Author: Stan Arts
Date: 05:34:04 09/22/04
Go up one level in this thread
On September 22, 2004 at 07:53:37, Andrew Platt wrote: >Actually, more accurately a problem with null move and a position that appeared >in this week's Washington Post Chess column: > >[D] 3K4/8/2ppp3/2pkp3/2ppp3/8/4P3/3Q4 w - - 0 1 > >The main thing to do is to wait for black to commit suicide, something that null >moves are very bad at doing! The solution is to make an active waiting move that >will position the Queen directly later: > >1. Qa1 d3 2. Qc3 e3 3. exd3 cxd3 4.Qxd3# > >However, Schola does not want to get this mate. I can run it as long as I like >and it still goes for the longer mate: > >1. Kd7 d3 2. e3 c3 3. Qb3+ c4 4. Qb4 c2 5. Qxd6# > >If I disable null move it finds this very quickly. I suspect that there's a bug >in the search somewhere though I can't find it and the traces look sane. The >main problem is that the null move for Qa1 always cuts off so it retains Kd7 >which is the first move it considers. > >Has anyone else had problems with positions like these and null moves? BTW, I >have verified null move implemented but that doesn't help either. That does make >a small difference if I jump the position to 1. Qa1: It then finds the fastest >mate which the regular null move still doesn't do. I've also disabled some other >things like the hash table which sometimes get in the way for me but that didn't >change anything. > >Andy. Hi, Thanks for the position. Maybe the first problem is doing nullmoves for black, even when he has no pieces? That seems very risky, just think of pawn-endgames where the kings are involved, then it often gets down to pawn-races with move order and zugzwangs desiding the outcome. Using nullmove there will make your program blind to this, even with some form of verificationsearch I would suspect it does more harm then good overall. (Because you'll often spot tactics some depths later, (till the verificationsearch with reduced depth finds it) and nullmove probably isn't so efficient because there aren't that many pieces to leave en-prise every other move. But I'm not sure about the last.) I suspect this is the problem, because how can you miss the mate by a nullmove by white? (Because if with a nullmove you get >=beta, the score will go up, and with a research or so find the mate. :) ) And then, shouldn't a verificationsearch eventually spot the shortest mate anyway? (Because of the (brute force?)verificationsearches getting deeper as searchdepth increases.) So maybe there is a problem with that. Neurosis 1.81 (P3-550) : Legal moves: 19 Static score: 274 Ply Score Time Nodes Best move and expected line ------------------------------------------------------------ 2 267 0 103 Kd7 c3 3 269 0 239 Kd7 c3 e3 3 281 0 481 Qd2 e3 Qd1 4 276 5 1349 Qd2 e3 Qa2 e4 4 350 5 1930 Kd7 d3 Qa4 Kd4 5 350 5 4490 Kd7 d3 Qa4 Kd4 Qxc6 6 402 11 11656 Kd7 < + > d3 Qa4 Kd4 Qxc6 c3 6 423 16 14701 Kd7 d3 Qa4 Kd4 Qxc6 Ke3 7 464 33 33348 Kd7 < + > e3 Qa4 Ke4 Qxc4 Kf4 Qxe6 7 555 44 42312 Kd7 < + > c3 Qb3+ c4 Qb4 e3 Qxd6+ 7 9989 44 42688 Kd7 < + > c3 Qb3+ c4 Qb6 d3 Qxc6+ 7 9989 44 43665 Kd7 <M6> c3 Qb3+ c4 Qb6 d3 Qxc6+ Kd4 Qb6+ Kd5 Qxd6+ 7 9993 66 80348 Qa1 < + > d3 Qc3 e3 exd3 cxd3 Qxd3+ 7 9993 66 80650 Qa1 <M4> d3 Qc3 e3 exd3 cxd3 Qxd3+ 8 9993 88 109390 Qa1 <M4> d3 Qc3 e3 exd3 cxd3 Qxd3+ Total nodes:161619 n/sec:117900 (q-nodes:4% max depth:17) Greetings Stan
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.