Effects of reducing number of central points in central composite design? The central composite design method involves evaluating the experiment at the central point several times (typically 4 times). If this method was used for a deterministic computational experiment, then all center point results would be identical. In this case, is it best to just reduce the number of central points to 1, or maintain it at 4 to improve the representation of any correlations?
 A: For deterministic experiments, the additional runs do you no good, so you can drop them.  Unfortunately the central composite (with no repeated runs) has more distinct design points than model terms, so it's still possible to fit a model that does not interpolate the design points.  For a deterministic computer experiment, this is really bad.  Adding more center runs will tend to force the response surface to be nearer to the center run value at the cost of being further from some of the other points.
That said, people generally don't use ANOVA for deterministic computer experiments, partially for the reasons described above.  Instead many do Gaussian process regression and/or Kriging (which may or may not mean the same thing depending on who you talk to).  There's a literature on computer experiments you should check out, for example Design and Analysis of Simulation Experiments by Kleijnen  (2008).  The benefit of these is that they guarantee a response surface that interpolates each of your observations.
I know there's an R package for this, but I haven't used it.  I've used JMP, which does a nice job with these.  
