Windowing is the simplest of the following techniques. In an iterative deepening framework, we always have a value of the previous iteration. Therefore, we try to reduce the search effort by using a smaller alpha-beta window. Instead of using -INFINITY...+INFINITY we can use an interval lastvalue-WINDOW...lastvalue+WINDOW. If the true MiniMax value at this iteration is really inside this window, we will just need to search fewer nodes than with the larger window. On the other hand, our guess that the true value is inside this window might also be wrong. In this case, we will get a fail-high or a fail-low. In this case, we will have to do a re-search with a larger window. '

See also