38  Problem set 6 solutions

Q1. A van der Pol equation

  1. Determine \(f(x)\) so that this equation can be written as a Liénard phase plane system in the form \[ \begin{aligned} \epsilon \dot{x} &= f(x) + 4y, \\ \dot{y} &= -x. \end{aligned} \]

Start from the ODE and re-write \[ \frac{\mathrm{d}}{\mathrm{d}t} \left[\epsilon \dot{x} + (x^3 - 3x^2 - 9x)\right] = - 4x. \] Double check that the derivative of the LHS returns the original problem. Now set the quantity in brackets equal to one unknown: \[ \tilde{y} = \epsilon \dot{x} + (x^3 - 3x^2 - 9x), \] and solve for the derivative: \[ \epsilon \dot{x} = \tilde{y} - (x^3 - 3x^2 - 9x). \] The remaining equation is \[ \dot{\tilde{y}} = -4x. \]

Now in this question, we have chosen a slightly different choice to eliminate the factor of \(4\). Set \(\tilde{y} = 4y\) to get the system

\[\begin{align} \epsilon \dot{x} &= f(x) + 4y, \\ \dot{y} &= -x. \end{align}\]

It doesn’t really matter which choice you select. So here we see \(f(x) = -x^3 + 9x + 3 x^2\). What we often call the slow manifold in lectures is then \[ y = -f(x)/4 = \frac{x^3 - 3 x^2 - 9x}{4} \equiv S(x). \]

  1. For fixed \(\epsilon > 0\), find the equilibrium point(s) in the phase plane, find their eigenvalues, and classify their linear stability.

The only equilibrium point is \((x_*, y_*) = (0,0)\). Because the equilibrium point is at the origin, we can obtain the linear matrix by ignoring the higher-order powers in the differential equations. The matrix is \[ A = \begin{pmatrix} \frac{9}{\epsilon} & \frac{4}{\epsilon} \\ -1 & 0 \end{pmatrix}. \] Subtract \(\lambda\) from the diagonal elements and solve the resultant quadratic equation \(\lambda(\lambda - 9/\epsilon) + 4/\epsilon = 0\) to see that the eigenvalues are \(\lambda = (9 \pm \sqrt{81 - 16\epsilon}/(2\epsilon) > 0\), so the point is an unstable node.

  1. Use the expansions \(x(t) = x_0(t) + \epsilon x_1(t) + O(\epsilon^2)\) to determine the equations for the leading-order slow solution. Sketch the slow manifold, indicate the directions motion on each part, and identify the two attracting points on the curve.

As usual expand the solutions into a series in powers of \(\epsilon\). The slow manifold is \(y_0(x_0) = S(x_0) = (x_0^3 - 3 x_0^2 - 9 x_0)/4\). In image of the slow manifold is shown below in blue. Notice that for \(x < 0\), \(y\) is increasing and for \(x > 0\), \(y\) is decreasing (due to the second differential equation).

Figure 38.1: Illustration of slow manifold
  1. Use the expansions \(x(t) = X_0(T) + \epsilon X_1(T) + O(\epsilon^2)\) and \(y(t) = Y_0(T) + \epsilon Y_1(T) + O(\epsilon^2)\) with \(T = t/\epsilon\) to obtain the leading-order fast solution.

Rescale \(t = \epsilon T\) and re-write \(x = X(T)\) and \(y = Y(T)\). We get \[ \begin{align} \frac{\mathrm{d}X}{\mathrm{d}T} &= f(X) + 4Y \\ \frac{\mathrm{d}Y}{\mathrm{d}T} &= -\epsilon X. \end{align} \]

Now expand into powers of \(\epsilon\) and take the leading-order: \[ \begin{align} \frac{\mathrm{d}X_0}{\mathrm{d}T} &= f(X_0) + 4Y_0 \\ \frac{\mathrm{d}Y_0}{\mathrm{d}T} &= 0. \end{align} \]

The second equation indicates that \(Y_0\) is constant. Its value, found in part (e) is either \(5/4\) or \(27/4\). The remaining equation for \(X_0\) is nonlinear and there is not much you can do with it. The important item of understanding is to examine its form: \[ \frac{\mathrm{d}X_0}{\mathrm{d}T} = f(X_0) + 4 Y_0. \] Now on the upper branch of the picture shown above \(f(X_0) = 4 Y_0 > 0\) and therefore the solution increases its value in \(X_0\) until it intersects with the blue slow manifold, at which point its rate of change is zero and it descends the slow manifold. A similar behaviour occurs on the lower branch of the fast evolution.

  1. Use the phase plane to determine the maximum and minimum values of \(x(t)\) during an oscillation. Sketch \(x(t)\) as a function of time.

By checking the derivatives, we can see that the slow manifold has local extrema at \((-1, 5/4)\) and \((3, 27/4)\). In the fast evolution stages, \(y\) is constant and \(x\) evolves from one extrema value to the other on the slow manifold, satisfying \(y = S(x)\).

Here is a plot of the key points.

Q2. Fast-slow dynamics with three variables

\(\nextSection\)
  1. For the system \[ \begin{aligned}[c] \dot{x} &= 2 - y, \\ \dot{y} &= x-z, \\ \epsilon\dot{z} &= y - y^2 + \frac{1}{3}y^3 - z, \end{aligned} \] with the initial conditions of \(x(0) = 1\), \(y(0) = 3\), and \(z(0) = 0\).

Identify the surface \(z = S(x, y)\) that defines the slow manifold. Find the equilibrium point of the leading=order slow phase plane system and show that it is asymptotically stable for \(t \to \infty\). Also determine the form of the initial layer that describes the transition from the initial conditions to the slow manifold.

Setting \(\epsilon = 0\) we see that the slwo manifold is \(z = S(y) = y - y^2 + y^3/3\). The leading-order slow system is

\[\begin{align} \frac{\mathrm{d}x_0}{\mathrm{d}t} &= 2 - y_0 \\ \frac{\mathrm{d}y_0}{\mathrm{d}t} &= x_0 - y_0 + y_0^2 - y_0^3/3. \end{align}\] $$ The only equilibrium point is \((x^*, y^*) = (2/3, 2)\). Linearisation gives eigenvalues of \(\lambda = 1/2(-1 \pm 3i)\) so the point is a stable spiral. Consequently the \(t \to \infty\) solution will be \((x, y, z) = (2/3, 2, 2/3)\). Note also that \(2/3 = S(2)\).

Near the initial condition there is a boundary layer, found by setting \[ t = \epsilon T, \] and transforming \(x = X(T)\), \(y = Y(T)\), \(z = Z(T)\). We get at leading order,

\[\begin{align} \frac{\mathrm{d}X_0}{\mathrm{d}T} &= 0 \\ \frac{\mathrm{d}T_0}{\mathrm{d}T} &= 0 \\ \frac{\mathrm{d}Z_0}{\mathrm{d}T} &= Y_0 - Y_0^2 + Y_0^3/3 - Z_0. \end{align}\]

The initial condition sets \(X_0 = 1\) and \(Y_0 = 3\). We can finally solve the equation for \(Z_0\) to get \[ Z_0 = 3(1 - e^{-T}). \]

  1. For the system \[ \begin{aligned}[c] \dot{x} &= 2 - y, \\ \epsilon\dot{y} &= x-z, \\ \dot{z} &= y - y^2 + \frac{1}{3}y^3 - z, \end{aligned} \] with the initial conditions of \(x(0) = 0\), \(y(0) = 3\), and \(z(0) = 1\).

Show that the slow manifold reduces to a curve that could be written in parametric form as \(x = x(z), y = y(z), z = z\). Determine the asymptotic solution for \(t \to \infty\). Also determine the form of the initial layer that describes the transition from initial conditions to the slow manifold.

Setting \(\epsilon = 0\), we see that the second equation gives \(x = z\). Now if you use this result in the first equation you obtain \(\dot{z} = 2 - y\). This is now equated with the third equation which yields an algebraic equation: \[ 2 - y = y - y^2 + \frac{y^3}{3} - z \Longrightarrow z = \frac{y^3}{3} - y^2 + 2y - 2. \tag{38.1}\] Therefore we conclude that \(z = z(y)\). We can verify that this equation is monotonically increasing in \(y\), so we can invert without issue, yielding \(y = y(z)\). Similarly, we obtain \(x = z\). So the slow manifold is given by \[ (x, y, z) = (z, y(z), z), \] which is a curve in 3D.

In order to determine what happens as \(t\to\infty\), differentiate \[ \frac{\mathrm{d}y}{\mathrm{d}t} = \frac{\mathrm{d}y}{\mathrm{d}z} \frac{\mathrm{d}z}{\mathrm{d}t} = \frac{2 - y}{2 - 2y + y^2}, \] which follows from using the first equation of the system of ODEs and also the \(z\) derivative of (Equation 38.1). In particular, note that we do not have to invert this latter equation and we can take the \(y\)-derivative of (eq-PS5-Q3-tmp) and use \[ \frac{1}{\frac{\mathrm{d}z}{\mathrm{d}y}} = \frac{1}{2 - 2y + y^2}. \]

Take \(dy/dt = 0\) for the steady state gives \[ y^* = 2, \] as the only fixed point. At this value of \(y\), notice that \(z^* = 2/3\) from (Equation 38.1). So the fixed point is \[ (x^*, y^*, z^*) = (2/3, 2, 2/3). \]

The procedure for the initial layer is the same as in part (a). This time, the fast system satisfies at leading order,

\[\begin{align} \frac{\mathrm{d}X_0}{\mathrm{d}T} &= 0 \\ \frac{\mathrm{d}Y_0}{\mathrm{d}T} &= X - Z \\ \frac{\mathrm{d}Z_0}{\mathrm{d}T} &= 0. \end{align}\]

So we obtain \(X_0 = 0\) and \(Z_0 = 1\) using the initial condition. Solving for the final equation for \(Y_0\) with \(Y_0(0) = 3\) gives \[ Y_0 = -T + 3. \]