It took me some time for the point of the sac to sink in. The advent of 'e' P on to 'f' file is the fulcrum of this game.

For one thing, he might not have resigned. In the final position ...

click for larger view

Instead of tipping his king, black could (should) have played 25...Rg8. Black has to weather a bit of a storm but it should fizzle out into a fairly level position.

Fritzie offers this line: 25... Rg8 26. fxe7 f6 27. Rxh5 Qxe7 28. Qxh7+ Qxh7 29. Bxf6+ Rgg7 30. Bxg7+ Kxg7 31. Rxh7+ Kxh7

click for larger view

And it's one of those horribly difficult to win rook and pawn endings. They are all supposed to be drawn but I manage to lose an awful lot of them ;-)

I wonder why players do this. Nothing for Black was vaguely tricky to see. In fact the hardest move is probably White's Qxh7+ to avoid difficulties. I can't believe that a 2385 player just thinks - 'well there are a lot of threats so i'll just resign without calculating them'.

click for larger view

White threatens Rxh5 followed by Bf6+. It all looks pretty horrible and unstoppable. The black queen cannot get to the defence of the kingside, g7 looks terribly weak and the Nh5 is "pinned" against the mate threat on h7.

If it was white to move his attack would indeed be impossible to defend against.

But it's black to move and he has just one move that saves him. 26...f6 blocks the bishop check. At first glance this move looks impossible because the only thing protecting the f6 pawn is the "pinned" Nh5. But the knight is not pinned if it can land on f6 without being captured, because from f6 it protects against Qxh7#

So I think that black looked at these variations and missed 26...f6. He didn't realise that the Nh5 which was pinned in most variations was not pinned in this one.

In situations like this, I tend to apply the patzer principle. Only resign when the mate is so absolutely crystal clear that the rawest patzer in the room can understand why you are resigning.

If I were black I would have played 25...Rg8 as the only move that doesn't lose instantly and then thought about it again after white had moved. And hopefully I might have spotted 26...f6, again as the only move which doesn't lose immediately.

Houdini 1.5a agrees with both Fritz 12 and Rybka 4, evaluating the position after 25...Rg8 26.fxe7 f6 as a miniscule advantage for White, [+0.11], d=30.

click for larger view

Houdini agrees with Fritz on the best moves for both sides through 27.Rxh5 Qxe7 28.Qxh7+ Qxh7 29.Bxf6+ Rgg7 30.Bxg7+ Kxg7 31.Rxh7+ Kxh7

click for larger view

The rest of Houdini's PV goes as follows: 32.h4 Rb6 33.Kh2 Kh6 (I would think that 33...Kf6 centralizing the king would be better, although it probably makes no difference) 34.Kh3 (likewise 34.Kg3) 34...g5 (I don't see why Houdini as Black wants to help White establish a passed pawn on the k-side) 35.Rd3 a5 36.g3 gxh4 (36...a4 immediately seems more to the point) 37.Kxh4 Kg6 38.g4 a4 39.g5 e4 40.Rd2 Kf7 41.Kg4 Rb3 42.Rxd6

click for larger view

Now I'm no longer sure who stands better; Houdini has managed to unbalance the position so that both sides have winning chances with White's passed g-pawn and Black's soon to be passed a-pawn. Restarting Houdini's analysis from this position results in an eval of [+0.28], d=30 after either 42...Ke7 or 42...Rxa3, and [+0.29] after 42...e3.

But it's White who stands substantially better. Houdini's top 3 lines go as follows:

(1) 42...Ke7 43.Ra6 Rxa3 44.Kf5 Rf3+ 45.Kxe4 Rxf2 46.g6 Rg2 47.Kf5 Rf2+ 48.Kg5 Ra2 (maybe 48...Rg2+?) 49.Ra7+ Kf8 50.Rf7+ Kg8 51.Rb7 (maybe a critical loss of tempo, 52.Ra7 immediately looks better) 51...a3 52.Ra7 Rg2+ 53.Kf6 Rf2+ 54.Ke6 Rd2 (Black must also cut off White's king from capturing his c-pawn) 55.Rxa3 Rd4 56.Rc3 Kg7 57.Kf5 Rd6 58.Rg3 Rd1 (58...Rd4 seems more direct since 59.Rg4 Rxg4 is a draw per the 6-man Nalimov tablebases, but Houdini only has access to 5-man Gaviota tablebases so maybe it couldn't see this) 59.Ra3 Rf1+ 60.Ke4 Kxg6, and this is a mate in 33 for White per the Nalimov tablebases.

click for larger view

(2) 42....Rxa3 43.Kf5 Rf3+ 44.Kxe4 Rxf2 45.Ra6 Rd2 46.Kf5 (a slight waste of time since Black won't allow White's king to help support the passed g-pawn) 46...Rf2+ 47.Ke5 Rg2 48.Ra7+ Kg6 49.Rxa4 Rxg5+ and this is a mate for White in 44 moves per the Nalimov tablebases.

click for larger view

(3) 42...e3 43.Rf6+ Kg7 44.f4 (!) Rxa3 45.Re6 Rd3 46.Kf3 e2+ 47.Kxe2 Rc3 48.Ra6 a3 (48...Rxc4 looks hopeless for Black since his rook is then stuck guarding his 2 passed pawns and White's king can support his passed pawns by 49.Kf3 and 50.Kg4) 49.Kf2 (this seems like a waste of time, 49.f5 immediately accomplishes the same thing) 49...Rh3 50.Kg2 Re3 51.f5 Re5 52.Ra7+ Kg8 53.Kf3 Rxf5+ 54.Kg4 Re5 55.Kf4 Re2 56.Rxa3 Rc2 57.Ra8+ (some more waste of time at best, why allow Black an additional move to bring his king closer to the action?) 57...Kg7 58.Ra4 Kg6 59.Ke5 Rf2 (59...Rd2 cutting off White's king from the c-pawn doesn't work since after 60.Ra5 Kxg5 it's a mate for White in 33 moves per the Nalimov tablebases) 60.Ra5 Rf8 61.Rxc5 and this is a mate for White in 29 moves per the Nalimov tablebases.

click for larger view

None of this, of course, is forced, but merely an indication of how the game could go if Black had not resigned after 25.f6, particularly since IMO Houdini is not the best chess engine endgame player. It's clear that there was still a lot of fight left in the position, and Black's resignation was premature as far as practical chances. But perhaps understandable, given that Black had been under attack since 19.Rh3 and who knows what the time control situation was. I suspect, as <Once> said, that Black simply may not have had the time to look at all the consequences of 25...Rg8 26.fxe7, particularly after evaluating many continuations with the knight "pinned" against the mate threat on h7.