Incorporating a predetermined model in the searching procedure usually
produces a robust algorithm. The model constrains *p* to be certain type of
functions of (*t*,*x*), for example, smoothed functions. This constraint
excludes many possible estimation errors caused by noise and aliasing.
Of course, the choice of predetermined model
depends on the individual application.
Ideally, the function should be constrained in both axes,
however, doing this significantly complicates the search algorithm.
Thus, I decide to
constrain the solution in *t* axis only,
which proves to be practical for many applications.

The constrained optimization can be formally stated as follows: for
each *x*, find a function *p*(*t*) that satisfies certain constraints and
minimizes

12/18/1997