Neat position: Rybka claims win but it seems a draw

[Thomas Mayer]

forget it, choose life

Well, I analyzed a bit down with Quark, concentrating on a line like
Rxg3 f4 Rg1 Kf5

it seems that in this line black will always lose the f-pawn. Anyway, I wonder whether the resulting 6-men is still a draw. Can anyone try that out with a full set of 6-men ?

Greets, Thomas

hehe, touché...

[Dann Corbit]

Dan check this position after your line up to Kg8

[d] 5k2/1R6/7P/5b2/8/5p2/4pK2/8 b - - 0 10


1. Rxg3 Ke5 2. Ke1 f4 3. Rg5+ Kf6 4. Rg1 f3 5. h4 Bc2 6. Kf2 Be4 7. h5 Kf7 8.h6 Kf8 9. Rg7 Bf5 10. Rb7 e1=Q+ 11. Kxe1 f2+ 12. Kxf2 Kg8 *

Yes, it looks drawn. Not only that, Rybka sees the draw, but he doesn't know it:

Code: Select all

Analysis from C:\test\MAYBE.EPD   
10/29/2009 7:25:11 PM Level: 600 Seconds
Analyzing engine: Rybka 3

1)                      
    Avoid move: 
    Best move (Rybka 3): Kf8-g8
    Not found in: 10:00
      2	00:00	         222	13.372	-2.24	Kf8g8
      3	00:00	         434	26.142	-2.75	Kf8g8
      4	00:00	         541	32.587	-2.74	Kf8g8
      5+	00:00	         830	49.995	-2.54	Kf8g8
      5+	00:00	         935	56.320	-2.34	Kf8g8
      5	00:00	       1.004	60.476	-2.64	Kf8g8 Rb7e7
      6	00:00	       1.863	59.616	-2.84	Kf8g8 Rb7e7 Bf5d3 Kf2xf3
      7	00:00	       4.347	139.104	-2.66	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Kg8f8 Re7e3
      8	00:00	       4.422	70.752	-2.73	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Kg8f8 Re7e3 Kf8g8 Kf3xe2 Kg8h7
      9	00:00	       5.807	54.057	-2.94	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Kg8f8 Re7e3 Kf8g8 Kf3xe2 Kg8h7 Re3h3 Bc2f5
     10	00:00	       6.047	56.292	-2.94	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Kg8f8 Re7e3 Kf8g8 Kf3xe2 Kg8h7 Re3h3 Bc2f5
     11	00:00	       6.686	43.608	-2.94	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Kg8f8 Re7e3 Kf8g8 Kf3xe2 Kg8h7 Re3h3 Bc2f5
     12	00:00	       8.004	40.176	-3.08	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Kg8f8 Re7e3 Kf8g8 Kf3xe2 Kg8h7 Re3h3 Bc2f5 Rh3h4 Bf5e6 Ke2e3 Be6b3 Rh4h3
     13	00:00	      22.544	35.190	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     14	00:00	      22.549	35.198	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     15	00:01	      24.270	37.884	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     16	00:01	      28.059	39.961	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     17	00:01	      33.046	37.350	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     18	00:01	      42.082	38.856	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     19	00:01	      57.143	40.748	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     20	00:02	      84.496	45.443	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     21	00:03	     123.070	49.228	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     22	00:05	     202.336	38.829	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     23	00:06	     292.107	45.862	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     24	00:08	     494.310	60.079	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     25	00:11	     746.684	71.438	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     26	00:14	   1.147.795	81.711	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     27	00:21	   1.978.209	96.465	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     28	00:30	   3.345.774	112.911	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     29	00:41	   5.082.656	125.561	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     30	00:54	   7.109.979	133.532	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     31	01:26	  12.006.562	143.266	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     32	01:56	  18.993.403	167.819	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3
     33	02:45	  30.217.538	187.174	-3.43	Kf8g8 Rb7e7 Bf5c2 Kf2xf3 Bc2d3 Kf3f2 Bd3c4 Kf2e1 Bc4d3 Re7d7 Bd3c4 Rd7c7 Bc4d3

[Michel]

[d] 5k2/1R6/7P/5b2/8/5p2/4pK2/8 b - - 0 10


RobboLito seems to give a similar analysis with a similar large negative score for black.

info time 234472 nodes 67423093 nps 287000 score cp -307 depth 26 seldepth 46 pv f8g8 b7e7 f5c2 f2f3 c2d3 f3f2 g8h8 e7g7 d3b5 f2e1 b5c4 g7e7 c4d3 e7c7 d3b5 c7a7 b5c4 a7g7 c4b5 e1f2 b5c4 g7c7 c4b5 f2e1 b5d3 c7e7 d3b5 e7f7 b5d3 f7b7 d3c4 b7a7 c4b5 a7b7 b5d3 b7e7

(this is on a slow computer).

One would think this is a bug. Where could this large negative possibly come from?

[Thomas Mayer]

[d] 5k2/1R6/7P/5b2/8/5p2/4pK2/8 b - - 0 10


RobboLito seems to give a similar analysis with a similar large negative score for black.

info time 234472 nodes 67423093 nps 287000 score cp -307 depth 26 seldepth 46 pv f8g8 b7e7 f5c2 f2f3 c2d3 f3f2 g8h8 e7g7 d3b5 f2e1 b5c4 g7e7 c4d3 e7c7 d3b5 c7a7 b5c4 a7g7 c4b5 e1f2 b5c4 g7c7 c4b5 f2e1 b5d3 c7e7 d3b5 e7f7 b5d3 f7b7 d3c4 b7a7 c4b5 a7b7 b5d3 b7e7

(this is on a slow computer).

One would think this is a bug. Where could this large negative possibly come from?

well, thats simply static evaluation, most of it piece scores. In all given lines the f-pawn got captured by white, therefor you have something rook+pawn against bishop+pawn. Now it depends on the engine how it scores that. One of the possibilities to solve such things is to have some kind of pattern recognition, e.g. when king of the stronger side must stop pawn of the weaker side and king of the weaker side can walk ahead of the pawn (clearly some more patterns needed) of the stronger side etc. then lower score to draw, someting like this. As far as I know older programs are full of stuff like this, but after all it seems that this all makes you just slower without increasing your strength, so it seems newer "better" programs know less and lesser about such exceptions.

Greets, Thomas

[Uri Blass]

[d] 5k2/1R6/7P/5b2/8/5p2/4pK2/8 b - - 0 10


RobboLito seems to give a similar analysis with a similar large negative score for black.

info time 234472 nodes 67423093 nps 287000 score cp -307 depth 26 seldepth 46 pv f8g8 b7e7 f5c2 f2f3 c2d3 f3f2 g8h8 e7g7 d3b5 f2e1 b5c4 g7e7 c4d3 e7c7 d3b5 c7a7 b5c4 a7g7 c4b5 e1f2 b5c4 g7c7 c4b5 f2e1 b5d3 c7e7 d3b5 e7f7 b5d3 f7b7 d3c4 b7a7 c4b5 a7b7 b5d3 b7e7

(this is on a slow computer).

One would think this is a bug. Where could this large negative possibly come from?

well, thats simply static evaluation, most of it piece scores. In all given lines the f-pawn got captured by white, therefor you have something rook+pawn against bishop+pawn. Now it depends on the engine how it scores that. One of the possibilities to solve such things is to have some kind of pattern recognition, e.g. when king of the stronger side must stop pawn of the weaker side and king of the weaker side can walk ahead of the pawn (clearly some more patterns needed) of the stronger side etc. then lower score to draw, someting like this. As far as I know older programs are full of stuff like this, but after all it seems that this all makes you just slower without increasing your strength, so it seems newer "better" programs know less and lesser about such exceptions.

Greets, Thomas

The problem here is that even if white get KRP vs KB it is still a draw based on tablebases and the reason is that KR+ pawn at h6 against king and bishop is usually a draw if the bishop and the king control h7.

I do not believe that the reason for not having endgame knowledge about simple endgames is that knowledge does the program weaker and I guess that the simple reason is that in order to do it right you need a lot of work for a small improvement and it may be better to work for bigger improvements.

Uri

[Dann Corbit]

[d] 5k2/1R6/7P/5b2/8/5p2/4pK2/8 b - - 0 10


RobboLito seems to give a similar analysis with a similar large negative score for black.

info time 234472 nodes 67423093 nps 287000 score cp -307 depth 26 seldepth 46 pv f8g8 b7e7 f5c2 f2f3 c2d3 f3f2 g8h8 e7g7 d3b5 f2e1 b5c4 g7e7 c4d3 e7c7 d3b5 c7a7 b5c4 a7g7 c4b5 e1f2 b5c4 g7c7 c4b5 f2e1 b5d3 c7e7 d3b5 e7f7 b5d3 f7b7 d3c4 b7a7 c4b5 a7b7 b5d3 b7e7

(this is on a slow computer).

One would think this is a bug. Where could this large negative possibly come from?

Not a bug at all. There is a large material advantage, but the program cannot figure out how to exploit it. Quite possibly (but not certainly) there is no way to exploit it.

When you see ten deep plies of analysis in a row where the score does not change, know that it is probably a draw (that is what I meant when I said Rybka found that it was a draw but did not realize it).

Actually, the analysis given is *exactly* what I would want the analysis to be.

To me it says the following:

"I have a big material advantage but I cannot see any way to make progress."

Sometimes, I can figure out a way to make progress by myself (though often I can't). I cannot think of any score that a program could give that would be superior to the score given by Rybka.

P.S.
It may or may not be a draw. We do not know for sure.

[adieguez]

Not a bug at all. There is a large material advantage, but the program cannot figure out how to exploit it. Quite possibly (but not certainly) there is no way to exploit it.

When you see ten deep plies of analysis in a row where the score does not change, know that it is probably a draw (that is what I meant when I said Rybka found that it was a draw but did not realize it).

Actually, the analysis given is *exactly* what I would want the analysis to be.

To me it says the following:

"I have a big material advantage but I cannot see any way to make progress."

Sometimes, I can figure out a way to make progress by myself (though often I can't). I cannot think of any score that a program could give that would be superior to the score given by Rybka.

P.S.
It may or may not be a draw. We do not know for sure.

This may not be the root position and you may have no time to check too many lines. I understand the ideal could be to have an option to disable non-trivial endgame knowledge, or tablebases and some draw recognizers at once.

Anyway, I have no tablebases. You neither? I guess 6 men would help to solve it with more confidence. Altough it really looks like a draw and already made amyan to score it a bit faster as a draw. (yes in a previous post i said i was going to remove amyan source from my hd but didn't ) would be great if it's not a draw someone corrects.

[Mark]

[d] 5k2/1R6/7P/5b2/8/5p2/4pK2/8 b - - 0 10


RobboLito seems to give a similar analysis with a similar large negative score for black.

info time 234472 nodes 67423093 nps 287000 score cp -307 depth 26 seldepth 46 pv f8g8 b7e7 f5c2 f2f3 c2d3 f3f2 g8h8 e7g7 d3b5 f2e1 b5c4 g7e7 c4d3 e7c7 d3b5 c7a7 b5c4 a7g7 c4b5 e1f2 b5c4 g7c7 c4b5 f2e1 b5d3 c7e7 d3b5 e7f7 b5d3 f7b7 d3c4 b7a7 c4b5 a7b7 b5d3 b7e7

(this is on a slow computer).

One would think this is a bug. Where could this large negative possibly come from?

Not a bug at all. There is a large material advantage, but the program cannot figure out how to exploit it. Quite possibly (but not certainly) there is no way to exploit it.

When you see ten deep plies of analysis in a row where the score does not change, know that it is probably a draw (that is what I meant when I said Rybka found that it was a draw but did not realize it).

Actually, the analysis given is *exactly* what I would want the analysis to be.

To me it says the following:

"I have a big material advantage but I cannot see any way to make progress."

Sometimes, I can figure out a way to make progress by myself (though often I can't). I cannot think of any score that a program could give that would be superior to the score given by Rybka.

P.S.
It may or may not be a draw. We do not know for sure.

With tablebases, Rybka sees the draw for the above position pretty fast:

[+0.00] d=22 10...e1Q+ 11.Kxe1 (0:00:01) 12kN
[+0.00] d=21 10...e1Q+ 11.Kxe1 (0:00:00) 8kN
[+0.00] d=20 10...e1Q+ 11.Kxe1 (0:00:00) 7kN
[+0.00] d=19 10...e1Q+ 11.Kxe1 (0:00:00) 5kN
[+0.00] d=18 10...e1Q+ 11.Kxe1 (0:00:00) 5kN
[+0.00] d=17 10...e1Q+ 11.Kxe1 (0:00:00) 4kN
[+0.00] d=16 10...e1Q+ 11.Kxe1 (0:00:00) 4kN
[+1.40] d=7 10...Kg8 11.Re7 (0:00:00) 2kN
[+1.87] d=7 10...Kg8 (0:00:00) 1kN
[+2.27] d=7 10...Kg8 (0:00:00) 0kN
[+2.47] d=7 10...Kg8 (0:00:00) 0kN
[+2.67] d=6 10...Kg8 11.Re7 (0:00:00) 0kN
[+2.47] d=5 10...Kg8 11.Re7 (0:00:00) 0kN
[+2.74] d=4 10...Kg8 (0:00:00) 0kN
[+2.75] d=3 10...Kg8 (0:00:00) 0kN
[+2.75] d=2 10...Kg8 (0:00:00) 0kN