A funny tablebase cook of a Pogosyants study

[yrobinso]

Almost always, tablebases play the endgame perfectly, giving a definitive result for each position. However, tablebases do not account for the castling rule because it is a little complicated to include it, and it is not really necessary for practical purposes. Nevertheless, I was amused to find this study, which seemed to expose this weakness of the tablebase:

http://www.gadycosteff.com/eg/eg56.pdf#page=5

"Castling in Studies"
E. Pogosyants
EG 56, June 1978, page 5
FEN: 8/8/7r/3N4/8/8/7P/R3K2k

8/8/7r/3N4/8/8/7P/R3K2k


The composer's solution is 1. Ne3 Rxh2 2. O-O-O mate! Side variations are 1...Kxh2 2. Ng4+ and 1...Re6 2. Kf2+ Kxh2 3. Ng4+ Kh3 4. Kf3 Kh4 5. Kf4 and either 5...Kh3 6. Ra2 or 5...Kh5 6. Kf5.

Naturally, the tablebase thinks that 1. Ne3 Rxh2 is a draw because it assumes that White cannot castle. However, the tablebase gets the last laugh: after 1. h4!! White mates in 33 moves. Black's best defense is 1...Kg2 2. Ne3+ Kf3 3. Ra3! Re6 4. Kf1 Kg3 5. Ng2 etc. Of course, Black can force an ending of R+N vs. R, but that doesn't save him either: 1...Rxh4 2. Kf2+! Kh2 3. Ra6 Kh3 4. Kf3! Rh7 5. Nf4+ Kh2 6. Kf2 Rh8 7. Rf6 Rh4 8. Rf5 Rh6 9. Ng6 Kh1 10. Nh4 Ra6 (10...Rxh4 is no better) 11. Kg3 Ra3+ 12. Nf3 Ra5! (sets a stalemate trap if White takes the rook) 13. Rf4 Ra4! 14. Nd4 Ra3+ 15. Kf2 Ra2+ 16. Ne2! and White mates shortly.

It is beside the point that Pogosyants's study can be corrected by placing a Black pawn on g4. This adds an insignificant side variation 1. Ne3 g3 2. Ke2+ Kxh2 3. Ng4+, but otherwise the solution does not change.

The point is that there is an endgame study with two solutions. The human (Pogosyants) could not find the tablebase solution, and the tablebase ( http://www.k4it.de/index.php?topic=egtb&lang=en ) could not find the human solution! This situation may be unique, or almost unique, in the history of chess. Can anyone find another example?

[guyhaw]

Great work: delightful find.
You should relay your cook (and repair) of the study to the editor of the EG (Studies) Quarterly's observation column, Jarl Henning Ulrichsen.
EGs 001-152 are available in .pdf form from the ARVES site, and can be scanned in one folder - so it's not necessary to consolidate the 152 files into one as I did.
Would be worth searching for 'O-O', '0-0', 'O-O-O' and '0-0-0' for similar.
Congratulations again - g

[guyhaw]

A search on 'O-O' mainly yields irrelevant matches like "co-operation" but does come up with:
- #2527 V.Evreinov; A.Selesniev on 'p553' of some volume;
- #13762 I.Jarmonov; #13865 (a sideline); #13877 J.Timman (lots of men)

A search on '0-0' produces 216 matches, most relevant, apart from the odd ISBN or telephone number.

Neither search comes up with the cited study by Pogosyants in his 'Castling in Studies' article in EG 056 - because he has "Castles" rather than 0-0-0! Study #256 by van Emde allows for castling on either side, a possibly unique composition.
g

[guyhaw]

http://www.arves.org/English/index.htm

URL does not change - - so use menu on left:
--> "EndgameStudies", scroll to the foot, Item 9, "Index on subject EG", "downloaded the EG.pdf", "EG Archive", d'load what you want

g

[guyhaw]

http://www.xs4all.nl/~timkr/chess2/diary.htm ... Item 324

g