Parameter sensitivity equation (source code)

= Parameter sensitivity equation

For the <initial value problem> $y'=f(t,y,\theta)$, $y(t_0,\theta)=y_0(\theta)$, assume $f$ is continuous in all variables with continuous state and <parameter> derivatives, $y_0$ is continuously differentiable, and solutions exist on a common interval. The sensitivity $z=\partial_\theta y$ obeys the <inhomogeneous linear differential equation>
$$
z'=D_yf(t,y,\theta)z+\partial_\theta f(t,y,\theta),\qquad z(t_0)=\partial_\theta y_0(\theta).
$$
This follows by differentiating the original equation and applying the <chain rule>. For several <parameters>, the columns of the sensitivity <matrix> satisfy the corresponding equations. The homogeneous part is the <variational equation>, while explicit parameter dependence supplies the <forcing term>.

For $y'=-\theta y$ with parameter-independent $y(0)=b$, the exact solution is $y=be^{-\theta t}$ and $z=-tbe^{-\theta t}$. Substitution gives $z'=-\theta z-y$ and $z(0)=0$, verifying both the evolution and its differentiated initial condition.