Author: Dan Newman
Date: 11:19:14 09/01/98
Go up one level in this thread
On August 30, 1998 at 09:46:29, Enrique Irazoqui wrote:
>[Event ""Rigasche Rundschau""]
>[Site "?"]
>[Date "1907.??.??"]
>[Round "?"]
>[White "Behting"]
>[Black "?"]
>[Result "1-0"]
>[SetUp "1"]
>[FEN "8/8/8/5pp1/8/1K6/2pp1Q2/2k5 w - - 0 1"]
>[PlyCount "5"]
>[EventDate "1907.??.??"]
>
>1. Qe3 f4 2. Qf2 d1=Q (2... f3 3. Qe3) 3. Kc3 f3 4. Qe3+ Kb1 5. Qb6+ Kc1 6. Qb2#
>1-0
>
>This is a relatively simple position where white wins in 12 plies or less, but
>several programs fail to find Qe3, even after a search deeper than 12 plies.
>
>After at least 5 minutes on a PII-400 with 256 MB RAM, Hiarcs 6, Fritz 5, Mchess
>7.1, Nimzo 98, Crafty 15.17, Exchess 2.34 and Junior 4.6 don’t find the
>solution. Junior plays instantaneously Kc3?? and announces mate in 4,
>overlooking the obvious underpromotion d1N+.
>
>Time to find Qe3 with a winning evaluation:
>Genius 5W, 13 secs.
>Rebel 9, 30 secs.
>Rebel 10, 24 secs.
>Shredder 2, 24 secs.
>
>Strange that there is so much difference between programs in this position.
>
>What would Ferret, Dark Thought and others play?
>
>Enrique
Very interesting position. I think it has a bit of null-move-itis.
My current program (Skyblue) doesn't find Qe3 after completing a
14 ply search (4+ minutes)--it likes Qxf5, bleh.
I became suspicious of null move, so I turned it off, and sure
enough it failed high on Qe3 in 10 seconds (actually it starts
out liking Qe3 immediately, switches to Qg3, then to Qxf5 for a
few plies, then to Kc4, and finally to the Qe3 fail-high in the
10 ply search). It holds Qe3 all the way through 16 plies of
search (15 minutes) without finding the mate... It does however
see that it can gobble 3 pawns way out there at 13, 15, and 19
plies. Since I don't have any reductions turned on, it's doing
those 16 plies full-width, and if I don't have any bugs in my code
that screw up the tree (a big if), it ought to find all mates in
15 or less.
Here's the 16 ply search's PV with null move turned off:
1. Qe3 f4 2. Qf2 f3 3. Qe3 g4 4. Kc3 Kb1
5. Qb6+ Kc1 6. Qb2+ Kd1 7. Qxc2+ Ke2 8. Qxd2+ Kf1
9. Qf4 Kf2 10. Qxg4 Ke3
which seems to follow your alternate (2... f3 3. Qe3) path.
If I make these moves, it puts us at "8/8/8/8/6Q1/2K1kp2/8/8 w - -"
which is a mate in 7 for white. Of course black may try something
different in the first 10 moves and drive the mate deeper or prevent
it altogether -OR- white may find something better...so I won't call
it a mate in 17.
I suspect Qe3 wins it for white, but not in 6 (12 plies).
I don't know whether it's the horizon effect from the null move's
reduced search depth that's the problem or some sort of zugzwang
condition. Perhaps the programs that get this one either don't have
the null move or manage it better than those that fail.
-Dan.
PS. I wrote the above a couple of days ago (but didn't manage
to get it posted) and I see in another message that it is (at
worst) mate in 17... Tablebases sure work wonders.
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.