Author: Richard Pijl
Date: 08:57:18 12/19/02
Go up one level in this thread
>POSITION:
>
>[D]1r2q1k1/pnp2ppp/3p4/3N4/Pp6/5N1P/1P1Q1KP1/4R3 b - -
>
>Sorry to leave you waiting a couple of days for my reply.
>
>Well, I thought there was only one main line leading to fireworks at the end,
>but it looks like there are variations. The analysis below is with the help of
>HIARCS 7, by moving back and forth between variations to see what came up.
>
>Let's look at the line I was considering, which may be the most forceful. It
>starts with 24 ... Qxa4, which isn't the best move. There follows
>
>25. Ng5 Rf8
>26. Qd3 f5
>27. Qc4 ...
>
>with at least a mate in 8 following. Black can hold the mate off longest by
>sacrificing the Queen with 27. ... Qc2+. There follows
>
>28. Qxc2 Nc5
>29. Qc4 g6
>30. Nf4+ d5
>31. Qxc5 Rf7
>32. Qxd5 Kg7
>33. Nfe6+ Kh6
>34. Nxf7+ Kh5
>35. Nf4+ Kh4
>36. g3++
>
>which is 24 ply from start to finish, although there might be a shorter mate
>that I didn't find.
>
>The position came up in a game between the shareware program Sigma Chess 5.1.1
>for the Macintosh and my program, at 10 seconds per move. My program, as Black,
>played 24. ... Qxa4, Sigma found the continuation 25. Ng5 and 26. Qd3 and
>proceeded to plaster my program. I was very impressed by its tactical prowess
>but now think it was more of an accident that the checkmate fell into place the
>way it did. Some really strong programs appear to need some time
>(more than 10 seconds) to see that they should avoid 24. ... Qxa4.
>
>In the game itself, my program didn't see the need to sacrifice the Queen
>immediately. There followed
>
>27. ... Kh8
>28. Nf4 Qe8
>29. Rxe8 Rxe8
>30. Nf7+ Kg8
>31. Nh6+ Kh8
>32. Qg8+ Rxg8
>33. Nf7++ !!
>
>With a beautiful smothered mate. Since there is a pattern to the last few moves
>I wondered whether some extensions might be able to handle it (at the time I
>though Sigma had seen through to the end, but I've changed my mind), hence my
>question.
>
>I see that Baron finds the 25. Ng5 Rf8 continuation, but prefers 26. Ne7+ which
>is weaker. Then it changes its mind to 25. Ng5 h6 26. Ne7+ Kf8 which is also
>winning very strongly for White. 25. ... h6 is weak, so I wonder why it switches
>to that move.
As Uri said: Not the whole truth is printed in PV's. So I started Baron again in
the position after 25.Ng5. Baron initially selected 25..Rf8, but got a massive
fail low on it.
ply time nodes score pv
4( 8)- 0 7148 -94 b7d8 d5c7
4( 9)& 0 9004 -178 b7d8 d5c7 d8b7 e1e7
4(10)& 0 20114 -117 b8f8 d5c7 a4d7 d2c2
4(10)& 0 24044 -104 a4b3 d5c7 h7h6 g5f3
4(10). 0 24103 -104 a4b3 d5c7 h7h6 g5f3
5(11)& 0 40740 -106 a4b3 e1e7 c7c6 d5b4 b7c5
5(12). 0 75549 -106 a4b3 e1e7 c7c6 d5b4 b7c5
6(15)& 3 303562 -118 b8f8 e1e7 b7c5 d5c7 c5e6 f2g1
6(15). 4 416268 -118 b8f8 e1e7 b7c5 d5c7 c5e6 f2g1
7(17)- 10 1003846 -148 b8f8 d5e7
7(17)- 18 1803476 -238 b8f8 d5e7
7(17)& 19 1972757 -303 b8f8 d5e7 g8h8 d2f4 a4b3 e7g6 h8g8 g6f8 g8f8
f2f1
7(18). 27 2952351 -303 b8f8 d5e7 g8h8 d2f4 a4b3 e7g6 h8g8 g6f8 g8f8
f2f1
8(18)+ 33 3632654 -273 b8f8
8(18)& 34 3900874 -238 b8f8 d5e7 g8h8 d2f4 a4b3 e7g6 h8g8 g6f8 g8f8
g5h7 f8g8 f4e4 b3b2 f2f3
8(18). 63 7034778 -238 b8f8 d5e7 g8h8 d2f4 a4b3 e7g6 h8g8 g6f8 g8f8
g5h7 f8g8 f4e4 b3b2
9(20)- 133 14646427 -268 b8f8 d5e7
9(20)- 195 21455204 -358 b8f8 d5e7
9(20)& 228 25690886 -542 b8f8 d2d3 g7g6 d5e7 g8g7 d3d4 f7f6 g5e6 g7f7
e6f8
9(21)& 297 34674983 -442 b7c5 d5e7 g8f8 g5h7 f8e8 e7c6 e8d7 c6b8 d7c8
e1e7 c8b8 d2e2
9(21)& 318 38251809 -432 h7h6 d5e7 g8f8 g5h7 f8e8 e7c6 e8d7 c6b8 d7c8
d2e2 b7c5 e2e8 c8b7 f2f1 a4e8 e1e8
Following your line further I decided to force Rf8. But the score got even
worse:
ply time nodes score pv
4( 9)& 0 10253 69 d2d3 f7f5 d5e7 g8h8 g5e6 b7c5
4( 9)& 0 12374 83 d2b4 a4b4 d5b4 b7c5
4( 9)& 0 17323 91 e1e7 b7c5 d5c7 c5e6
4( 9). 0 24522 91 e1e7 b7c5 d5c7 c5e6
5(11)& 0 33494 118 e1e7 b7c5 d5c7 c5e6 f2g1
5(13). 0 79867 118 e1e7 b7c5 d5c7 c5e6 f2g1
6(13)& 1 147784 96 e1e7 h7h6 g5f3 a4b5 d5c7 b5c4
6(15)+ 1 180214 148 d5e7
6(15)+ 2 192667 238 d5e7
6(16)& 2 281438 303 d5e7 g8h8 d2f4 a4b3 e7g6 h8g8 g6f8 g8f8 f2f1
6(16). 3 324563 303 d5e7 g8h8 d2f4 a4b3 e7g6 h8g8 g6f8 g8f8 f2f1
7(18)- 8 1013292 273 d5e7 g8h8
7(18)& 10 1202056 238 d5e7 g8h8 d2f4 a4b3 e7g6 h8g8 g6f8 g8f8 g5h7
f8g8 f4e4 b3b2 f2f3
7(18). 11 1338025 238 d5e7 g8h8 d2f4 a4b3 e7g6 h8g8 g6f8 g8f8 g5h7
f8g8 f4e4 b3b2 f2f3
8(18)+ 12 1500200 268 d5e7
8(19)+ 23 2787322 358 d5e7
8(19)& 26 3198117 457 d5e7 g8h8 d2d3 f7f5 d3c4 a4e8 e7g6 e8g6 g5f7
h8g8 f7e5 g6f7 e5f7 f8f7
8(19)& 39 4703661 524 d2d3 f7f5 d3c4 g8h8 d5f4 a4e8 e1e8 f8e8 g5f7
h8g8
8(19). 39 4767191 524 d2d3 f7f5 d3c4 g8h8 d5f4 a4e8 e1e8 f8e8 g5f7
h8g8
9(20)& 48 6030479 542 d2d3 g7g6 d5e7 g8g7 d3d4 f7f6 g5e6 g7f7 e6f8
a4c2 e1e2 c2c5 d4c5
9(21)+ 66 8418929 554 d5e7
9(21)& 76 9570252 610 d5e7 g8h8 d2d3 f7f5 d3c4 a4e8 e7g6 e8g6 g5f7
h8g8 f7e5 g8h8 e5g6 h7g6 c4c7 b7c5 c7d6
9(21). 77 9854230 610 d5e7 g8h8 d2d3 f7f5 d3c4 a4e8 e7g6 e8g6 g5f7
h8g8 f7e5 g8h8 e5g6 h7g6 c4c7 b7c5 c7d6
10(22)& 127 16473110 610 d5e7 g8h8 d2d3 f7f5 d3c4 a4e8 e7g6 e8g6 g5f7
h8g8 f7e5 g8h8 e5g6 h7g6 c4c7
10(22)& 275 34436324 622 d2d3 g7g6 d3d4 f7f6 d5f6 f8f6 d4f6 a4d7 g2g4
b7c5 e1e7 c5d3 f2e3
10(22). 284 35633597 622 d2d3 g7g6 d3d4 f7f6 d5f6 f8f6 d4f6 a4d7 g2g4
b7c5 e1e7 c5d3 f2e3
11(23)& 371 47479835 622 d2d3 g7g6 d3d4 f7f6 d5f6 f8f6 d4f6 a4d7 g2g4
b7c5 e1e7 c5d3
Again, Baron deviates from your main line with g6, instead of f5.
Forcing f5 gets the following (shocking) result:
ply time nodes score pv
4( 9)+ 0 9151 141 d3c4
4( 9)& 0 9168 113 d3c4
4(10). 0 17126 113 d3c4
5(11)+ 0 19297 143 d3c4
5(13). 0 20763 233 d3c4
6(13)+ 0 28105 263 d3c4
6(16)+ 0 42564 353 d3c4
6(16)& 0 62663 352 d3c4 g8h8
6(16)& 0 108773 621 d5e7 g8h8 d3c4 a4e8 e7g6 e8g6
6(16). 0 119182 621 d5e7 g8h8 d3c4 a4e8 e7g6 e8g6
7(17)+ 2 359514 651 d3c4
7(17). 2 414454 741 d3c4
8(18)+ 7 939715 771 d3c4
8(19)+ 8 1112966 861 d3c4
8(19)& 9 1270258 1087 d3c4 g8h8 d5f4 a4c2 c4c2 b7c5 c2c4 c5e4 g5e4
f5e4
8(19). 11 1533184 1087 d3c4 g8h8 d5f4 a4c2 c4c2 b7c5 c2c4 c5e4 g5e4
f5e4
9(20)+ 29 3900333 1117 d3c4
9(20). 39 5192716 1207 d3c4
10(22)+ 105 14019499 1237 d3c4
10(22)+ 141 18292632 1327 d3c4
10(22)& 828 103828911 32748 d3c4 g8h8 d5f4 g7g6
(MATE in 10 moves)
The reason that no real PV's are given is that Baron doesn't resolve a second
fail high on odd search depths. It rather spends the time doing an additional
ply (and fail high there again ...) so it might be that the mate score could
already be found at search depth 9
>
>I hope you don't think this is overkill in a position where most programs will
>avoid 24. ... Qxa4 anyway, but by taking things apart we might be able to find
>ways to improve our programs' thinking.
>
>So, are there any extensions which could deal with this position, i.e., to allow
>programs to see through to the end after 24. ... Qxa4?
The version I used here only has standard extensions (check, single reply,
recapture, pawn push (to 7th rank), mate threat. Those extensions are limited
when a certain depth is reached.
Seeing through to the end after Qxa4 is not possible with the Baron (at the
moment) as there seems to be better alternatives than the ones you just gave. Or
Baron is mistaken and misevaluating the position.
Bye,
Richard.
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.