# Studies that have been cooked by computers

A study is unlike a regular chess problem. Instead of "mate in 3" (for instance), a study's instructions are something like "White to draw" (for instance). What this means is to find a line of play after which White reaches a position that is known to be a draw. But there is a problem with this notion because computers have shown us that many situations that were "known to be a draw" aren't really (same for theoretical wins).

Are there any famous studies that have been cooked (solution is not valid) due to a computer showing that the final position did not satisfy the conditions of the study?

(By famous I mean either a classic that appears in many books or a study that won prizes in problem tournaments.)

• I distinctly recall that many moves in older books have been cooked. All responsibly produced chess books now have been checked by computer. Most of the time the old books are right. But know that they were written in a different era with different standards. Sep 21 '13 at 12:20
• There is another issue you don't mention in your post. The solution to a study is supposed to be unique. Computers have sometimes found alternative solutions, making the study unsound. Sep 22 '13 at 13:50

I don't know how famous it is, but here is a study shown to be incorrect by computers:

`````` [Title "J. van Reek, 1987. White to play and draw."]
[FEN "8/5n2/5N2/3K4/8/3p4/Rn6/1k6 w - - 0 1"]

1. Ra3 d2 2. Rd3 Nxd3 3. Ne4 d1=N! 4. Ke6 Nd8+ (4... Nh6!) 5. Kd7 Nb7 6. Kc6 Na5+ 7. Kb5 Nb3 8. Nc3+ Nxc3+ 9. Kc4 Kc2 1/2-1/2
``````

The intended solution is given in the diagram. It has a very nice stalemate finish.

Unfortunately black can play `4... Nh6!`, and according to the tablebases there is a win for black in 39 moves.

So three knights can win against one. Having to play this ending would give me nightmares!

• The ending is similar to that of Kubbel & Herbstmann, 1st Prize, Troitzky Tournament 1937; Nunn, Solving in Style, 121. 8/8/8/7n/8/7N/3kp1K1/5n2 Draw. 1 Ng1! Ne3+ 2 Kh3! Nf4+ 3 Kh2! Ng4+ 4 Kh1! Nf2+ 5 Kh2 e1=N 6 Nf3+ Nxf3+ 7 Kg3 Ke3 stalemate. Aug 7 '17 at 8:38
• I have lost the source. Does anyone know where this was published? May 3 at 14:01
• @DagOskarMadsen I can't find any matches in YACPDB or the Die Schwalbe PDB. May 3 at 14:32

If a study contains only a small number of pieces it is always possible to check for accuracy. (Right now the number is six, with seven in the works.) See for example: http://www.k4it.de/?topic=egtb&lang=en

Here is a study that has been busted. Perhaps it is not a famous one (it seems not to be in PDB or YACPDB). Tim Krabbé featured it in entry 376 of Open Chess Diary. Tim reported that Olli Heimo busted it. The annotations are Tim's.

``````[Title "B. Badai, Shakhmaty v SSSR, 1965. Draw"]
[fen "5k2/8/7P/6pP/8/8/8/4K1n1 w - - 0 0"]

1.Kf2 Nh3+ 2.Kf3! (2.Kg3? Nf4 3.Kg4 Ne6 4.Kf5 Kf7 {and White is in zugzwang and must allow...} 5.Ke5 g4 6.Ke4 Kg8) 2...Nf4
3.Ke4! Ne6! {bust found by Olli Heimo} (3... Kf7? {the intended mainline} 4.Ke5! {Again, not 4.Kf5 Ne6 and White is in Zugzwang.} 4...Ne6 (4...Kg8 5.Kf5 {is the same}) 5.Kf5 {And now Black is in zugzwang.} 5...Kg8 6.Kxe6 g4 7.Kf5 (7.Kf6?? Kh7) g3 8.Kg6 g2 9.h7+ Kh8 10.Kh6 g1=Q =)
4.Ke5 Nd8! {and after 5... Nf7 Black wins the struggle to keep his last pawn.}
``````

Two corrections are involved in the following very famous study. Its complete history, including how it was adapted from the conclusion of a game, are beyond the scope of this answer, but may be read on web pages from Wikipedia and the archive of Tim Krabbé's Open Chess Diary. When first published in study form, it was a draw-study as follows:

``````[Title "G.E. Barbier. Glasgow Weekly Citizen, 4 May 1895. Black to play and draw."]
[StartFlipped "0"]
[fen "8/2P5/1K6/3r4/8/8/8/k7 b - - 0 0"]

1... Rd6+ 2. Kb5 Rd5+ 3. Kb4 Rd4+ 4. Kb3 Rd3+ 5. Kc2 Rd4! 6. c8=Q Rc4+ 7. Qxc4 =
``````

Rev. Saavedra read the above solution, given by Barbier the following week. He then cooked Barbier's draw-study, showing how White could force a win:

``````[Title "G.E. Barbier, corr. Rev. Saavedra. Glasgow Weekly Citizen, 18 May 1895. Black to play and White to win."]
[StartFlipped "0"]
[fen "8/2P5/1K6/3r4/8/8/8/k7 b - - 0 0"]

1... Rd6+ 2. Kb5 Rd5+ 3. Kb4 Rd4+ 4. Kb3 Rd3+ 5. Kc2 Rd4! 6. c8=R Ra4 7. Kb3 {and wins}
``````

(The study is usually given in the version by Lasker; White's pawn is moved to c6, the stipulation is changed to "White to play and win", and there is an extra move 1. c7.)

Saavedra's move is beautiful in being a rook-promotion which, in a position with only four units, is the only winning move. And he transformed a draw-study into a win-study which is beautiful in that it puts White in that position and forces that promotion.

Except ... it doesn't. Tablebases have shown that White can indeed force a win, but Black avoids the Saavedra position, and White can win only by promoting to queen. At move 3, Black's best defence is not `3. ... Rd4+` but `3. ... Kb2` (deferring mate for 3 more moves) giving White only one winning move: `4. c8=Q!`. If Black does play `3. ... Rd4+` then White's best attack, supposing that Black always plays the best defence, is either `4. Kb3 Rd3 5. Kc2! Rf3 6. c8=Q!` or `4. Kc3 Rd1 5. Kc2 Rf1 6. c8=Q!`.

• Well, that's not a cooked study. Saavedra merely omits a line which would be preferred by computers (but is still winning for White), not by humans. May 30 '16 at 13:04