Author: Robert Hyatt
Date: 10:35:13 11/25/98
Go up one level in this thread
On November 24, 1998 at 22:05:52, Jeremiah Penery wrote:
>On November 24, 1998 at 21:19:24, Robert Hyatt wrote:
>
>>On November 24, 1998 at 20:28:50, Quenton Fyfe wrote:
>>
>>>Afraid I can't reproduce this either. > 6.5 hours on a P6 200, and CM5000 is
>>>still planning Qb6. That's >500m positions examined. So unless there were
>>>special settings involved (CM5K is quite configurable) then it's no go.
>>>
>>>I've also run Fritz for about 36 hours on the same hardware, and it wanted to
>>>play Qb6 too.
>>>
>>>Bob - can we work out from the Deep Blue node rate the maximum time it should
>>>take on with brute force search to find out what's wrong with Qb6? I guess it'd
>>>be weeks huh?
>>>
>>>Cheers
>>>
>>>Quenton Fyfe
>>
>>
>>It is difficult to do. IE take the nolot positions and try them on all the
>>programs. You will find one program that gets one particular position quicker
>>than all the others, yet it gets killed on the other positions. CM5xxx is a
>>good example... it finds lots of good tactical shots that take otherprograms
>>minutes or hours to find. That's why it isn't easy to figure out how deep we
>>have to search... The main programs searching on Qb6/axb5 are/were mine,
>>Bruce's and Ernst's... and all are null-move programs. It is possible that
>>this is a null-move killer position as we have seen in the past. and if you
>>have followed null-move "problems" there are some positions a null-move search
>>won't *ever* solve...
>
>Can you give an example (or more) of a position like this?
>Also, what factor(s) cause these to never be solved when using null-move?
>
>Thanks.
>Jeremiah
Here are two classics:
+---+---+---+---+---+---+---+---+
8 | | | | | *K| *B| R | |
+---+---+---+---+---+---+---+---+
7 | R | | | | | | | |
+---+---+---+---+---+---+---+---+
6 | | | | | | | | *P|
+---+---+---+---+---+---+---+---+
5 | | | | | | | | |
+---+---+---+---+---+---+---+---+
4 | | | | | | | | P |
+---+---+---+---+---+---+---+---+
3 | | | | | | | P | |
+---+---+---+---+---+---+---+---+
2 | | | | | | P | | |
+---+---+---+---+---+---+---+---+
1 | | | | | | | K | |
+---+---+---+---+---+---+---+---+
a b c d e f g h
fen: 4kbR/R/7p//7P/6P/5P/6K/ w
In the above position what is going on is obvious. Black is zugged
if he can't play h5.. and if he does he is zugged the next move. Crafty
finds lots of moves (eventually) lead to a zug and mate... but it takes
8 plies to find this simple mate in 2: 1. h5 Kd8 2. Rxf8# Because the
null-move search fails since not moving is the best thing black can do
here.
Here is another one:
+---+---+---+---+---+---+---+---+
8 | *K| | | | | | | |
+---+---+---+---+---+---+---+---+
7 | *P| *P| | | | | | |
+---+---+---+---+---+---+---+---+
6 | | | | | | | | |
+---+---+---+---+---+---+---+---+
5 | | | | | | *Q| | |
+---+---+---+---+---+---+---+---+
4 | | *P| | | | | | |
+---+---+---+---+---+---+---+---+
3 | *P| P | *R| | | | | |
+---+---+---+---+---+---+---+---+
2 | P | | P | R | | | | |
+---+---+---+---+---+---+---+---+
1 | | B | K | | | | | |
+---+---+---+---+---+---+---+---+
a b c d e f g h
fen: k/pp//5q/1p/pPr/P1PR/1BK b
This is another zug case... Qf1+ Rd1 Qf4+ Rd2 is the key stuff. Now
black plays a waiting move with the king or pawns, and white is forced
to play Kd1, which leads immediately to Qf1#. But if white can play a
"null" when he is zugged, then this doesn't happen, because after (say)
Kb8 <null> Qf1 leads to nothing but a repetition. So the null move search
fails high, and we conclude black's position is gross...
This is an easy mate, and doesn't happen to hurt me too badly, because
black is just low enough in material that some "tricks" I use happen to
work here. But you should get the idea now... if you can find a zugzwang
position in the middlegame (ie with at least a queen left on the board,
so that null-move programs don't stop using the heuristic, you have found a
position that will give a null-mover problems. These are very rare... but
they can and do happen.
Other serious cases happen in a different way...
suppose black has played g6 after castling kingside... if you reach a position
in the tree with white having a queen at h6 and a pawn/bishop at f6, with no
way for black to defend g7, this is a mate in 1. But if you reach this position
with black to move, near the depth limit on the tree, black can play a
null-move, which reduces the depth by 2 extra plies, which can take you right
to the capture search where you don't find Qg7#. This used to kill me until
I got my speed up to the point where I don't do 4-5 ply searches, even in
bullet, any longer. If you do a 5 ply search, and reach this at ply=2, you are
dead, because white's next move will be a quiesce() move... and it will likely
overlook Qg7# unless you include non-captures in the q-search...
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.