Constrictor

Initial version published in MatPlus MP 35 as No. 1354 PDF article

Final version published in Glarean (Ref. 5)

One side of the story (author’s side)

The idea of creating tough puzzles for chess-engines was there for a long time, but only at the beginning of 2009 was I able to put together (in a puzzle) such concepts as long-term strategy, fortress and zugzwang, all of which are difficult for computers to comprehend. This became the initial version of the endgame study.
In winter 2008-2009, Matplus agreed to publish my previous 3 puzzles (of type “white to move and mate in N”, MP 33-34/09, puzzles 1229, 1230 and 1233 – ref.1), so I naturally sent the initial version of this endgame study there as well. After a short mail exchange with editor Iuri Akobia (ref.2), in which I showed him 2 winning variation that were previously evaluated by his chess computer program as dead draws, he agreed to publish it, and in September it appeared in the autumn issue of the magazine (MP 35/09 puzzle 1354 – ref.1).
Chess is fun, and this study was my humble attempt to spread the fun of chess (and mock chess engines J ). So in the meantime, I opened topics about this study (well, the initial version of it) in various chess forums, mostly without telling the solution, but mentioning that chess engines (badly) misevaluate the position long time into the solution. However, in none of these forum was the study thoroughly analysed, until Jonas took it from Schachmatt.de (ref.3) and moved it to the Rybkaforum, where…

The other side of the story (Peter’ side)

…I found a thread in Rybkaforum, titled “Can Rybka solve this?” (ref.4). A member brought it here from a German chess forum, where the author showed the starting position, but did not give any solution or hints.
For several days, members of Rybkaforum could only guess what this position was about, but then someone found a breakthrough which got the discussions going: “Rybka came up with an amazing position where the queen was helplessly out of the game. After that the real fight starts and the question becomes: Must White accept a draw or can he win? If you are a strong player or a good engine analyst, you will enjoy this challenge!” The debate continued, questioning whether the solution starting as above was correct (leading to win for white) and unique, as computer programs had absolutely no idea what to do with the resulting position. After another week, fascinated by the improvability of the solution by computers, and after I already started looking for the author in the German forum where the issue came from, MoKy found in Rybkaforum a well-hidden refutation (Black held the draw, even if it was by a single tempo and in a totally un-computer-like playing style). I didn’t want to accept this was it, the author did not give any sign yet, so I started slightly modifying the initial position, in hope I would find a version where the drawing variation wouldn’t work – and I did find it!
In neither forum was another refutation, but the variation tree was still huge. Then the author finally posted another message in the German forum; he simply had not checked his messages there for a long time, and he only learnt about the discussion in the Rybkaforum from my posts in Schachmatt.de. He first requested some time to analyse the disputed variations, and half a day later, he thanked everybody for their efforts and acknowledged the refutation of the initial version of the study, as well as the correctness of my suggestion.
Initially, the black pawn was on a7, and in my suggestion it ended up on a5, but he pushed it back to a6, which I also see as the clearest (and aesthetically better) way of presenting the study. Computer programs still cannot see through the position, and only after showing them the first approximately 10 moves in each variation (depending on software, hardware and analysis time), they start evaluating the position as a win for White. We’ll probably have to wait for another generation of chess software to see machines reliably solve such positions… 

The drawings show:

• The initial version of the study, published in Matplus;
• The final (corrected) version, published in Glarean (ref.5);
• Peter’s version with pawn on a5, which is a little less complicated than the final version;
• A computer-friendly version, which is within the grasping capacity of the machines

Special thanks to:
Peterhiarcs, whose perseverance made this story possible
Jonas, who opened the discussion in Rybkaforum
MoKy, who found the refutation in the initial position
Rybkaforum members, for their enthusiastic analysis and great help
Zibbit and Dennis, the first rated players who expressed encouraging words about the study
Iuri Akobia and Walter Eigenmann, who agreed to publish the study in Matplus / Glarean respectively

References:
Ref.1 Matplus home page: http://www.matplus.net/pub/start.php
Ref.2 Iuri Akobia on Wiki: http://en.wikipedia.org/wiki/Iuri_Akobia
Ref.3 Schachmatt.de: http://www.schachmatt.de/69-schachraetsel/4240-schachraetsel-gesucht-bei-dem-selbst-schachprogramme-versagen.html#post47091
Ref.4 Rybkaforum: http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=12580
Ref.5 Glarean article:  http://glareanverlag.wordpress.com/2009/10/01/studie_glarean_martan_neghina/