(k1i + e1i)X1i + (k2i + e2i)X2i + uX1iX2i

if u=0 then the equation can be estimated by standard 'bivariate probit'. However, when u>0 the decision of (X1i, X2i) = (1,1) or (0,0) will be biased; when u>0 the decision of (X1i, X2i) = (1,0) or (0,1) will suffer the biaseness also. 

My first question is of how and where to put the restriction of u > 0 and u< 0 when i am programming ml.

Second question is that do i have to write theta3 for uX1iX2i? if i do, how should i put it?  any suggestion or help will be very much appreciated. thank you!