Test Position

[Daniel]

Geschrieben von: / Posted by: Daniel at 19 January 2004 17:57:22:

Hi
8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74


This is a position that occurs in a game b/n DanChess and Aristarch.
DanChess(white) thinks it's down by more than -7 here and resigned.
I think this position is a draw after Rxe3!.I tested also aristarch(as white) at this position,it also resigned.
Can any engine see the draw here?
regards
Daniel

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 19 January 2004 18:22:14:
Als Antwort auf: / In reply to: Test Position geschrieben von: / posted by: Daniel at 19 January 2004 17:57:22:

Hi Daniel,

8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74

This is a position that occurs in a game b/n DanChess and Aristarch.
DanChess(white) thinks it's down by more than -7 here and resigned.
I think this position is a draw after Rxe3!.I tested also aristarch(as
white) at this position,it also resigned.
Can any engine see the draw here?

Well, I agree with you, after Rxe3 it looks very drawish...
the only engine I found that seems to have an idea that it might be a draw is Little Goliath 2000 v3.9 -> Afaik Little Goliath has some special stuff to see draws by continued checks. (Or how "Dauerschach" is called in english)
Analysis:

[Anthony Cozzie]

Geschrieben von: / Posted by: Anthony Cozzie at 19 January 2004 18:23:00:
Als Antwort auf: / In reply to: Test Position geschrieben von: / posted by: Daniel at 19 January 2004 17:57:22:

Hi
8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74


This is a position that occurs in a game b/n DanChess and Aristarch.
DanChess(white) thinks it's down by more than -7 here and resigned.
I think this position is a draw after Rxe3!.I tested also aristarch(as white) at this position,it also resigned.
Can any engine see the draw here?
regards
Daniel

What is your drawing line?
Rxe3 dxe3 Rh3+ Kf2 and white loses since Rxe3 is met by Rh4 and black mates.
This is all human analysis, so it may contain (many) errors.
anthony

[Tord Romstad]

Geschrieben von: / Posted by: Tord Romstad at 19 January 2004 18:44:00:
Als Antwort auf: / In reply to: Re: Test Position geschrieben von: / posted by: Anthony Cozzie at 19 January 2004 18:23:00:

Hi
8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74


This is a position that occurs in a game b/n DanChess and Aristarch.
DanChess(white) thinks it's down by more than -7 here and resigned.
I think this position is a draw after Rxe3!.I tested also aristarch(as white) at this position,it also resigned.
Can any engine see the draw here?
regards
Daniel

What is your drawing line?
Rxe3 dxe3 Rh3+ Kf2 and white loses since Rxe3 is met by Rh4 and black mates.
This is all human analysis, so it may contain (many) errors.

Yes, but what if white plays Rf3+ instead of Rxe3? I don't see how black can
escape the checks.
This is also human analysis, and is equally likely to contain errors.
Tord

[Tord Romstad]

Geschrieben von: / Posted by: Tord Romstad at 19 January 2004 19:08:01:
Als Antwort auf: / In reply to: Test Position geschrieben von: / posted by: Daniel at 19 January 2004 17:57:22:

Hi
8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74


This is a position that occurs in a game b/n DanChess and Aristarch.
DanChess(white) thinks it's down by more than -7 here and resigned.
I think this position is a draw after Rxe3!.I tested also aristarch(as white) at this position,it also resigned.
Can any engine see the draw here?

Mine can't.
It's funny that I recieved a game from Leo today where Gothmog crashed in
a position similar in nature to the one above. The position was this
(sorry, I don't know how to make a nice graphical board):
7k/5B1p/1P2R1pP/1P4P/r2K1P/// w
My maximal search depth is 60, and the size of my PV[] array is also 60.
I always thought this meant that there were no risk of array-out-of-bound
errors. I overlooked that I could theoretically end up with PVs longer
than 60 plies, because I build my PVs from the hash table.
The above position (which occured in a game Gothmog-Ktulu) made me
discover that this risk is not only theoretical. Gothmog produced
a 60+ move PV, and instantly crashed. For a long time, I wondered
how it could be possible to obtain such a long PV from such a shallow
search (the crash occured at the 18th iteration). Then I suddenly
realized that the search tree would probably contain nodes where
the white rook and black where located virtually anywhere on the board,
and that these nodes would be stored in the hash table with a best
move.
The bug will of course be fixed in the next version of Gothmog,
which will also include some evaluation bugfixes, some new
positional knowledge, a better qsearch, "lazy" null move searches,
and some new, cool features. It will also be marginally less
clueless in the endgame.
Tord

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 19 January 2004 19:14:35:
Als Antwort auf: / In reply to: Re: Test Position geschrieben von: / posted by: Tord Romstad at 19 January 2004 18:44:00:

Hi Tord,
8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74

Can any engine see the draw here?

What is your drawing line?
Rxe3 dxe3 Rh3+ Kf2 and white loses since Rxe3 is met by Rh4 and black mates.
This is all human analysis, so it may contain (many) errors.

Yes, but what if white plays Rf3+ instead of Rxe3? I don't see how black can
escape the checks.
This is also human analysis, and is equally likely to contain errors.

but your analysis seems to be correct - I did try several engines to get a win against another engine after Rxe3... none did win... and I did try myself and draw quite easy with white... but you must be careful that the black rooks can not walk between king and the white rook. I did also try to win with black against engines and it was not possible - even when some show -10 at some point... (e.g. Patzer)
Greets, Thomas

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 19 January 2004 19:59:39:
Als Antwort auf: / In reply to: Re: Test Position geschrieben von: / posted by: Anthony Cozzie at 19 January 2004 18:23:00:

8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74

What is your drawing line?
Rxe3 dxe3 Rh3+ Kf2 and white loses since Rxe3 is met by Rh4 and black mates.

[Event "Computer chess game"]
[Site "ARNOLD7"]
[Date "2004.01.19"]
[Round "-"]
[White "Quark v2.29beta"]
[Black "Yace Paderborn"]
[Result "1/2-1/2"]
[TimeControl "1800+5"]
[FEN "8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 1"]
[SetUp "1"]
{--------------
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . R
. . . . . p r .
. . R . p . k .
. . . . . . r .
. . . . . . . K
white to play
--------------}
1. Rxe3+ fxe3 2. Rh3+ Kf2 3. Rf3+ Ke1 4. Rf1+ Kd2 5. Rf2+ Kc3 6. Rc2+ Kb4
7. Rb2+ Kc4 8. Rc2+ Kb5 9. Rc5+ Kb4 10. Rb5+ Kc4 11. Rb4+ Kd5 12. Rb5+ Ke4
13. Re5+ Kf3 14. Rxe3+ Kf4 15. Re4+ Kg5 16. Re5+ Kg6 17. Re6+ Kh7 18. Rh6+
Kxh6
{Stalemate} 1/2-1/2
of course Yace could have avoided the stalemate for some more moves - then it would end in a fifty-move-draw...
Greets, Thomas

[Dieter Bürßner]

Geschrieben von: / Posted by: Dieter Bürßner at 20 January 2004 00:16:46:
Als Antwort auf: / In reply to: Test Position geschrieben von: / posted by: Daniel at 19 January 2004 17:57:22:

8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74

Interesting!
Yace PB, with TBs and "bitbases", P4, 2.53 GHz, 200 M hash (may be a bit lucky here):

[   nodes   time    score depth]   
   8178764   8.863  -5.77 17t  74.Rxe3+ fxe3 75.Rh3+ Kf2H 76.Rf3+H Ke1H
                               77.Rf1+H Kd2H 78.Rd1+H Kc3H 79.Rd3+H Kb4H
                               80.Rb3+H Ka4H 81.Ra3+H Kb5H 82.Ra5+H Kc4H
                               83.Rc5+H Kb4H 84.Rb5+H Ka4H 85.Ra5+H Kb3H
                               86.Ra3+H Kb2H 87.Ra2+H Kc3H 88.Rc2+H Kd4H
                               89.Rb2H {HT}
   9165228  10.403  -5.77 17.  74.Rxe3+ fxe3 75.Rh3+ Kf2 76.Rf3+ Ke1 77.Rf1+
                               Kd2 78.Rd1+ Kc3 79.Rd3+ Kb4 80.Rb3+ Ka4 81.Ra3+
                               Kb5 82.Ra5+ Kc4 83.Rc5+ Kb4 84.Rb5+ Ka4 85.Ra5+
                               Kb3 86.Ra3+ Kb2 87.Ra2+ Kc3 88.Rc2+ Kd4 89.Rb2
   9396354  10.645  -5.37 18++ 74.Rxe3+ fxe3 75.Rh3+ Kf2 76.Rf3+ Ke1 77.Rf1+
                               Kd2 78.Rd1+ Kc3 79.Rd3+ Kb4 80.Rb3+ Ka4 81.Ra3+
                               Kb5 82.Ra5+ Kc4 83.Rc5+ Kb4 84.Rb5+ Ka4 85.Ra5+
                               Kb3 86.Ra3+ Kb2 87.Ra2+ Kc3 88.Rc2+ Kb4 89.Rb2+
                               Ka5 90.Rxg2 Rg3
   9708394  10.925   0.00 18++ 74.Rxe3+ fxe3 75.Rh3+ Kf2 76.Rf3+ Ke1 77.Rf1+
                               Kd2 78.Rd1+ Kc3 79.Rd3+ Kb4 80.Rb3+ Ka4 81.Ra3+
                               Kb5 82.Ra5+ Kc4 83.Rc5+ Kb4 84.Rb5+ Ka4 85.Ra5+
                               Kb3 86.Ra3+ Kb2 87.Rb3+ Kxb3 {-1080}
   9774036  10.991   0.00 18t  74.Rxe3+ fxe3 75.Rh3+ Kf2H 76.Rf3+H Ke1H
                               77.Rf1+H Kd2H 78.Rd1+H Kc3H 79.Rd3+H Kb4H
                               80.Rb3+H Ka4H 81.Ra3+H Kb5H 82.Ra5+H Kc4H
                               83.Rc5+H Kb4H 84.Rb5+H Ka4H 85.Ra5+H Kb3H
                               86.Ra3+H Kb2H 87.Rb3+H Kxb3H {-1080}
  12324183  15.098   0.00 18.  74.Rxe3+ fxe3 75.Rh3+ Kf2 76.Rf3+ Ke1 77.Rf1+
                               Kd2 78.Rd1+ Kc3 79.Rd3+ Kb4 80.Rb3+ Ka4 81.Ra3+
                               Kb5 82.Ra5+ Kc4 83.Rc5+ Kb4 84.Rb5+ Ka4 85.Ra5+
                              Kb3 86.Ra3+ Kb2 87.Rb3+ Kxb3 {-1080}
  15816541  18.482   0.00 19t  74.Rxe3+ fxe3 75.Rh3+ Kf2 76.Rf3+ Ke1 77.Rf1+
                               Kd2 78.Rd1+ Kc3 79.Rd3+ Kb4 80.Rb3+ Ka4 81.Ra3+
                               Kb5 82.Ra5+ Kc4 83.Rc5+ Kb4 84.Rb5+ Ka4 85.Ra5+
                               Kb3 86.Ra3+ Kb2 87.Rb3+ Kxb3 {-1080}

Regards,
Dieter

[N.K.]

Geschrieben von: / Posted by: N.K. at 20 January 2004 06:20:28:
Als Antwort auf: / In reply to: Re: Test Position geschrieben von: / posted by: Tord Romstad at 19 January 2004 19:08:01:

The position was this
(sorry, I don't know how to make a nice graphical board):
7k/5B1p/1P2R1pP/1P4P/r2K1P/// w

It seems that Yace understands this (short) variant of FEN notation,
but Ufim does not.
I wonder which programs generate short FENs?
Where can I read about this (short) standard?

[Peter Fendrich]

Geschrieben von: / Posted by: Peter Fendrich at 20 January 2004 12:18:00:
Als Antwort auf: / In reply to: A question about FEN geschrieben von: / posted by: N.K. at 20 January 2004 06:20:28:

The position was this
(sorry, I don't know how to make a nice graphical board):
7k/5B1p/1P2R1pP/1P4P/r2K1P/// w

It seems that Yace understands this (short) variant of FEN notation,
but Ufim does not.
I wonder which programs generate short FENs?
Where can I read about this (short) standard?

I don't think it's a documented standard. It's a short hand FEN that some programmers implemented.
/Peter

[Daniel]

Geschrieben von: / Posted by: Daniel at 20 January 2004 14:39:57:
Als Antwort auf: / In reply to: Re: Test Position geschrieben von: / posted by: Tord Romstad at 19 January 2004 19:08:01:

Hi
8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74


This is a position that occurs in a game b/n DanChess and Aristarch.
DanChess(white) thinks it's down by more than -7 here and resigned.
I think this position is a draw after Rxe3!.I tested also aristarch(as white) at this position,it also resigned.
Can any engine see the draw here?

Mine can't.
It's funny that I recieved a game from Leo today where Gothmog crashed in
a position similar in nature to the one above. The position was this
(sorry, I don't know how to make a nice graphical board):
7k/5B1p/1P2R1pP/1P4P/r2K1P/// w
My maximal search depth is 60, and the size of my PV[] array is also 60.
I always thought this meant that there were no risk of array-out-of-bound
errors. I overlooked that I could theoretically end up with PVs longer
than 60 plies, because I build my PVs from the hash table.
The above position (which occured in a game Gothmog-Ktulu) made me
discover that this risk is not only theoretical. Gothmog produced
a 60+ move PV, and instantly crashed. For a long time, I wondered
how it could be possible to obtain such a long PV from such a shallow
search (the crash occured at the 18th iteration). Then I suddenly
realized that the search tree would probably contain nodes where
the white rook and black where located virtually anywhere on the board,
and that these nodes would be stored in the hash table with a best
move.
The bug will of course be fixed in the next version of Gothmog,
which will also include some evaluation bugfixes, some new
positional knowledge, a better qsearch, "lazy" null move searches,
and some new, cool features. It will also be marginally less
clueless in the endgame.
Tord

Hi
Gothmog 0.4.5 as white doesn't crash at this position.May be it is another one.
Though I extract PV's from hash tables,I don't need to store it in a pV[] array.
Just extract and print the PV when needed or return at MAXPLY-1 in the search.
Good luck!
regards
Daniel

[Peter Fendrich]

Geschrieben von: / Posted by: Peter Fendrich at 20 January 2004 15:15:43:
Als Antwort auf: / In reply to: Test Position geschrieben von: / posted by: Daniel at 19 January 2004 17:57:22:

Hi
8/8/8/7R/5pr1/2R1p1k1/6r1/7K w - - 0 74


This is a position that occurs in a game b/n DanChess and Aristarch.
DanChess(white) thinks it's down by more than -7 here and resigned.
I think this position is a draw after Rxe3!.I tested also aristarch(as white) at this position,it also resigned.
Can any engine see the draw here?
regards
Daniel

Terra doesn't find anything without TB's but with TB's on a Dell laptop 2.0 GHz:

Terra 3.3B4. Start Analyze
 7      0.56   -735  135428 Rh8 Kf2 Rc2+ e2 Rxe2+ Kxe2 Re8+ Kd3 Rd8+ Ke4 Rd7
 7      0.61   -627  158557 Ra3 f3 Rxe3 Kf2 Rxf3+ Kxf3 Rf5+ Kg3 Rf3+ Kh4 Rf7
 7      0.72   -621  205109 Rxe3+ fxe3 Rh3+ Kf4 Rf3+ Ke4 Rxe3+ Kf4 Re1 Kg3 Re7
 8      0.85   -621  249786 Rxe3+ fxe3 Rh3+ Kf4 Rf3+ Ke4 Rxe3+ Kd4 Rd3+ Ke5
                            Re3+ Kf4 Re1  (Kf4g3 Re1e7)
 9      1.05   -621  325884 Rxe3+ fxe3 Rh3+ Kf4 Rf3+ Ke4 Rxe3+ Kd4 Rd3+ Ke5
                            Re3+ Kf4 Re1  (Rg2f2 Re1e7)
10      1.74   -576  652245 Rxe3+ fxe3 Rh3+ Kf4 Rf3+ {H} (Kf4f3)
11      3.40   -576 1472041 Rxe3+ fxe3 Rh3+ Kf4 Rf3+ Ke4 Rxe3+ Kf5 Re5+ Kf4
                            Re4+  (Kf4e4)
12      4.46    -38 2009834 Rxe3+ fxe3 Rh3+ Kf2 Rf3+ Ke1 Rf1+ Kd2 Rf2+ e2
                            Rxe2+ Kc3 Rxg2 Re4 Rg8 Re1+ Kh2 Re2+ Kh3 Kd4
13      5.00    -37 2287645 Rxe3+ fxe3 Rh3+ Kf2 Rf3+ Ke1 Rf1+ Kd2 Rf2+ e2
                            Rxe2+ Kd3 Rxg2 Re4 Rg8 Re2 Rd8+ Kc4 Rc8+ Kd5 Rc7
14      6.02    -30 2859848 Rxe3+ fxe3 Rh3+ Kf2 Rf3+ Ke1 Rf1+ Kd2 Rf2+ e2
                            Rxe2+ Kd3 Rxg2 Re4 Rg7 Re2 Rd7+ Kc4 Kg1 Kb4 Rc7
15      7.94    -22 3968719 Rxe3+ fxe3 Rh3+ Kf2 Rf3+ Ke1 Rf1+ Kd2 Rf2+ e2
                            Rxe2+ Kd3 Rxg2 Re4 Kh2 Rh4+ Kg3 Rh1 Rf2 Rg1+
                            Rg2 Re1 Rf2
16     12.34    -22 6511499 Rxe3+ fxe3 Rh3+ Kf2 Rf3+ Ke1 Rf1+ Kd2 Rf2+ e2
                            Rxe2+ Kd3 Rxg2 Re4 Kh2 Rh4+ Kg3 Rh1 Ra2 Rg1+
                            Kf3 Rf1+ Rf2 Rg1 Rg2
17     21.27   -15 11847140 Rxe3+ fxe3 Rh3+ Kf4 Rf3+ Kg5 Rf5+ Kg6 Rg5+ Kh6
                            Rg6+ Kh7 Rg7+ Kh8 Rg8+
18     43.23   -15 24418467 Rxe3+ fxe3 Rh3+ Kf4 Rf3+ Kg5 Rf5+ Kg6 Rg5+ Kh6
                            Rg6+ Kh7 Rg7+ Kh8 Rg8+

/Peter

[Ron Murawski]

Geschrieben von: / Posted by: Ron Murawski at 20 January 2004 23:21:30:
Als Antwort auf: / In reply to: Re: A question about FEN geschrieben von: / posted by: Peter Fendrich at 20 January 2004 12:18:00:

The position was this
(sorry, I don't know how to make a nice graphical board):
7k/5B1p/1P2R1pP/1P4P/r2K1P/// w

It seems that Yace understands this (short) variant of FEN notation,
but Ufim does not.
I wonder which programs generate short FENs?
Where can I read about this (short) standard?

I don't think it's a documented standard. It's a short hand FEN that some programmers implemented.
/Peter

I've never seen short forms documented either.
My engine assumes no pieces on a rank, no ep square, and no castling allowed. Half-moves and full-moves are assumed to be zero. If a field is missing from the FEN the defaults are held.
This is the only description of FEN that I know:
http://www.very-best.de/pgn-spec.htm#16.1 (link is below)
It is part of the PGN Specification
Ron



FEN Notation