Suposse that you have two systems with A1(x)= G*(x+1) and A2(x) = G*(x+1)^2 representing its respectives behaviour. Now, introduce an input x= Cos(wt). Then, A1(x) = G*B*Cos(wt) + G and A2(x)= G*(B^2*Cos^2(wt)+2*B*Cos(wt)+1) =(after some algebra).. = G*(3/2+ 2*B*Cos(wt)+ B^2*Cos(2wt)/2). From above we see that if in the first system we introduce a signal with the form Cos(wt), we obtain at the output the same signal scaled with a constant G displaced by G, while in the second, we obtain components with multiples of the fundamental frecuency (2wt, 3wt, etc, depending of the system´s grade). So, we can conclude that the first system es linear while second is not. A linear (and time invariant, which is another topic) CANT generate harmonics while a non-linear can. The THD is the measure of how much a signal is distorted. So, non-linearity is a property while THD is a measure. I think that the offset in the multiplier is the level of DC generated at the output (0 volts ideal) but, due mistmach you will have a non-zero DC voltage at the output.
Hope will help
Best regards
Jesus