Not sure I follow the intend exactly, but if I understand you wish to pick the subset of (8) values from each of two larger sets of (13) that minimizes a function of the two. Since
>> nchoosek(13,8)
ans =
1287
and there are two combinations of these, brute-force enumeration will be time-consuming.
The thought that comes to mind would be to use one of the minimization routines and solve in a least squares sense (linear or nonlinear, depending on the functional form is the reason for the genericity rather than a specific routine) and the choose the coefficients that are nearest in magnitude to those. That doesn't guarantee that is the minimum but would likely be a reasonable guesstimate unless the model is extremelty sensitive to one (or more) coefficient.
