Htydjtk

I played a game recently on chess.com, where the most detailed analysis identified a single blunder in the game -- 23. d6 The analysis given by the computer suggested that 23 Qd1 was better, with the line it gave then chasing the black knight around the board until it was captured, and freeing the Black queen rook in the process.

Myself and my opponent both like 23. d6 (well, I liked it more than he did) as it locks out the entire Black queenside for the rest of the game. From my perspective I'm a rook and bishop up at this point, which makes it hard to me to see this as a blunder.

Could anyone perhaps point out better reasons why I should regard this move as a blunder?

[fen ""]

[Event "Let's Play!"]

[Site "Chess.com"]

[Date "2019.02.19"]

[Round "-"]

[White "postmortes"]

[Black "Glebbf"]

[Result "1-0"]

[WhiteElo "1341"]

[BlackElo "859"]

[TimeControl "1/86400"]

[EndDate "2019.02.21"]

[Termination "postmortes won by checkmate"]



1. g3 e5 2. Bg2 Nf6 3. Nc3 Bd6 4. e3 Na6 5. Nge2 O-O 6. O-O c6 7. a3 Qc7 8. b4

Re8 9. d3 c5 10. Re1 h5 11. b5 Nb8 12. Rb1 Re6 13. Bd2 c4 14. a4 cxd3 15. cxd3

a6 16. b6 Qc5 17. d4 exd4 18. exd4 Qf5 19. Nf4 Bxf4 20. Bxf4 Nc6 21. d5 Rxe1+

22. Qxe1 Nd4 23. d6 h4 24. Ne4 h3 25. Nxf6+ Qxf6 26. Qe8+ Kh7 27. Be4+ Nf5 28.

Qe7 Qg6 29. Qe5 Kg8 30. Bxf5 f6 31. Qd5+ Qf7 32. Qxf7+ Kxf7 33. a5 g6 34. Bxh3

f5 35. Re1 Kf6 36. Bf1 Kf7 37. Bc4+ Kf6 38. h4 Kg7 39. Re7+ Kf8 40. Bh6# 1-0

This may be the horizon effect at work.

I analyzed the position a bit with Lichess' Stockfish, which at depth ~20 likes 23. Rd1 (+8.0~8.5) even more than 23. Qd1 (+7.0~7.5). The idea is the same: White catches Black's knight, for an immediate material advantage (e.g. 23. Qd1 Nc2 24. Rb2 Na3 25. Qb3).

After 23. d6 on the other hand, Stockfish still gives a solid advantage of +4.0~5.0 (no human would call that "blunder" since it's still clearly winning) at depth ~20, but the following lines are rather unforcing.

A likely explanation is that in either variation there are so many defensive possibilites left (there's a lot of open space to manoeuvre!) that Stockfish is unable to find a concrete way to force mate or even only increase White's advantage substantially within its search horizon. It does of course realize that White has the much better position, which it prices at around +5.0~6.0, but that's it. In the Qd1/Rd1 line, it then simply adds the material advantage of +2.0~3.0 of that missing knight. However, it won't do the same for the queenside rook and bishop in the d6 line since it has no way to tell that these pieces are buried forever (that would require it to go deeper in its search). The engine cannot exclude that these pieces will become active again at depth 50 or something, so it still counts them in. We as humans do of course know better (one of the last cases we do).

Case in point: While writing this, the evaluation of 23. d6 increased to +6.5 at depth 30. It will most likely converge with the one of 23. Qd1 / 23. Rd1 at some point.

From a human point of view, you should not view this move as a blunder.

Update:

Depth 32 starting from the position after 22...Nd4:

1 23. Rd1 +8.4

2 23. Qd1 +7.7

3 23. Qe3 +7.1

4 23. d6 +6.9

5 ...

We're getting closer.

I wouldn't call your move a blunder, it seems just as good as winning the Black knight. Stockfish puts you at roughly +2.5 with 23.d6, while winning the knight with 23.Qd1 yields roughly +5.5. Honestly not such a huge difference - at such extreme evaluations you should generally prefer the move that's easier to play.

It's easier for engines to evaluate being a clean piece up, rather than a few of your opponent's pieces being indefinitely trapped. This is a well known flaw in the top well known engines (Stockfish, Komodo, Houdini, etc). They're very good at searching, but their evaluation functions aren't perfect.