1: \begin{abstract}

2: 	The purpose of this  paper is to present an inexact version of the scaled gradient projection method on a convex set, which  is inexact in two sense.  First, an inexact projection on the feasible set is computed, allowing for an appropriate relative error tolerance. Second, an  inexact non-monotone line search scheme is employed to compute  a step size which defines the next iteration. It is shown that the proposed   method has similar  asymptotic convergence properties and iteration-complexity bounds as the usual  scaled gradient projection method employing monotone line searches.

3: \end{abstract}

4: