Dependence on Parameters

Consider the differential equation \begin{equation} \dot x = f(t, x, \alpha),\qquad x(0) = x_0, \label{eq-ivp-with-parameter} \end{equation} where $\alpha$ is a parameter (i.e. a constant that appears in the differential equation).

We assume that the function $f$ is a continuously differentiable function of $(t, x, \alpha)$.

Theorem. Assume that for some choice $(\bar \alpha, \bar x_0)$ of the parameter and initial value, the initial value problem \eqref{eq-ivp-with-parameter} has a solution $\bar x(t)$ that is defined on a closed interval $[0, t_1]$. Then there is an $\varepsilon\gt0$ such that for all $\alpha$ and $x_0$ with $|\alpha-\bar \alpha| \lt \varepsilon$ and $|x_0-\bar x_0|\lt \varepsilon$ the equation \eqref{eq-ivp-with-parameter} has a solution $x(t, \alpha, x_0)$ that is defined on the same interval $[0, t_1]$. The equation $x(t, \alpha, x_0)$ is a continuously differentiable function of $(t, \alpha, x_0)$.
The partial derivatives with respect to the parameters $\alpha$ and $x_0$ satisfy their own differential equation. This equation is easily derived by “differentiating the equation”. For example, $y=\frac{\pd x} {\pd \alpha}$ satisfies \[ \frac{\pd y} {\pd t} = f_x(t, x(t), \alpha) y(t) + f_\alpha(t, x(t), \alpha), \qquad y(0) = 0, \] while $z = \frac{\pd x} {\pd x_0}$ satisfies \[ \frac{\pd z} {\pd t} = f_x(t, x(t), \alpha) z(t), \qquad z(0) = 1. \] Each of these two equations is called “the variational equation” for the differential equation \eqref{eq-ivp-with-parameter}.

The variational equation is always a first order linear differential equation: the equation for $y$ is inhomogeneous, the equation for $z(t)$ is homogeneous. For a review of how to solve linear first order equations, follow this link.

Example: solving the variational equation

Let $x(t, \alpha)$ be the solution of \[ \dot x = x(1-x) + \alpha\sin kt, \qquad x(0)=0. \] Here $k$ and $\alpha$ both are constants, but we will only vary $\alpha$ (in this example—you could also consider variations in $k$.)

For general $\alpha$ we do not have a formula for the solution to this equation, but for $\alpha=0$ we do have one, namely: $x(t, 0) = 0$. For small values of $\alpha$ the solution should be approximately given by

\begin{equation} x(t, \alpha) = x(t, 0) + \frac{\pd x}{\pd \alpha}(t, 0)\,\alpha +\dots \label{eq-Taylor-in-alpha} \end{equation}
by Taylor’s formula (the dots “$\dots$” contain the higher order terms, and the remainder term). Here we have used Taylor’s formula in the following way. Freeze the variable $t$ at some value, and think of $x(t, \alpha)$ as a function of $\alpha$ only. Taylor then says \[ x(t, \alpha) = x(t, 0) + \frac{\pd x}{\pd\alpha}(t,0) \alpha +\frac{\pd^2 x}{\pd\alpha^2}(t,0) \frac{\alpha^2}{2!} +\frac{\pd^3 x}{\pd\alpha^3}(t,0) \frac{\alpha^3}{3!} +\cdots \] The derivative \[ y(t) = \frac{\pd x}{\pd \alpha}(t, 0) \] satisfies the equation \[ \dot y = \bigl(1-2x(t, 0)\bigr) y(t) + \sin kt, \qquad y(0)=0 \] or, taking into account that our known solution is $x(t, 0) = 0$, \[ \dot y = y + \sin kt, \quad y(0)=0. \] The solution of this linear equation is (see below for the details) \[ y(t) = \frac{-\sin kt- k\cos kt + k e^{t}}{1+k^2}. \] Conclusion: for small values of $\alpha$ the solution $x(t, \alpha)$ is approximately given by the Taylor expansion \eqref{eq-Taylor-in-alpha} \[ x(t, \alpha) = \alpha y(t) + \cdots = \alpha \frac{k e^{t} -\sin kt- k\cos kt}{1+k^2} + \cdots \]

Solving that linear equation

Rewrite it as \[ \dot y - y = \sin kt \] and multiply both sides with the integrating factor $e^{-t}$: \[ \frac{d}{dt}\bigl(e^{-t}y\bigr) = e^{-t}\dot y - e^{-t}y = e^{-t}\sin kt. \] Now integrate $\frac{d}{dt}(e^{-t}y)$ to get $e^{-t}y$: \[ e^{-t}y = \int e^{-t}\sin kt \, dt = - e^{-t} \frac{\sin kt + k\cos kt}{1+k^2} + C. \] Don’t forget the “$+C$”'! (How do you do the integral? Integrate by parts twice, look it up, or ask Wolfram).

The initial condition $y(0)=0$ tells us that \[ 0 = -\frac{k}{1+k^2} + C \implies C=\frac{k}{1+k^2}. \] Therefore the solution is \[ y(t) = - \frac{\sin kt+ k\cos kt}{1+k^2} + C e^t = \frac{-\sin kt- k\cos kt + k e^{t}}{1+k^2}. \]