This paper focuses on a subclass of box-constrained, non-linear optimization problems. We are particularly concerned with settings where gradient information is unreliable, or too costly to calculate, and the function evaluations themselves are very costly. This encourages the use of derivative...