Partial decoupling of slow and fast variables (Q752943)

Consider a linear system of ordinary differential equations which contains a small positive parameter \(\epsilon\), \((*)\quad (y'_ 1,\epsilon y'_ 2)^ T=A(t)(y_ 1,y_ 2)^ T+(F_ 1(t),F_ 2(t))^ T,\) where \(t\in (a,b)\subset R\); \(y_ 1,y_ 2\) are real vector functions of dimensions m and n; \(A(t)=(A_{ij}(t))\) is a real \(2\times 2\) matrix function whose elements are of appropriate dimensions; \(F_ 1,F_ 2\) are real vector functions of dimensions m,n, respectively. It is assumed that: A1. The matrix functions \(A_{ij}(t)\) are continuously differentiable for \(i,j=1,2\); A2. There are constants \(b_{ij}>0\) and \(c_{ij}>0\) such that \(\| A_{ij}(t)\| \leq b_{ij}\) and \(\| A'_{ij}(t)\| \leq c_{ij}\), where \(\| \cdot \|\) is some matrix norm; A3. There are constants \(c_ 3>0\) and \(c_ 4>0\) such that any eigenvalue \(\lambda\) (t) of \(A_{22}(t)\) satisfies Re \(\lambda\) (t)\(\leq -c_ 3\) or Re \(\lambda\) (t)\(\geq c_ 4\). Under these assumptions, a partial decoupling transformation is constructed which decouples the slow component from (*), in the sense that, after the transformation is applied, the slow component becomes independent whereas the equations for the fast component still contain the slowly varying component.