On the Fedosov-Hörmander formula for differential operators (Q1030496)

The classical Fedosov-Hörmander formula for the index of a Fredholm operator in a certain class of pseudodifferential operators can be applied to elliptic linear differential systems where the coefficients are smooth, bounded matrix-valued functions and their partial derivatives have a certain decay at infinity. Formally, the index formula makes sense under much weaker smoothness conditions which are more appropriate in many applications. In the present paper, its validity is shown for linear systems with bounded \(C^1\)-coefficients of vanishing oscillation at infinity. The proofs rely on the construction of suitable homotopies to reduce the linear system to one which fits into the classical framework and then to apply the original formula. The author avoids pseudodifferential techniques and instead makes use of the Banach algebra approach to PDEs developed by H.\,O.\thinspace Cordes and others, which seems more suitable for dealing with weak smoothness assumptions. The generalized version of the Fedosov-Hörmander formula is then used to study the \(t\)-dependence of the index for a family of systems parametrized by \(t\in [0,1]\). More precisely, the invariance of the index is proved in cases where typical homotopy invariance arguments do not apply. Then various sufficient conditions for a linear system to have index \(0\) are provided in the case where the coefficients are merely continuous and bounded with vanishing oscillation at infinity. In these cases, the index formula is not directly applicable. The paper ends with two appendices; one of them provides necessary conditions for the Fredholmness of a linear system.