Checking out the position for a minute one notices the following features:

1/ Black can sac a piece on e3 and get three pawns for the piece.

2/ White's pieces have generally been drawn away from the defence of his king.

3/ The black queen can jump into the attack along the d8-h4 diagonal.

4/ The white queen is committed to the defence of the Na4, and can be further drawn away from the defence of his king by Rxa4.

All of these features suggest a violent sac on e3. My first thought was Nxe3 because I wanted the rook to end up there, building up the pressure on the pinned Ne3. The queen is then over-committed defending the two knights. But on further investigation, this line leads nowhere.

So let's look at:

22...Rxd3
23. fxe3 Nxe3
24. Qb3 (to maintain defence of the Na4)
24...Bxd4 (aha! now we have a discovered attack looming against the white king) 25. Kh1 (I think this is forced)

So far so easy - or at least, not too difficult. But now we're in a situation where black has a whole bunch of candidate moves, and none of them look immediately decisive. Eg: Qg5 Qh4 Nxd1 Rxa4.

Of these, I don't like Nxd1 because I don't think black should be trying to recover material at this point. The initiative is much more important.

The best option looks like Qg5, since black can follow up with Be4. However, white has (for example) Ne1, defending g2 and allowing the Bb1 to neutralise the black Bf5.

So let's look at 25....Qh4. Now we see the possibility of mating on h2 after 26...Ng4 or 26...Bxd3 and 27...Be5.

We also have the possibility of Rxa4 at some point, drawing away the white queen.

White also has a bunch of candidate moves, and this is getting to the limit of my ability to analyse positions in my head, so I'm going with this:

22...Rxd3
23.fxe3 Nxe3
24.Qb3 Bxd4
25.Kh1 Qh4
26.(any)

when black follows by 26...Ng4 or 26...Qf2 or (most probably) 26...Bxd3 and 27...Be5 depending on white's 26th.

But what is with 24. Qd2. I see no clear win for Black afterwards.

Neither 24...Nxd1 is convincing (see my first post) nor 24..Rxa4 25. Nc5 Nxd1 26. Nxa4.

I found also no good continuation after 24...Bxd4 25. Ndc5 Qg5 26. Kh1

24. Qd2 Nxd1 25. Qxd1 Bxd4+ 26. Kh1 Be3 (now the white queen is overworked and the knight on a4 has to run) 27. Nc3 Bxc1 28. Qxc1

click for larger view

I tried to make this line work, but I have no glue if there is any mistake or better defence:

24. Qd2 Bxd4 25. Ndc5 Qg5 26. Kh1 Nxd1 27. Qxd1 Be3! 28. Rc3 d4 29. Rb3 b5! and this seems to be enough advantage for Black

click for larger view

I agree that black's ahead and should win, but after 24 Qf2 Nxd1 25 Rxd1 Rxa4 26 Ne5 he still has to play carefully and simplify.

click for larger view

He then should end up a couple of pawns ahead but with the queens still on the board.

yes, a whole new game, lost for white.

Scrub that Master Chess (32 bit 20 Mhz) says 26 Rg1 leads to mate in 3 after 26...Ng4, and so does 26 Rc2. and 26 Rd2.

Material: B for N+P. The Black Bf5 pins Nd3 to Qc2. The Ra7 threatens Na4, so Qc2 has several burdens preventing untrammeled movement. The Black Re8 and Nc4 attack Pe3, so Nxe3 can fork Qc2 and Rd1. The Nc4 also attacks Pb2, defended by Na4. The Black Qd8 and Bg7 require activation. Black can regain the P immediately with

22…Bxd3 23.Rxd3 [Qxd3 Rxa4 24.b3 Rxa3]

23…Rxa4 24.Qxa4 Nxb2 (forking Qa4 and Rd3)

but is there something better? With all the White pieces on the Q-side, the White Kg1 has no defenders. No candidate pops out, however. Enumerate the usual suspects: checks (0), captures (feasibly Rxa4 and Bxd3, as mentioned, and Nxe3 or Rxe3), and threats (distantly, Qg5 and Bh6). The dark-squares pieces Qd8 and Bg7 seem to invite activation, which occurs naturally if the White P chain is shortened.

Candidates (22…): Rxa4, Bxd3, Nxe3, Rxe3

22…Rxe3 (threatening 23…Rxd3 24.Rxd3 Bxd3 25.Qxd3 Rxa4, etc.)

The capture 22…Nxe3 is inviting, but of the possibilities for the final piece on e3, N is more threatening and forceful than R.

23.fxe3

[else, submit meekly to the loss of a P with an inferior position]

Refusal of the sacrifice with material balance is possibly best. Even then, if Black does not initiate a sacrificial attack with Bxd4, Bg7 ties down a major piece to the defense of the isolated Pd4 for the foreseeable future and has a strong initiative on the K-side.

23…Nxe3 (threatening 24…Nxc2, 24…Bxd4, and 24…Qg5 25…Qxg2#)

The Qc2 must move. If White maintains lateral defense of g2, then:

(1) 24.Qe2 [or Qd2 or Qf2]

24...Rxa4 (threatening 25…Nxd1, 25…Bxd4, 25…Be4)

The Ne3 is immune because of Bxd4, so Black will recover the material investment and retain a strong initiative. The alternative, passive defense of Na4, leaves Kg1 in the lurch anyway (with apologies to <YouRang>):

(2) 24.Qb3 Bxd4 25.Kh1 Be4 (threatening 26…Nxg2 or 26…Qg5)

To all appearance, the position is still insane, but White is still tied up on the Q-side with defense of Na4, whereas Black has activated all his pieces and is storming an isolated K.

<1. d4 Nf6 2. c4 g6 3. Nf3 Bg7 4. Nc3 d5 5. Bg5>

Most often played, and my preference, is 5. Qb3 as in Onischuk vs L Dominguez, 2008 or Team White vs Team Black, 2007.

Another solid alternative is 5. cxd5 as in Y Pelletier vs L Dominguez, 2008 and P Cramling vs D Harika, 2008.

<5...Ne4 6. cxd5 Nxg5 7. Nxg5 e6 8.Nf3 exd5 9. e3 a5>

Apparently opening book programers like this line for both sides, as evidenced by the game Shredder vs HIARCS, 2008.

<10. Be2>

Out of 16 games in the Opening Explorer with this move, the score is 8 Black wins and 8 draws, but zero White wins.

<10...O-O 11. O-O c6 12. a3>

Perhaps an alternative worth considering here is 12. Ne1 as in the two draws S Temirbaev vs Yermolinsky, 1986 and Flear vs Gulko, 1986.

<12...Re8 13. Ne1>

Maybe 13. Rc1 = is worth considering.

<13...Bf5 14. Qb3>

Perhaps 14. Bg4 =, with the idea of exchanging off White's active Bishop and getting the White Knights more room to maneuver, is a better idea.

<14... Ra7 15. Bd3 Be6 16. Rd1 Nd7 17. Bb1?!>

IMO this impedes development and serves no purpose other than to free the d3 square for the Knight, in which case 17. Be2 makes more sense.

<17...Nb6 18. Nd3>

Well this was White's plan, but executing it now is not so good. Better I think is admitting the previous move was a mistake and playing 18. Bd3 .

<18...a4 19. Qc2 Bg4 20. Rc1>

Maybe 20. Rde1= would help avert the coming disaster for White.

<20...Nc4 21. Nxa4>

Instead of grabbing the pawn, maybe White could try to improve his survival chances with 21. Ba2 Qg5 22. Rfe1 b5 23. f4 Qh4 24. Qf2 Qxf2+ 25. Kxf2 Bf5 26. Bxc4 bxc4 27. Ne5 Ra6 28. e4 Be6 29. g4 =.

<21... Bf5 22. Rfd1?>

This appears to be the losing move.
Instead, maybe White can improve his drawing chances with 22. Nac5 b6 23. Nb3 c5 24. Rfe1 Rae7 25. Qc3 Nd6 26. Nf4 cxd4 27. Nxd4 Nb5 28. Qd2 Nxd4 29. exd4 Rxe1+ 30. Rxe1 Rxe1+ 31. Qxe1 Bxb1 32. Qxb1 Bxd4 33. g3 Qf6 34. Qd1 h6 35. b4 g5 36. Nd3 Kg7

<22... Rxe3!!>

This demolition apparently is the winning move.
However, Black may also have winning chances after 22... Nxe3!? 23. fxe3 Rxe3 24. Nac5 Bxd4 25. Kh1 Qe7 26. Rf1 Bxc5 27. Rxf5 gxf5 28. Qxc5 Qxc5 29. Rxc5 Ra4 30. Kg1 Rd4 31. Rc3 Re2 32. h3 f4 33. Kf1 Rd2 34. Ne1 Rxb2 35. Bd3 Kg7 .

<23. fxe3 Nxe3 24. Qb3?!>

White can put up much more resistance, but is still losing after 24. Qd2 Nxd1 25. Qxd1 Bxd4+ 26. Kh1 Be3 27. Nc3 (27. Rc3 d4 28. Rc5 Bxd3 29. Bxd3 b5 30. b3 bxa4 31. b4 Qe8 32. Rc4 Bh6! 33. Bf1 Re7 34. Qxd4 Re1 35. Qf2 Qe5 36. Rxc6 Bg5 37. Rc8+ Kg7 38. Rc7 Rxf1+ 39. Qxf1 Qxc7 ) 27... Bxc1 28. Nxc1 Bxb1 29. Nxb1 c5 30. Nd3 c4 31. Nb4 d4 32. Nd2 c3 33. Nb1 cxb2 34. Nd3 Qb6 35. a4 Ra5 36. h3 Qc6 37. Nxb2 Rg5 38. Qd2 Rf5 39. Kg1 Qb6 40. g4 d3+ 41. Kg2 Qc6+ 42. Kg1 Qc5+ 43. Kg2 Qd5+ 44. Kg1 Rf3 .

<24... Bxd4 25. Kh1>

Putting up more resistance, but also losing is 25. Nf2 Qg5 26. g3 Nxd1 27. Rxd1 Bxf2+ 28. Kxf2 Bxb1 29. Nc3 Qf5+ 30. Kg1 Bc2 31. Qb6 Bxd1 32. Qxa7 Bc2 33. Qxb7 d4 34. Qxc6 dxc3 35. Qxc3 Bd3 36. Qc1 h5 37. b4 Qg4 38. Qb2 Qf3 39. Qc1 Be4 40. Qf1 Qxa3 .

<25... Qh4 26. Ndc5 Qf2 27. Rg1 Bh3 28. Qxe3 Bxe3 29. Nd3 Qh4> 0-1.

White resigns as he cannot stop Black from winning decisive material.