Stiff oscillatory systems, delta jumps and white noise (Q5944039): Difference between revisions

Consider the scalar ordinary differential equation (ODE) \[ dz_{N}/dt=f(z_{N})+H_{n}(t),\;0\leq t\leq \pi,\quad z_{N}(0)=z_{0},\tag{*} \] (\(f\) of class \(C^{\infty }\) global Lipschitz) with the oscillatory forcing term \(H_{N}(t)=\sum_{j=0}^{N}u_{j}(t)\), \(N\gg 1\), where \(u_{j}(t)\) are harmonic oscillator solutions to \[ d^{2}u_{j}/dt^{2}+j^{2}u_{j}=0,\quad u_{j}(0)=a_{j},\qquad du_{j}(0)/dt=0.\tag{**} \] Two cases are considered: (a) \(a_{0}=1/2\), \(a_{j}=1\), \(j=1,\dots ,N\), which approximates \(dz/dt=f(z)\), \(z(0)=z_{0}+\pi /2\) (delta forcing term); (b) \(a_{0}=\eta _{0}/\sqrt{\pi }\), \(a_{j}=\eta _{j}\sqrt{2/\pi}\), \(j=1,\dots ,N\), with \(\eta _{0},\eta _{1},\dots ,\eta _{N}\) i.i.d. standard Gaussian r.v., which approximates the SDE \(dz=f(z)dt+dw(t), z(0)=z_{0}\), where \(w(t)\) is a standard Wiener process. Section 2 compares \(z_{N}\) with \(z\), establishing, as \(N\rightarrow \infty \), the rate of strong convergence and, for the case \(f\equiv 0\), the rate of weak convergence (and, as a byproduct, the rate of convergence of the pdf's). In section 3, (*) is solved numerically by the \(\theta -\)method using the exact values of \(H_{N}(t)\) and, for the regime \(N\Delta t\) fixed, \(\Delta t\rightarrow 0\), \(N\rightarrow \infty \), the rate of convergence of the numerical solution to \(z(t)\) is established (in the case (a) with \(\theta \neq 1/2\), the convergence is not to \(z(t)\) but rather to \(\overline{z}(t)\) solution to \(dz/dt=f(z)\), \(z(0)=z_{0}+\pi /2+r(1-2\theta)/2\)). For the same regime, the following sections consider weak and strong (for \(f\equiv 0\)) convergence results and rates for numerical solutions of (*) by the \(\theta\)-method using approximate values of \(H_{N}(t)\) obtained by numerically solving (**). The situation dealt with in the paper has easy solvable limiting equations for the purpose of allowing comparisons. This enables us to get an idea of what may be going on when the limiting behavior is more complicated.