Why is this the solution to this puzzle?

Question

[FEN "2r3k1/5pp1/1q1bpn1p/pB1pN3/P2P4/5R2/1P3PPP/3Q2K1 w - - 0 1"]

1. Rxf6 Qc7 2. Rf3

This is a puzzle (and its solution) I found on lichess.

But I simply do not see the expected logic of this puzzle. The Qc7 move doesn't do much from my POV, and refraining from following up with gxf6 also doesn't make sense as that move wins a rook, and while it does open the king up to check, that check doesn't force any other captures, nor does it lead to mate. In fact, when I analyse the game with an engine, it even prefers gxf6 over Qc7.

Answer 1

There is a lot more to this, and this is really not a great tactics problem as much as it is a continuing the attack problem. The difference is that with a tactic problem, you can see the result all the way to the end. Other positions, you have just a feel for it, and you know that you are probably winning, but you have to play what you can calculate, you get there, and then you calculate more of the attack. Often, you do this while calculating places you could take a forced draw if it is not working out.

A great example of that is the game Dubov-Svane (here with copyrighted annotations), which I will give below. It was one of the best games of the year.

The move Qc7 is an example of why I am not a fan of online puzzles that have been computer checked, but no strong player has evaluated. After Rf6, if you do not take, you simply lose, and thus, Qc7 is not even an option to consider. The only lines that are important are the ones that include gf, but more importantly, the ones that include Bxe5, which comes closer to defending for a human than any other move.

 [FEN "2r3k1/5pp1/1q1bpn1p/pB1pN3/P2P4/5R2/1P3PPP/3Q2K1 w - - 0 1"]

 1. Rxf6 Bxe5 {This may be the most testing try in practice.} (1... gxf6 2. Qg4+ Kh7 (2... Kh8 3. Nxf7+ Kh7 4. g3 {The threat is Qe6.} Bb8 5. Qh4 Kg7 6. Qxh6+ Kxf7 7. Qh7+ Kf8 8. Qh8+ Kf7 9. Qxc8 $18) (2... Kf8 $2 3. Nd7+ Ke7 4. Nxb6 Rc1+ 5. Bf1 $18 {This is obviously important to see.}) 3. Bd3+ f5 4. Bxf5+ exf5 5. Qxf5+ Kh8 6. Qf6+ Kh7 7. Qxf7+ Kh8 8. Ng6#) 2. dxe5 gxf6 3. exf6 Kf8 (3... Kh8 4. Qd2 Kh7 5. Bd3+ Kg8 6. Qxh6 $18) (3... Kh7 4. Qd3+ Kh8 5. Qd2 Kh7 6. Bd3+ Kg8 7. Qxh6 $18) 4. Qd2 Rc6 5. Qxh6+ Ke8 6. g3 d4 {It is clear that that is heading to a queend ending, so maybe d4 would give some counter-chances under slightly more favorable circumstances.} 7. Qh7 d3 8. Qg8+ Kd7 9. Qxf7+ Kc8 10. Qg8+ Kb7 (10... Kc7 11. Qg7+ Kb8 (11... Kd6 12. Qe7+ Ke5 13. Bxc6 Qxc6 14. f7 d2 15. Qg5+ Ke4 16. Qe3+ Kd5 17. Qxd2+ $18) 12. Bxc6 Qxc6 13. f7 $18) 11. Bxc6+ Qxc6 12. Qh7+ Kb6 13. Qxd3 {There are other lines in here, but this line represents a moves that try to fight back until it is finally over totally.}

Here is that fantastic attack my Danil Dubov, which illustrates my point. Dubov most likely did not see everything from move 22 through to move 39, but rather, he saw chunks at a time, and there were probably places he could have bailed out with a draw. This type of attack, while containing tactics, is not one big tactic. I would describe the problem above in the same way, assuming that black played the best practical defense with 1...Bxe5 trying to keep equal material while trading off the key Ne5 attacker.

 [Event "EU-chT 22nd"]
 [Site "Batumi"]
 [Date "2019.10.31"]
 [Round "7.4"]
 [White "Dubov, Daniil"]
 [Black "Svane, Rasmus"]
 [Result "1-0"]
 [ECO "D37"]
 [WhiteElo "2699"]
 [BlackElo "2592"]
 [PlyCount "77"]
 [EventDate "2019.10.24"]
 [EventType "team-swiss"]
 [EventRounds "9"]
 [EventCountry "GEO"]
 [WhiteTeam "Russia"]
 [BlackTeam "Germany"]
 [WhiteTeamCountry "RUS"]
 [BlackTeamCountry "GER"]
 [FEN ""]

 1. c4 e6 2. Nc3 d5 3. d4 Nf6 4. Nf3 Be7 5. Bf4 O-O 6. e3 b6 7. Qc2 Ba6 8. O-O-O dxc4 9. Ng5 Nc6 10. a3 g6 11. h4 Bd6 12. g3 Qe7 13. h5 e5 14. hxg6 hxg6 15. Bg2 exf4 16. Bxc6 fxg3 17. Kb1 Rad8 18. f4 Bc8 19. Rde1 Kg7 20. Nd5 Nxd5 21. Rh7+ Kg8 22. Rxf7 Rxf7 23. Qxg6+ Kf8 24. Qh6+ Rg7 25. Bxd5 Ke8 26. Qh5+ Kd7 27. Qh3+ Ke8 28. Qh5+ Kd7 29. Be6+ Kc6 30. Qf3+ Kb5 31. Bxc4+ Ka5 32. Qd5+ Bc5 33. b4+ Ka4 34. Qg2 Bxb4 35. Qc6+ Kxa3 36. Bb3 Bd7 37. Qc1+ Kxb3 38. Qc2+ Ka3 39. Qa2# 1-0

P.S. The only "logic" to it is that someone, who was not that strong fed it to a computer, and posted only the computer's best moves, but not the ones that would challenge a human the most.

Answer 2

I'm going to assume you meant ...Qc7 instead of ...Qc6, since the latter move blunders the queen.

After 1...gxf6 2.Qg4+, Black has three moves:

1) 2...Kf8 loses the queen to 3.Nd7+.

2) 2...Kh7 leads to mate after 3.Bd3+ f5 4.Bxf5+ exf5 5.Qxf5+ Kh8 (5...Kg7 6.Qxf7+ Kh8 7.Ng6#) 6.Qf6+ Kg8 7.Qxf7+ Kh8 8.Ng6#.

3) 2...Kh8 is Black's best move. After 3.Nxf7+ Kh7, White must play 4.g3 to keep his advantage. In this position White has a knight and pawn for a rook, but Stockfish evaluates as over +4. White's immediate threat is to play Qh5-Qxh6+, but there's also some ideas with Bd7-Bxe6 or Qxe6. You can analyse this position further with your engine if you want.

Stockfish thinks 1...Qc7 is very slightly a better move than 1...gxf6, but it basically evaluates them at the same (both are +4 for White). Here such small differences in evaluation aren't that relevant. In a practical game with humans of course 1...gxf6 should be preferred. In the line after 1...Qc7 2.Rf3 Bxe5 3.dxe5 Qxe5, White is up a bishop for a pawn.

By the way, I'm curious about the level of this problem. For line 3 I gave, some calculation after the position I stopped at is still needed.

Answer 3

It wins a piece. If you take the rook black gets mated in two more moves else he wins the queen too if you try to avoid the mate.