Comments by: GM Michal Krasenkow

1.e4 c5 2.Nf3 Nc6 3.d4 c:d4 4.N:d4 Nf6 5.Nc3 e5 6.Ndb5 d6 7.Bg5 a6 8.Na3 b5 9.B:f6 g:f6 10.Nd5 f5 11.Bd3 Be6 12.0–0 B:d5 13.e:d5 Ne7 14.We1 Bg7 15.c3 0–0 16.Qh5 e4 17.Bf1 Re8 18.Rad1 Ng6

WHITE TO MOVE

However frightened you may be to sac a pawn in a computer-assisted game (something your engine - even Rybka - will hardly approve {editor comment: hrrmmm J}), your human evaluation is mostly correct! Black activates his pieces and obtains good chances for a kingside attack. 19.Q:f5 Re5 20.Qh3 Nf4 21.Qg3 Qf6 22.c4 b4 23.Nc2 a5 24.Qe3?! Underestimating Black's reply. [24.Qb3!? was interesting]

BLACK TO MOVE

24...Nd3! 25.B:d3 e:d3 26.Q:d3 R:e1+ 27.R:e1 Q:b2 28.Ne3

BLACK TO MOVE

White is a pawn up but Black's queenside pawns are very dangerous. Such double-edged positions are very difficult both for humans and computers. For humans - because of abundance of complicated variations. For engines - because they can hardly evaluate the strength of passed pawns correctly (unless they see the pawns queening by force, of course). Black has 4 alternatives here:

A) 28…Qc3?! Was played in the game and led to an equal endgame.

B) “Human move" 28...Re8 would be met by 29.Qb3;

C) 28...Q:a2 !? was interesting. We didn't dare to play this move as we couldn't come to a definite conclusion as for the credibility of Rybka's evaluation. It now seems to me that the position remained balanced. I'll not cite all complex variations, here is the one I like most: 29.Nf5 Qb2 30.Qg3

BLACK TO MOVE

30…Rd8 (30...Qc3 31.Nh6+ Kh8 32.N:f7+ Kg8=) 31.Re8+ (31.h3!? a4 32.c5 is a real mess) 31...R:e8 32.Nh6+ Kh8 33.N:f7+ Kg8 34.Nh6+ Kf8 35.Q:d6+ Re7 36.Qd8+ Re8 37.Qd6+=;

D) However, 28...a4! was probably much stronger (perhaps winning). We rejected it because of 29.Nf5 Qc3

WHITE TO MOVE

30.Qe3 (30.Qe4 b3 31.a3 b2 is favorable for Black as 32.N:d6 (32.N:g7 Rb8! 33.Rf1 K:g7 –+) doesn't work due to 32...Qe5! 33.Q:e5 B:e5 34.Ne4 (34.Nb5 Rc8 –+) 34...Rc8 35.c5 f5 –+) , which was evaluated in White's favor by Rybka. It's curious that after 30...Q:e3 31.f:e3 Bc3 (31...b3? 32.N:g7) 32.Rb1 (32.Rc1 b3! 33.a:b3 Bb2) 32...b3 33.a3 she still prefers White, not seeing 33...Bb4! (see diagram below) { editor comment: Rybka to find 33… Bb4 needs 14 ply search, or 22 seconds on top hardware.}.

WHITE TO MOVE (after 33... Bb4!)

(Rybka's evaluation is immediately reversed once this move is shown to her!) 34.a:b4 a3 –+ Obviously, Rybka's calculating algorythm should be adjusted.] Back to the game: 29.Q:c3 B:c3 30.Rb1

BLACK TO MOVE

30… a4?! Missing another chance: [ 30...Rc8! 31.Rd1 (31.Rc1 a4 32.Kf1 a3 33.Ke2 Bb2 34.Rd1 b3 35.a:b3 Rb8 with better position for black; 31.Kf1 Bd2 32.Ke2 B:e3 33.K:e3 R:c4 with a superior rook ending) 31...a4 32.Kf1 Rb8 (32...a3? 33.Rd3; 32...b3?! 33.a:b3 a3 34.Nc2 a2 35.Ke2 Ra8 is something any engine can hardly evaluate.) 33.Ke2 b3 34.a:b3 a3 35.Nc2 R:b3 36.c5 Kf8 keeping winning chances - again too difficult for an engine and too frightening for a human to decide on "against" his engine. After the text move the game safely reaches a harbor of peace. ] 31.Nd1 Bd2 32.Kf1 Re8 33.Ne3 Re4 34.Ke2 B:e3 35.Kd3 Rd4+ 36.K:e3 R:c4

WHITE TO PLAY

Unlike in the endgame arising in the 30...Rc8! line, the b4 pawn is not protected... 37.Kd3 Rc3+ 38.Kd2 Rc4 39.g3 Kg7 40.Rb2 Kf6 41.Kd3 Rc3+ 42.Kd4 a3 43.R:b4 Rc2 44.Rb6 R:f2 45.R:d6+ Ke7 46.Ra6 R:a2 47.h3 Ra1 48.Kc5 Rc1+ ½–½