Author: Robert Hyatt
Date: 11:55:35 06/07/01
Go up one level in this thread
On June 06, 2001 at 14:24:03, Scott Gasch wrote: >Hi, > >Most of us do not allow two nullmoves in a row in one search line. I know >Vincent has some speciual trick for zugzwang detection... but other than him I >don't know anyone using this technique. > >Consider, then, what happens when you make a nullmove in a parent node, recurse >into a child node with the reduced depth, and hit a usable lower bound in the >hash table. That lower bound you hit could be the result of a previous nullmove >indicated by a value with no move attached to it. That is a dangerous assumption. "UPPER" positions _also_ have no move attached since there is no best move in an UPPER position. > Right now, in this case, my >engine happily fails high in the child and returns a value that causes the >nullmove parent in the node above to fail low. > >My question is this: How is this situation different from allowing two nullmoves >in a row? The child node has in essence said "doing nothing is excellent, doing >something must be better" just the same as if it had nullmoved. Should hashed >nullmove lower bounds be allower in nodes directly under a nullmove? What is >the reasoning, if so? > >Thanks, >Scott The plus for two consecutive null moves is that if you are truly in zugzwang, where the side on move loses, then the second null-move search will fail high, which will return a beta value to the previous null-move search which will then fail low and be ignored... eliminating the zugzwang error. however, there is an issue here. Most of the null-move failures are _not_ zugzwang related. They are depth-related. By chopping off 2-3 plies of the search, you hide difficult-to-detect tactical threats and thereby overlook them. The double-null might be cute in pawn-only endings, allowing the use of null-move searches there. But in the middlegame, there is a cost associated with the second search that is not 'cheap'... And the double-null does nothing to handle the lost depth issue. In fact, it hides it further.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.