On the use of antithetic variates

The use of antithetic variates as a variance reduction method is investigated by numerical experimentation. It is found that if variance reduction is not carefully defined, it can go undetected. With this caveat in mind, it is shown that antithetic variates can reduce variance up to a point. The phenomenon of spontaneous correlation is identified as the cause of failure. The surprising result that it can pay to use non-scoring histories is demonstrated. A total energy deposition problem is studied. (orig.)