:

In [6]:
@constraint(model, V >= sum(x.^2))
Out[6]:
$$ (_[1] - 1)x_{1}^{2} + (_[2] - 1)x_{2}^{2} + (_[3] - 1)x_{3}^{2} \text{ is SOS} $$

We now compute $\text{d}V/\text{d}x \cdot f$.

In [7]:
using LinearAlgebra # Needed for `dot`
dVdt = dot(differentiate(V, x), f)
Out[7]:
$$ \frac{(-2 _[1])x_{1}^{4}x_{3}^{2} + (-2 _[2])x_{1}^{2}x_{2}^{2}x_{3}^{2} + (-2 _[1] + 6 _[3])x_{1}^{2}x_{3}^{4} + (-2 _[1])x_{1}^{4} + (-2 _[2])x_{1}^{2}x_{2}^{2} + (-2 _[1] + 6 _[3])x_{1}^{2}x_{3}^{2} + (-2 _[2])x_{2}^{2}x_{3}^{2} + (-2 _[3])x_{3}^{4} + (-2 _[2])x_{2}^{2} + (-8 _[3])x_{3}^{2}}{x_{3}^{2} + 1} $$

The denominator is $x[3]^2 + 1$ is strictly positive so the sign of dVdt is the same as the sign of its numerator.

In [8]:
P = dVdt.num
Out[8]:
$$ (-2 _[1])x_{1}^{4}x_{3}^{2} + (-2 _[2])x_{1}^{2}x_{2}^{2}x_{3}^{2} + (-2 _[1] + 6 _[3])x_{1}^{2}x_{3}^{4} + (-2 _[1])x_{1}^{4} + (-2 _[2])x_{1}^{2}x_{2}^{2} + (-2 _[1] + 6 _[3])x_{1}^{2}x_{3}^{2} + (-2 _[2])x_{2}^{2}x_{3}^{2} + (-2 _[3])x_{3}^{4} + (-2 _[2])x_{2}^{2} + (-8 _[3])x_{3}^{2} $$

Hence, we constrain this numerator to be nonnegative:

In [9]:
@constraint(model, P <= 0)
Out[9]:
$$ (2 _[1])x_{1}^{4}x_{3}^{2} + (2 _[2])x_{1}^{2}x_{2}^{2}x_{3}^{2} + (2 _[1] - 6 _[3])x_{1}^{2}x_{3}^{4} + (2 _[1])x_{1}^{4} + (2 _[2])x_{1}^{2}x_{2}^{2} + (2 _[1] - 6 _[3])x_{1}^{2}x_{3}^{2} + (2 _[2])x_{2}^{2}x_{3}^{2} + (2 _[3])x_{3}^{4} + (2 _[2])x_{2}^{2} + (8 _[3])x_{3}^{2} \text{ is SOS} $$

The model is ready to be optimized by the solver:

In [10]:
JuMP.optimize!(model)

We verify that the solver has found a feasible solution:

In [11]:
JuMP.primal_status(model)
Out[11]:
FEASIBLE_POINT::ResultStatusCode = 1

We can now obtain this feasible solution with:

In [12]:
value(V)
Out[12]:
$$ 15.718362431619164x_{1}^{2} + 12.28610732110516x_{2}^{2} + 2.995845325502817x_{3}^{2} $$

This notebook was generated using Literate.jl.