tournament selection for ga algorithm

Hi all,

I am using a the ga (genetic algorithm) to optimize a function with non linear constraints. I want to use the penalty function instead of the augmented Lagrangian to handle the non-linear constraints. I have read in the doc that the penalty function uses the "selection tournament" function. I would like to get a but more information about how this "selection tournament" function works: - How does it select the two results to be compared ? - Why does it sometimes not keep the population constant ? ( I usually start with a population of 50 but very quickly after the third or the fourth generation my population has drastically decreased ) - Does it still uses crossover or mutation ?

Many thanks, Etienne