Proc NLIN is used for non-linear regression... but say one parameter
is constrained in terms of other. eg. if we put bounds as

x <= y <=z , where both x and y are the parameters to be estimated..

in this case, many times the solution comes out to be x = y = z. But
it should not happen that way and solving the same thing in matalb
confirms that they need not be the same.

Any suggestion on - if this problem can be tackled and how?