Response surface methodology's steepest ascent and step size revisited

Response Surface Methodology (RSM) searches for the input combination maximizing the output of a real system or its simulation. RSM is a heuristic that locally fits first-order polynomials, and estimates the corresponding steepest ascent (SA) paths. However, SA is scale-dependent; and its step size is selected intuitively. To tackle these two problems, this paper derives novel techniques combining mathematical statistics and mathematical programming. Technique 1 called 'adapted' SA (ASA) accounts for the covariances between the components of the estimated local gradient. ASA is scale-independent. The step-size problem is solved tentatively. Technique 2 does follow the SA direction, but with a step size inspired by ASA. Mathematical properties of the two techniques are derived and interpreted; numerical examples illustrate these properties. The search directions of the two techniques are explored in Monte Carlo experiments. These experiments show that-in general-ASA gives a better search direction than SA.