Author: Uri Blass
Date: 15:54:59 09/17/05
Go up one level in this thread
On September 17, 2005 at 18:47:39, Uri Blass wrote:
>On September 17, 2005 at 18:03:16, Dieter Buerssner wrote:
>
>>On September 16, 2005 at 06:29:42, Thomas Lagershausen wrote:
>>
>>>[D]1r1bnrk1/1qp2ppp/2bp3B/2p2N2/p3P1P1/P4P2/1PPQ3P/2KN2RR w - - 11 20
>>>
>>>20.Bxg7 !! wins in all variations.
>>>
>>>Any programs of today are able to solve this?
>>
>>Some maybe interesting observation for this position, I just discussed with
>>Stefan Knappe ("Matador", he brought my attention to this) . Both Shredder and
>>Fruit WCC did solve it in a similar manner. They both needed some time, and then
>>solved it. They both got a rather huge fail high directly (to 6.x and to 3.x,
>>respectively). At the time we discussed this, I just had forced the moves Bxg7
>>Nxg7 Qh6 in Yace, and after the first fail low, the score went up down
>>continuesly (from black's point of view). I wonder, how jumps in score of Fruit
>>and Shredder can be explained? Is it a typical selective search issue? I cannot
>>imagine, that null move is the reason here, however, in other cases I could
>>imagine such a score jump be caused by null move.
>
>Fruit is using history based pruning when after fail high it does research with
>the original depth.
>
>I believe that it is possible that history based pruning is responsible here.
>It may be interesting to see output of Fruit WCCC without history based pruning.
>
>I thought that part of the difference between Fruit and Yace may be because of
>better evaluation of Fruit of king safety but inspite of the fact that I
>consider yace as worse in king safety evaluation it seems not to be the case in
>this position based on scores at small depths.
>
>
> Some move will be seemingly
>>refuted by a null move (possibly deeper down in the variation) at a much reduced
>>search depth. At some point, the null move will not be able to refute, and now a
>>real search starts. Now we search this line a couple of plies deeper
>>immediately, which could cause a much deeper "understanding" of the position,
>>and such a score jump. What technique are Fruit and Shredder using, that may
>>explain the score jump here?
>>
>>It might be interesting to see analysis different engines for the root position
>>(possibly with lots of patience) compared to the analysis after Bxg7 Nxg7, and
>>after Bxg7 Nxg7 Qh7. Do Shredder and Fruit show similar huge score jumps now?
>>
>>Yace is rather predictable here. Afte Qh6, it shows very small advantage for
>>white at depth 12, somewhat larger advantage at depth 13. With the recapture at
>>the start of the line, one could guess, that at depth 14 (when no other move
>>shows positive score) or at depth 15 in the root position the move will be
>>found. Exactly that happened. Analysis further down.
>>
>>Regards,
>>Dieter
>>
>>From the root position:
>>
>>1r1bnrk1/1qp2ppp/2bp3B/2p2N2/p3P1P1/P4P2/1PPQ3P/2KN2RR w - -
>>
>>Engine: Yace via ssh (700 MB)
>>von Dieter Buerssner
>>
>>14/44 5:20 +0.08 1.Lg5 f6 2.Le3 Tf7 3.Te1 Td7 4.h4 Db5
>> 5.De2 Dxe2 6.Txe2 Lb5 7.Td2 g6 8.Sg3 (288.780.070) 902
>>
>>14/46 7:58 +0.08 1.Lg5 f6 2.Le3 Tf7 3.Te1 Td7 4.h4 Db5
>> 5.De2 Dxe2 6.Txe2 Lb5 7.Td2 g6 8.Sg3 (456.342.873) 954
>>
>>15/46 13:34 +0.07 1.Lg5 Lxg5 2.Dxg5 f6 3.Dh4 Ld7
>> 4.Se7+ Kf7 5.Sd5 Kg8 6.g5 Tf7 7.gxf6 Sxf6
>> 8.Sxf6+ Txf6 (742.157.917) 911
>>
>>15/55 36:30 +0.08++ 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Le8 4.Tg3 Kh8
>> 5.Thg1 c6 6.Sxd6 Lxg5+ 7.Txg5 Td8
>> 8.Sxb7 Sxg5 (2.124.011.250) 969
>>
>>15/55 39:41 +0.48++ 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Le8 4.Tg3 Kh8
>> 5.Thg1 c6 6.Th3 f6 7.g6 Dg7 8.Sxg7 Sxg7 (2.299.314.625)
>>965
>>
>>15/55 39:45 +0.48 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Le8 4.Tg3 (2.304.130.942) 965
>>
>>15/55 40:35 +0.48 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Le8 4.Tg3 (2.349.987.716) 964
>>
>>16/55 44:50 +0.88++ 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Le8 4.Tg3 Kh8
>> 5.Thg1 c6 6.Th3 f6 7.g6 Lxg6 8.Txg6 Dg7
>> 9.Txg7 Sxg7 (2.591.912.344) 963
>>
>>16/55 50:48 +1.36 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Lxe4 4.g6 fxg6
>> 5.Txg6+ Kh8 6.Txe6 Lg5+ 7.Dxg5 Txf5
>> 8.Dg4 Lxc2 9.Te7 Tg8 10.Dh4 Tg7
>> 11.Kxc2 (2.922.985.737) 958
>>
>>16/55 55:47 +1.36 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Lxe4 4.g6 fxg6
>> 5.Txg6+ Kh8 6.Txe6 Lg5+ 7.Dxg5 Txf5
>> 8.Dg4 Lxc2 9.Te7 Tg8 10.Dh4 Tg7
>> 11.Kxc2 (3.208.273.117) 958
>>
>>17/57 70:00 +1.76++ 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Lxe4 4.g6 fxg6
>> 5.Txg6+ Kh8 6.Txe6 Lg5+ 7.Dxg5 Txf5
>> 8.Dg4 Lxc2 9.Tg1 c6 10.Kxc2 Db3+
>> 11.Kb1 Dd3+ 12.Ka2 Te8 13.Txe8+ Tf8 (4.022.260.535) 957
>>
>>17/60 99:16 +2.76++ 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Lxe4 4.g6 fxg6
>> 5.Txg6+ Kh8 6.Txe6 Lg5+ 7.Dxg5 Txf5
>> 8.Dh6 Txf3 9.Te1 Ld5 10.Te8+ Lg8
>> 11.Txg8+ Txg8 12.Te7 Db3 13.cxb3 Txb3 (1.384.992.478) 232
>>
>>17/60 127:12 +3.14 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Lxe4 4.g6 fxg6
>> 5.Txg6+ Kh8 6.Txe6 Lg5+ 7.Dxg5 Txf5
>> 8.Dh6 Txf3 9.Te1 Txa3 10.Df6+ Kg8
>> 11.Tg1+ Lg6 12.Txg6+ hxg6 13.Dxg6+ Kh8
>> 14.Dh6+ (2.935.261.196) 384
>>
>>17/60 136:35 +3.14 1.Lxg7 Sxg7 2.Dh6 Se6 3.g5 Lxe4 4.g6 fxg6
>> 5.Txg6+ Kh8 6.Txe6 Lg5+ 7.Dxg5 Txf5
>> 8.Dh6 Txf3 9.Te1 Txa3 10.Df6+ Kg8
>> 11.Tg1+ Lg6 12.Txg6+ hxg6 13.Dxg6+ Kh8
>> 14.Dh6+ (3.467.742.425) 423
>>
>>And after Bxg7 Nxg7 Qh6; score from black's point of view; hash tables cleared:
>>
>>
>>
>>
>>1r1b1rk1/1qp2pnp/2bp3Q/2p2N2/p3P1P1/P4P2/1PP4P/2KN2RR b - -
>>
>>11/34 0:13 +1.35 2...Se6 3.g5 Te8 4.Dh5 Te7 5.Sxe7+ Lxe7
>> 6.Tg4 Db5 7.Dh6 c4 8.Thg1 c3 (14.228.009) 1041
>>
>>12/37 0:21 +0.95-- 2...Se6 3.g5 Te8 4.g6 hxg6 5.Txg6+ fxg6
>> 6.Dxg6+ Kf8 7.Tg1 Lg5+ 8.Txg5 Lxe4
>> 9.fxe4 Sg7 (22.512.204) 1045
>>
>>12/37 0:32 -0.05-- 2...Se6 3.g5 Te8 4.g6 hxg6 5.Txg6+ fxg6
>> 6.Dxg6+ Kf8 7.Tg1 Lg5+ 8.Txg5 Db4
>> 9.axb4 Sxg5 (33.433.627) 1040
>>
>>12/37 0:33 -0.05 2...Se6 3.g5 Le8 4.Dg7+ Sxg7 5.Sxg7 (34.972.433) 1044
>>
>>12/38 0:39 -0.05 2...Se6 3.g5 Le8 4.Dg7+ Sxg7 5.Sxg7 (41.215.380) 1032
>>
>>13/40 0:57 -0.45-- 2...Se6 3.g5 Le8 4.Tg4 Kh8 5.Thg1 Lxg5+
>> 6.Txg5 Db3 (60.809.596) 1060
>>
>>13/40 1:15 -0.57 2...Se6 3.g5 Lxe4 4.g6 fxg6 5.Txg6+ Kh8
>> 6.Txe6 Lg5+ 7.Dxg5 Txf5 8.Dh6 Txf3
>> 9.Te7 Tg8 10.Te1 d5 11.T1xe4 Tg6
>> 12.Txh7+ Kg8 (80.836.344) 1076
>>
>>13/41 1:26 -0.57 2...Se6 3.g5 Lxe4 4.g6 fxg6 5.Txg6+ Kh8
>> 6.Txe6 Lg5+ 7.Dxg5 Txf5 8.Dh6 Txf3
>> 9.Te7 Tg8 10.Te1 d5 11.T1xe4 Tg6
>> 12.Txh7+ Kg8 (92.184.397) 1069
>>
>>14/47 2:04 -0.97-- 2...Se6 3.g5 Lxe4 4.g6 fxg6 5.Txg6+ Kh8
>> 6.Txe6 Lg5+ 7.Dxg5 Txf5 8.Dg4 Te5
>> 9.fxe4 Dc8 10.Tg1 Txe4 (133.076.466) 1073
>>
>>14/47 2:46 -1.36 2...Se6 3.g5 Lxe4 4.g6 fxg6 5.Txg6+ Kh8
>> 6.Txe6 Lg5+ 7.Dxg5 Txf5 8.Dg4 Lxc2
>> 9.Te7 Tg8 10.Dh4 Tg7 11.Kxc2 Txe7
>> 12.Dxe7 (179.467.154) 1081
>>
>>14/47 3:06 -1.36 2...Se6 3.g5 Lxe4 4.g6 fxg6 5.Txg6+ Kh8
>> 6.Txe6 Lg5+ 7.Dxg5 Txf5 8.Dg4 Lxc2
>> 9.Te7 Tg8 10.Dh4 Tg7 11.Kxc2 Txe7
>> 12.Dxe7 (200.601.272) 1077
>
>I have no Fruit WCCC but here is Fruit2.1's analysis and it also has a jump in
>score unlike yace.
>
>New game,
>[D]1r1b1rk1/1qp2pnp/2bp3Q/2p2N2/p3P1P1/P4P2/1PP4P/2KN2RR b - - 0 1
>
>Analysis by Fruit 2.1:
>
>2...Nxf5 3.gxf5+ Bg5+ 4.Rxg5+ Kh8
> ± (1.39) Depth: 1/5 00:00:00
>2...Ne6
> -+ (-1.78) Depth: 1/8 00:00:00
>2...Ne6 3.g5
> -+ (-1.55) Depth: 2/9 00:00:00
>2...Ne6 3.g5 Qa6
> -+ (-1.80) Depth: 3/9 00:00:00
>2...Ne6 3.g5 Kh8 4.h4
> -+ (-1.62) Depth: 4/10 00:00:00
>2...Ne6 3.g5 Kh8 4.h4 Qa6
> -+ (-1.87) Depth: 5/15 00:00:00
>2...Ne6 3.g5 d5 4.Rg4 dxe4 5.f4
> -+ (-2.52) Depth: 6/18 00:00:00 54kN
>2...Ne6 3.g5 Kh8 4.h4 c4 5.h5 Qa7
> -+ (-1.84) Depth: 7/27 00:00:00 121kN
>2...Ne6 3.g5 Kh8 4.Rg4 Rg8 5.Rhg1 Rg6 6.Qh5
> -+ (-1.83) Depth: 8/27 00:00:00 243kN
>2...Ne6 3.g5 Kh8 4.h4 Qa6 5.Rh2 Rg8 6.Qh5 Rg6
> -+ (-1.93) Depth: 9/32 00:00:01 511kN
>2...Ne6 3.g5 Be8 4.Rg3 Qc8 5.Nde3 Qd7 6.Nd5 Qb5 7.Nf6+ Bxf6 8.Qxf6
> µ (-0.77) Depth: 10/33 00:00:02 1611kN
>2...Ne6 3.g5 Be8 4.Rg3 Qc8 5.Nde3 Qd7 6.Nd5 Kh8 7.Nf6 Bxf6 8.Qxf6+ Kg8
> ³ (-0.67) Depth: 11/37 00:00:03 2754kN
>2...Ne6 3.g5 Bxe4 4.fxe4 Qxe4 5.g6 fxg6 6.Rxg6+ Kh8 7.Rxe6 Qxf5 8.Rg1 Bf6 9.Re7
>Bd4
> µ (-0.75) Depth: 12/40 00:00:11 9151kN
>2...Ne6 3.g5 Bxe4 4.g6 fxg6 5.Rxg6+ Kh8 6.Rxe6 Bg5+ 7.Qxg5 Bxf5 8.Rg1 Rf7 9.Qe3
>Bg6 10.Re8+ Rf8 11.Rxb8 Rxb8
> ³ (-0.26) Depth: 13/44 00:00:21 17574kN
>2...Ne6 3.g5 Bxe4 4.g6 fxg6 5.Rxg6+ Kh8 6.Rxe6 Bg5+ 7.Qxg5 Bxf5 8.Rg1 Rf7 9.Qe3
>Qd5 10.Qc3+ Qd4 11.Qxd4+ cxd4
> = (0.01) Depth: 14/52 00:00:52 43398kN
>2...Ne6 3.g5 Bxe4 4.g6 fxg6 5.Rxg6+ Kh8 6.Rhg1 Bg5+ 7.R1xg5 Bxf5 8.Rxe6 c6 9.Rf6
>Qe7 10.Rfxf5 Rxf5 11.Rxf5
> +- (2.77) Depth: 15/58 00:07:21 365277kN
>
>(, 18.09.2005)
History based pruning does not seem to be the reason
Fruit2.1 without history based pruning need more time to get depth 14 but also
find only score that is close to 0.00 and I do not believe that history based
pruning can help it to see more at depth 15 so probably it is going to fail low
at least to -2.77 without history based pruning.
New game,
1r1b1rk1/1qp2pnp/2bp3Q/2p2N2/p3P1P1/P4P2/1PP4P/2KN2RR b - - 0 1
Analysis by Fruit 2.1:
2...Nxf5 3.gxf5+ Bg5+ 4.Rxg5+ Kh8
± (1.39) Depth: 1/5 00:00:00
2...Ne6
-+ (-1.78) Depth: 1/8 00:00:00
2...Ne6 3.g5
-+ (-1.55) Depth: 2/9 00:00:00
2...Ne6 3.g5 Qa6
-+ (-1.80) Depth: 3/9 00:00:00
2...Ne6 3.g5 Kh8 4.h4
-+ (-1.62) Depth: 4/10 00:00:00
2...Ne6 3.g5 Kh8 4.h4 Qa6
-+ (-1.87) Depth: 5/15 00:00:00
2...Ne6 3.g5 d5 4.Rg4 dxe4 5.f4
-+ (-2.52) Depth: 6/18 00:00:01 64kN
2...Ne6 3.g5 Kh8 4.h4 c4 5.h5 Qa7
-+ (-1.84) Depth: 7/27 00:00:01 146kN
2...Ne6 3.g5 Kh8 4.Rg4 Rg8 5.Rhg1 Rg6 6.Qh5
-+ (-1.83) Depth: 8/27 00:00:01 316kN
2...Ne6 3.g5 Kh8 4.h4 Qa6 5.Rh2 Rg8 6.Qh5 Rg6
-+ (-1.93) Depth: 9/32 00:00:02 630kN
2...Ne6 3.g5 Kh8 4.g6 fxg6 5.Rxg6 Bxe4 6.fxe4 Qxe4 7.Rxe6 Qxf5 8.Rg1
³ (-0.67) Depth: 10/32 00:00:03 2069kN
2...Ne6 3.g5 Be8 4.Rg4 Qb5 5.Kb1 Qd7 6.Nc3 Kh8 7.Nd5 c4
µ (-0.87) Depth: 11/36 00:00:09 4849kN
2...Ne6 3.g5 Bxe4 4.g6 fxg6 5.Rxg6+ Kh8 6.Rxe6 Bg5+ 7.Qxg5 Rxf5 8.Qe3 Bxf3
9.Qc3+ Re5 10.Rxe5 Bxh1
³ (-0.56) Depth: 12/47 00:00:26 19188kN
2...Ne6 3.g5 Bxe4 4.g6 fxg6 5.Rxg6+ Kh8 6.Rxe6 Bg5+ 7.Qxg5 Bxf5 8.Rg1 Rf7 9.Qe3
Bg6 10.Re8+ Rf8 11.Rxb8 Rxb8
³ (-0.26) Depth: 13/47 00:00:43 32888kN
2...Ne6 3.g5 Bxe4 4.g6 fxg6 5.Rxg6+ Kh8 6.Rhg1 Bg5+ 7.R1xg5 Rxf5 8.Rg4 Rf4
9.fxe4 c4 10.Rg3
= (0.02) Depth: 14/50 00:02:29 116396kN
(, 18.09.2005)
Uri
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.