4.4 KiB

What is a limit cycle?

Isolated, closed trajectories.

  1. Not like a center.
  2. Centers are closed, but not isolated.
  3. Neighboring trajectories are NOT closed. Different forms:
  4. Stable - Trajectories pull onto the limit cycle
  5. Unstable - Trajectories are repelled by the limit cycle.

A imit cycle is a explicitly nonlinear phenomenon.

You can't identify if there is a limit cycle by using linearizing methods.

How do we find limit cycles?

How do we rule out a closed loop?

Bendixon's Criterion:

If we have some flow field:

\dot{\vec{x}}= f(\vec x)
  • If we can find a function \zeta(x,y) such that \nabla \cdot (\zeta f)) does not change sign in some region of R, then there's no limit cycle in that region.
  • If in some region R, \zeta(x,y) s.t : \frac{\partial}{\partial x} (\zeta(x,y) f_1(x,y)) + \frac{\partial}{\partial y}(\zeta(x,y) f_2(x,y)) is of constant sign, then there are no closed orbits in R. Finding \zeta is tricky.

Example: \dot x = y \dot y = -x -y + x^2 + y^2

Assume \zeta(x,y) = 1 \partial / \partial x (y) + \partial / \partial y (-x - y +x^2 +y^2) /rightarrow 0 + (-1+2y)

Assume \zeta(x,y) = e^{\alpha x}

\partial / \partial x (e^{\alpha x} y) + \partial / \partial y (e^{\alpha x} (-x - y +x^2 +y^2)) \alpha e^{\alpha x} y + 2 y e^{\alpha x} - e^{\alpha x} e^{\alpha x}((\alpha+2) y -1) Now let \alpha = -2 \nabla \cdot (\zeta f) = e^{-2 x}

Now a special note: These functions can define where limit cycles can't be. If the function doesn't change sign for a subset of R, there can't be a limit cycle contained in that subset. There CAN be a limit cycle that crosses the point the function changes sign.

Lyapunov Function

Aleksander Lyapunov (Liapunov) V(\vec x) = V(x,y) \leftarrow a scalar function V(\vec x) > 0 \forall \vec x\neq \vec x^* V(\vec x^*) = 0 \dot V = \frac{dV}{dt} <0 V(\vec x) is a positive definite function. Then the system is stable ISL (in the sense of Lyapunov). The system will always asymptotically approach the equilibrium point. \frac{dV}{dt} = \frac{dV}{dx} \frac{dx}{dt} + \frac{dV}{dy} \frac {dy}{dt} = \dot x \frac{dV}{dx} + \dot y \frac{dV}{dy}

Example: \dot x = y - x^3 \dot y = -x-y^3 V(x,y) = c_1 x^2 + c_2 y^2 \frac{dV}{dt} = 2 c_1 x \dot x + 2 c_2 y \dot y = 2 c_1 x(y-x^3) + 2c_2 y(-x-y^3) Assume c_1 = c_2 ... \therefore \frac{dV}{dt} = -2c(x^4+y^4) < 0 Therefore limit cycles are not possible.

Index Method

This is a method covered in the book. Sometimes is used to rule out limit cycles.

Poincare - Bendixon Theorem.

Book!

Perturbation Methods

  • Weakly nonlinear systems Linear Resonator: m \ddot x + b \dot x + kx = f Weakly Nonlinear: m \ddot x + b \dot x + kx + \alpha x^3 = f With a bookkeeping term: m \ddot x + b \dot x + kx + \epsilon \alpha x^3 = f

Asymptotic Expansion

x \neq x(t) \rightarrow x = x(t,\epsilon) x(t,\epsilon) = x_0(t) + \epsilon x_1(t) + \epsilon^2 x_2(t) + ... + \text{H.O.T.s} Looking for solutions that are like

x(t,\epsilon) ~ \sum_{k=0}^{\inf} x_k(t) \delta_c(\epsilon)

Where \delta is an asymptotically scaling number. This series sometimes doesn't converge but still gives useful information about the solution. Example: for x>=0

\dot x + x - \epsilon x^2 = 0, x(0) = 2

Develop a 3 term approximation using asymptotic expansion:

x(t,\epsilon) = x_o(t) + \epsilon x_1(t) + \epsilon^2 x_2(t) + ... \dot x(t,\epsilon) = \dot x_o(t) + \epsilon \dot x_1(t) + \epsilon^2 \dot x_2(t) + ...

Sub into the EOM:, and satisfy initial conditions x_0(2) = 0; x_1(0) = x_2(0) = 0

\dot x_o(t) + \epsilon \dot x_1(t) + \epsilon^2 \dot x_2(t) + x_o(t) + \epsilon x_1(t) + \epsilon^2 x_2(t) - \epsilon (x_o(t) + \epsilon x_1(t) + \epsilon^2 x_2(t))^2 = 0

Now that last term is going to yield higher order \epsilon terms (^2, ^4). We can't get rid of these, we'll need to keep them. Now collect terms:

Power Expression
\epsilon^0 \dot x_0 + x_0 = 0 \rightarrow x_0 = c_1e^{-t} \rightarrow x_0 = 2 e^{-t}
\epsilon^2 \dot x_1 + x_1 - x_0^2 = 0 \rightarrow \dot x_1 + x_1 - 4 e^{2t} = 0 \rightarrow x_1 = 4(e^{-t} - 2e^{-2t})
\epsilon^3 \dot x_2 + x_2 -2(2e^{-t})(4 e^{-t} - e^{-2t}) \rightarrow ...
Then we have an approximate solution for small \epsilon. What small means depends on the problem...