game link

Updated 2019-07-19

Lichess now has complete 7-piece tablebases online so you can follow the endgame from Black move 49.

Big Summary to July 2019 - in order of importance

Black move 37 ... e6 as played should LOSE, 37 ... e5 would DRAW.
White move 38. h6 as played should DRAW, 38. Rd1 would WIN.
Black move 54 ... b4 as played LOSES, 54 ... Qd3 or 54 ... Qd5 would DRAW.

Black move 26 ... f4 as played appears worse than 26 ... Bc5 - in response, White should have played 27. h4 instead of 27. Qf7.
Black move 21 ... Rxa4 as played appears worse than 21 ... f4.

Black move 29 ... Qc4 as played appears slightly worse than 29 ... b4 - perhaps an "easier" drawing path.
Black move 15 ... Ra8 as played appears slightly worse than 15 ... b5.
Black can head directly for a clear forced draw by 18 ... Bd4.
Black has "easier" drawing paths with moves like 34 ... Bh8, 35 ... Ne5, and 51 ... Ka1.
Black MAY have a better move with 19 ... Qd4.
Different ways to approach the position are 6 ... e5, 16. Be3, 19. Bd2, 25 ... Nd4 - unclear.

Updated 2017-10-25

This time I used Stockfish 111017 64 BMI2 and a c4.8xlarge instance with 5 piece tablebases. However, it became clear that for moves 36 and 37 I had been evaluating the position wrongly in the past.

37... e5 should draw and so 36... Kd5 is also a draw.

Move 6 - recommends ... e5

depth 49 seldepth 71 score cp -8 nodes 654516113531 nps 34771383 hashfull 999 tbhits 27675 time 18823413 pv e7e5 d2d3 g7g6 h2h4 f8g7

Move 15 - still recommends ... b5

depth 65 seldepth 39 score cp 0 nodes 1065040637566 nps 40112041 hashfull 999 tbhits 45776331 time 26551644 pv b6b5 f1e1 b5b4 c3a4 d7c7

if 15 ... b5 16. Be3 - recommends 16 ... b4

depth 61 seldepth 38 score cp 0 nodes 409892736169 nps 38645985 hashfull 999 tbhits 7266488 time 10606347 pv b5b4 c3a4 f6d5 a2a3 e7e6 a

after 16 ... Ra5 - recommends 16 ... Ne4

depth 65 seldepth 99 score cp 8 nodes 6023473162270 nps 42314416 hashfull 999 tbhits 651892519 time 142350376 pv f6e4 b2b3 c4e6 a1a2 e4

move 18 - still ... Bd4

depth 70 seldepth 27 score cp 0 nodes 1985738421206 nps 42115162 hashfull 999 tbhits 299818488 time 47150202 pv g7d4 b3f7 c6e5 f7b3 e5d

move 19 - now ... Qd4

depth 68 seldepth 40 score cp 0 nodes 1840270673233 nps 42387119 hashfull 999 tbhits 525725783 time 43415799 pv e4d4 f1e1 g7e5 a1d1 d4b

move 21 - still ... f4

depth 64 seldepth 88 score cp -13 nodes 868671243562 nps 41659027 hashfull 999 tbhits 385899090 time 20851933 pv f5f4 f7h7 b4e4 h7f7 e5

move 36 - still ... Bc3 - however, with 6 piece tablebases it recognizes that 37 ... e5 is a draw

depth 76 seldepth 95 score cp -44 nodes 1488368064318 nps 48412543 hashfull 999 tbhits 6166357726 time 30743439 pv d4c3 h5h6 c3d4 h6h7

move 37 - after 37 ... e5 it recommends 38. Bc1 but the five piece base tablebases don't score it as a draw yet. The six piece tablebases show it's a draw.

depth 64 seldepth 86 score cp 60 nodes 232879526594 nps 44578460 hashfull 999 tbhits 1194978077 time 5224037 pv f4c1 c6e7 f1f7 d5e6 f7f

Updated 2015-12-06

I spent a month using Stockfish 6 with five piece Syzygy tablebases on an m4.10xlarge instance (160 GB RAM, 40 cores of Intel Xeon E5-2676 v3 Haswell 2.4 GHz) on Amazon EC2. Older analysis was done with Stockfish 5, Stockfish DD, and Komodo 8.

I have the same conclusions as before really. The biggest mistakes were

38. h6 where 38. Rd1 wins

36 ... Kd5 which loses where 36 ... Bc3 should hold on, and then 34 ... Bh8 and 35 ... Ne5 are easier draws.

Of course 54 ... b4 loses immediately ...

Subtler are 18 ... Bd4 which is a clear draw that Stockfish prefers to 18 ... f5;

21 ... f4 that Stockfish prefers to 21 ... Rxa4; then after Kasparov's suggested 22. Rac1 it's another draw;

26 ... Bc5 that Stockfish prefers to 26 ... f4 (because of 26 ... f4 27. h4 Qf5 28. Qb1 Qe5 29. Qd3 b5 30. Qf3 b4 31. Bxf4 Qf5 32. h5 Ke8 33. Be3 Qxf3 34. gxf3 Bc3 and the White h-pawn is very dangerous (+106) vs 26... Bc5 27. Qd1 Qc4 28. h4 (+90)) ...

then move 29 where Stockfish suggests 29 ... b4 with a draw (see below) versus the less clear draw in the game.

Stockfish prefers

15 ... b5 (Paehtz, GM Chess Club). Eval 0. Kasparov in his book said he'd continue with 15. Be3, when Stockfish suggests 15 ... b4 16. Na4 Nd5 17. Rac1 Qxa2 18. Ra1 Qc4 19. Rac1 Nxe3 20. Nxb6+ Kd8 and the evaluation is still 0. Older analysis of 15 ... d5 with Stockfish 5 gave eval +13 after 16. Na4 Qb5 and 15 ... Ne4 (Barnet Chess Club) with Stockfish 5 and eval of 0 after 16. Nxe4 Qxe4 17. Be3 b5.

16 ... Ne4 - though if 16 ... Ra5 (Paehtz) was played Kasparov in his book said he would play 17. Nb5. Then Black plays 17 ... Ne4, and the evaluation for White is -10 after 18. Re1 f5. This is the only time in the game when following a line suggested by Kasparov in his book results in a negative score for White.
16 ... d5 (Bacrot) is also eval 0 with repetition after 17. Nb5 d4 18. Re1 Qd5 19. Ra3 e5 20. Rd3 Rae8 21. Nc3 Qc5 22. Nb5 Nd5 etc.
16 ... Nd4 (Felecan) is eval -58 after 17. Be3 Nd5 18. Bxd4 Bxd4 19. Nb5 Bc5.

18 ... Bd4 19. Qxf7 Ne5 20. Qb3 Nd3 21. Qb5+ Kc7 22. Bd2 Rf8 23. a5 Nxf2 24. axb6+ Kb8 25. Rxf2 Bxf2+ 26. Kh1 Qf5 27. Qb3 Qf7 28. Qb5 Qf5 with repetition. Kasparov's book has 21. Ra3 in this line also leading to a draw.
18 ... Nd4 (Paehtz, Felecan, Bacrot) is worse - eval +42 after 19. Qxf7 Nc2 20. Bd2 Nxa1 21. Re1 Qh4 22. Qd5 Rf8 23. Qxb7+ Ke8 24. Qc6+ Kf7 25. Qd5+ Ke8 26. Qb5+ Kf7 27. Ra1
Komodo 8 analyzed 18 ... e6 (King) to have eval +8 after 19. Qxb6 Nd4 20. Bg5 Qc6 - further back in the analysis Stockfish 6 has 18 ... e6 19. Bd2 Bd4 20. Rae1 Qd5 21. Qxd5 exd5 with eval 0.
Stockfish 6 recommends 19. ... Qb4 with eval -19 and PV 20. Qxf7 Be5 21. h3 f4 22. Qxh7 Qe4 23. Qf7 Bxb2 24. Re1 Qf5 25. Qxf5 gxf5 26. Rb1 Bd4 27. Bxf4 Rxa4 28. h4 Ne5 29. Bxe5 dxe5 30. Kh2 e4 (57 ply).
It's also a fan of 19 ... Qd4 (Bacrot) until 55 ply (eval -24) and 19 ... Be5 (Computer Chess Team, IM2429) until 48 ply (eval -40).
After 19 ... Nd4 (Paehtz, Felecan) the eval is +105 after White survives a fierce attack - 20. Qf7 Ne2+ 21. Kh1 h6 22. Be3 f4 23. Bxb6 Rf8 24. Qxg7 f3 25. Rg1 fxg2+ 26. Rxg2 Nf4 27. Rg1 Nxg2 28. Rxg2 Qb1+ 29. Rg1 Qe4 30. f3 Qxf3+ 31. Rg2 Rf4 32. Qxh6 Rxa4 33. Qc1 Qc6 34. Qxc6+ bxc6 35. Be3 Ra1+ 36. Rg1 Rxg1+ 37. Kxg1. Kasparov's book has a main line with 21 ... Bxb2 22. Ra2.

21 ... f4 with PV 22. Qxh7 Qe4 23. Qf7 Qf5 24. Qxf5 gxf5 (eval -23) or after Kasparov's suggested 22. Rac1 - a draw with perpetual check after 22.Rac1 Qe4 23.Rxc6 bxc6 24.Qxe7+ Kc8 25.Qe8+ Kb7 26.Qd7+ Ka6 27.b4 Qc4 28.Qxh7 Qxb4 29.Qxg6 Qxa4 30.Bf6 Qd4 31.Bxe5 Qxe5 32.Qc2 Kb7 33.Qh7+ Kc8 34.Qg8+ Kb7 35.Qh7+

Move 29: Stockfish suggests 29 ... b4 with a draw after 30.h6 Qc2 31.Bxf4 Nd8 32.Qh5 Bxf4 33.h7 Be5 34.Qxe5 dxe5 35.h8=Q b3 36.Qxe5 Nc6 37.Qd5+ Kc7 38.Re1 b2 39.Qd1 Qc3 40.Re3 Qc1 41.Re1 Qc3

Move 33: 33 ... Bxg3 would have lost quickly after 33...Bxg3 34.h6 Be5 35.h7 Bg7 36.Bh6 Bd4+ 37.Rf2 b4 38.Kf1 Nd8 39.Be3 Bg7 40.Bc1 d5 41.Bb2 d4 42.Ke2 Kd7 43.Kd3 e5 44.Rf8 Ke7 45.Rg8 Bf6 46.h8=Q Bxh8 47.Rxh8 Nf7 48.Ra8 Kd6 49.Kc4 Ng5 50.Kxb4 Kd5 51.Bc1 Nf3 52.Ra5+ Ke4 53.Kc4 Nh4 54.Rb5 Nf5 55.Kb3 Nh4 56.Kc2 Nf3 57.Rxb7 Ne1+ 58.Kd1 Nf3 59.Ke2 Nh4 60.Bg5 Nf5 61.Rb6 Ng3+ 62.Ke1 or 33...Bxg3 34.h6 Be5 35.h7 Bf8 36. Rf8 Bg7 37. Kf2 etc

Move 34: Black can hold it together (eval -23) after 34...Bh8 35.Rb1 Kf5 36.Kg2 Kg4 37.h6 Kh5 38.Bd2 Bd4 39.Bxb4 Kxh6 40.Bd2+ Kg6 41.Rxb7 Bf6 42.Bf4 Kf5 43.Rb5+ Ke4 44.Rb6 Nd4 45.Bc1 Kf5

Move 35: Stockfish liked 35. Kh1 and was predicting 35...Ne5 36.Bxe5 dxe5 37.Rb1 Bc3 38.Kg2 Kf5 39.g4+ Kg5 40.Kf3 e4+ 41.Kxe4 Kxg4 42.h6 Kg5 43.h7 Kg6 44.Rh1 Bh8 45.Rh2 b3 46.Kd3 (eval +23)

Move 36: the only move to hold it together was 36... Bc3 with eventual six-piece tablebase draw after 36...Bc3 37.h6 Bd4 38.h7 b2 39.g5 Bh8 40.Bd2 b5 41.g6 b1=Q 42.Rxb1 Kf6 43.Rc1 Ne5 44.Rc8 Nxg6 45.Bc3+ e5 46.Rg8 d5 47.Kg2 d4 48.Bb4 e4 49.Kf1 Bg7 50.Bf8 Bh8 51.Ke2 Nf4+ 52.Kd1 Ng6 53.Bh6 b4 54.Kc2 e3 55.Kb3 e2 56.Bd2 Bg7 57.Re8 d3 58.Kxb4 Kf7 59.Rg8 Nh8 60.Kc4 Kg6 61.Kxd3 Kxh7 62.Re8 Ng6 63.Rxe2
36...Kd7 37.Bh6 b2 38.g5 Be3 39.Bg7
36...Bh8 37.g5 Nd4 38.Bd2 Nf5 39.Kg2 Be5 40.Bf4 Bd4 41.Kf3 b2 42.Kg4 Ne3+ 43.Bxe3 Bxe3 44.g6 Bh6 45.Rb1 Bc1 46.h6 Bxh6 47.Rxb2 Kf6 48.Kh5 Bf8 49.Rxb7 Kg7 50.Kg5 Kg8 51.Kf5 Kg7 52.Rb8 winning
36 ... Kf7 37. Be5+
36 ... b2 (Felecan, Bacrot) ... 37. g5
36 ... Nb4 (Paehtz) ... 37. g5
36 ... Nd8 ... 37. g5
36 ... Ne5 ... 37. g5

Move 38. Rd1 is a clear win after e.g. 38.Rd1 Ke4 39.Bxd6 Kf5 40.g6 Bg7 41.Rg1 b5 42.Ba3 b4 43.Bc1 b2 44.Bd2 Ne7 45.Rf1+ Ke4 46.Bxb4 Nf5 47.Kh2 Ne3 48.Rb1 Nc4 49.Kg3 Kf5 50.Rh1 Kg5 51.Be7+ Kf5 52.Rf1+ Ke4 53.Bg5 Nd6 54.Kg4 Nf5 55.Rd1 Ke5 56.Rb1 Kd5 57.Bd2 Kc4 58.h6 Bh8 59.g7 Nxg7 60.Rxb2 Nf5 61.Rb8 Ba1 62.h7 Nd6 63.h8=Q Bxh8 64.Rxh8
38. Bc1 does not win because of a change to the line further down. 38. Bc1 b5 39.g6 Ne7 40.Rf7 Nf5 41.h6 Nxh6 42.Bxh6 b2 43.Rf1 Kc4 44.Rb1 b4! (instead of Kc3 as below) draws.
53.Qe4 also appears to only draw after 53 ... Qf1+ 54.Ke7 Qc4 55.Qg2+ (then moves to transpose to ...) Ka3 56.Qf3+ Ka4 57.g6 Qc7+ 58.Kf6 Qd8+ 59.Kf5 Qd7+ 60.Kg5 b4 61.Qa8+ Kb3 62.Qf8 Qh3 63.Qf7+ Ka3 contrary to the Regan page about it.

Updated 2012-11-14

For White's move 27, these four programs also all recommend 27. h4 and agree on the PV for the next moves: 27. h4 Qf5 28. Qb1


32/86 14:13:56 530,248,097,451 10,348,823 +0.87 h3-h4 Qe4-f5 Qb3-b1 Qf5-e5 Qb1-d3 b6-b5 Qd3-f3 b5-b4 Bg5xf4 Qe5-f5 h4-h5 Kd7-e8 g2-g4 Qf5-f7 Kg1-g2 b4-b3 Bf4-g3 b3-b2 Qf3-e4 d6-d5 Qe4-c2 Ke8-f8 Rf1-b1 Qf7-f6 Qc2-b3 Qf6-e6 g4-g5 Qe6-e4+ Qb3-f3+ Kf8-g7 Qf3xe4 d5xe4 Kg2-f1 e4-e3 f2xe3


41/94- 13:56:00 381,600,295,786 7,607,589 +1.05 h3-h4 Qe4-f5 Qb3-b1 Qf5-e5 Rf1-e1 Qe5-c5 Re1-e2 Qc5-d5 Bg5xf4 Qd5-h5 Qb1-e4 Qh5xh4 Qe4-f5+ Kd7-c7 g2-g3 Qh4-f6 Qf5xf6 e7xf6 Re2-e1 Kc7-d7 Re1-b1 f6-f5


35 8:13:02 331,213,123,943 11,196,150 +0.85 h3-h4 Qe4-f5 Qb3-b1 Qf5-f7 Qb1-e4 Bd4-e5 Rf1-b1 Kd7-c7 Kg1-f1 Nc6-d4 Rb1-d1 Qf7-c4+ Qe4-d3 Qc4-f7 Rd1-e1 Nd4-c6 Kf1-g1 Be5-h8 Qd3-e4 Bh8-e5 Re1-b1 e7-e6 Rb1-b3 d6-d5 Qe4-e2 Be5-d6 Qe2-g4 Nc6-d4 Rb3-c3+ Kc7-b8 h4-h5 b6-b5 Rc3-h3 e6-e5


34 5:37:08 29,339,330,966 1,450,415 +0.56 h3-h4 Qe4-f5 Qb3-b1 Qf5-f7 Qb1-e4 e7-e5 Qe4-d3 Kd7-e6 Qd3-h3+ Qf7-f5 Qh3-a3 Qf5-h7 Qa3-a2+ Ke6-d7 Qa2-a8 Kd7-c7 Qa8-e8 Qh7-g7 Qe8-e6 Qg7-h7 Qe6-h6 Qh7xh6 Bg5xh6 Kc7-d7 h4-h5 d6-d5 Rf1-b1 Kd7-e7 Kg1-f1 Bd4-c5 Rb1-d1 Bc5-d4 Kf1-e2 b6-b5 g2-g4 f4xg3/ep f2xg3 b5-b4 Bh6-e3 b4-b3 h5-h6 Ke7-f6 Be3xd4 Nc6xd4+ Ke2-d3

Updated 2012-11-09

After reading the first few chapters of Michael Nielsen's book "Reinventing Discovery", where he writes

(p26) "... after the game ended Krush singled out move number 26 as one of her three favourite World team moves."

(p64) "Yasha ... who contributed the crucial move number 26. Yasha would have been lost playing Kasparov on his own. But it was very helpful, perhaps vital, for the World Team to have access to Yasha's small contribution."

... I decided to look at Black's move 26 and White's move 27 some more.

I had previously found that both Houdini 1.5a and Rybka 4 recommended 26 ... Bc5. On the Computer Chess Rating List of 27 October 2012 there are now three free / open source programs rated between 3144 and 3205: Critter 1.6a (rated just above Rybka 4), Stockfish 2.3.1, Bouquet 1.5, and Komodo 3. After some time, they all prefer 26 ... Bc5 as well. The next step for me will be to see what they prefer for White's 27th move.


35 12:50:47 64,141,770,583 1,386,944 -0.59 Bd4-c5 Qb3-d1


32/75 10:44:03 365,835,399,022 9,466,582 -0.66 Bd4-c5 Qb3-d1 Kd7-e8 Rf1-e1 Qe4-c4 Qd1-h5+ Qc4-f7 Qh5-e2 e7-e5 Re1-a1 Nc6-d4 Ra1-a8+ Ke8-d7 Qe2-d1 b6-b5 Bg5-e3 Kd7-c7 Be3xd4 e5xd4 h3-h4 Qf7-e6 Ra8-a1 Qe6-e4 h4-h5 d4-d3


35 14:26:10 563,734,780,620 10,846,500 -0.64 Bd4-c5 Qb3-d1 Kd7-e8 Rf1-e1 Qe4-c4 h3-h4 Nc6-e5 Qd1-h5+ Ke8-d7 Qh5-e2 Qc4xe2


39/87+ 14:21:33 370,593,907,220 7,169,067 -0.80 Bd4-c5 Qb3-d1 Qe4-e6 h3-h4 Qe6-g6 Bg5-e3 Kd7-c7 h4-h5 Qg6-f6 Qd1-f3 Nc6-e5

Updated 2011-04-17

Most recently, I examined some moves with Houdini 1.5a.
My conclusions are the same, so probably if I do future analysis, I'll look at the unplayed analyst suggestions up to and including move 37.
In his book, Kasparov seems to be right about the following six moves for Black. They are the top choice of Houdini and Rybka in each case:
In the book, the following five moves seemed to be assessed wrongly:

Finally a variation that was not discussed in the book was 33 ... Bxg3 34. h6 Be5 35. h7 Bg7 36. Bh6 (Houdini assesses as +1.00). So probably 33 ... Bxg3 is an error.
Raw Houdini output:
27. h4 +0.62
16... Ne4 0.00
15... b5 -0.01
26... Bc5 -0.48
36... Kf7 -0.48
29... b4 0.00

36. Bh6 +1.00 (after 33... Bxg3 34. h6 Be5 35. h7 Bg7)

21... f4 -0.19
34... Bh8 -0.35
19... Qd4 -0.10
35... Ne5 -0.52

36/90 26:40:57 1,008,124,571,183 10,494,000 +0.62 h3-h4 Qe4-f5 Qb3-b1 Qf5-e5 Qb1-d3 b6-b5 Qd3-f3 b5-b4 Bg5xf4 Qe5-f5 h4-h5 Kd7-e8 g2-g4 Nc6-e5 g4xf5 Ne5xf3+ Kg1-g2 Nf3-h4+ Kg2-g3 Nh4xf5+ Kg3-f3 Bd4-c3 Kf3-e4 Nf5-g7 h5-h6 Ng7-e6 Bf4-e3 Ke8-f7 Rf1-g1 b4-b3 Ke4-d3 Bc3-f6 h6-h7 Ne6-f8 Rg1-b1 b3-b2 Rb1-h1 Kf7-g6 Rh1-h6+ Kg6-f7

37/88 42:28:10 1,539,696,208,259 10,070,000 0.00 Nf6-e4 Nc3xe4 Qc4xe4 Qd1-b3 Bg7-d4 Bc1-d2 Kd7-e8 Ra1-e1 Qe4-g4 h2-h3 Qg4-d7 Bd2-e3 Ke8-f8 Be3xd4 Nc6xd4 Qb3-c3 e7-e5 f2-f4 Ra8-c8 Qc3-d3 Rc8-e8 f4xe5 d6xe5 Rf1-f6 Qd7xa4 Rf6xb6 Qa4-d7 Qd3-a3+ Kf8-g7 Qa3-d6 Qd7-e7 Qd6-d5 Nd4-c6

35/81 43:11:09 1,541,896,368,946 9,917,000 -0.01 b6-b5 Rf1-e1 h7-h5 Bc1-e3 Nf6-g4 Ra1-c1 Ng4xe3 Re1xe3 Bg7xc3 a2-a3 Qc4-d4 Re3xc3 Qd4xd1+ Rc1xd1 Rh8-a8 Rc3-f3 Nc6-e5 Rf3-b3 Ra8-a5 f2-f4 Ne5-c4 Rb3-b4 Nc4-e3 Rd1-c1 Ne3-d5 Rb4-d4 Nd5-e3 Rd4-d3

35/88 24:35:45 883,545,304,470 9,978,000 -0.48 Bd4-c5 Qb3-d1 Kd7-e8 Rf1-e1 Qe4-c4 Qd1-f3 Qc4-f7 h3-h4 b6-b5 Qf3-h3 Ke8-f8 h4-h5 b5-b4 Qh3-h4 b4-b3 h5-h6 Kf8-g8 Bg5xe7 Kg8-h7 Qh4-f6 Qf7xf6 Be7xf6 Kh7xh6 Re1-b1 Nc6-a5 Bf6-c3 b7-b6 Kg1-f1 Kh6-g5 Kf1-e2 Kg5-f4

36/92 20:34:54 812,560,469,698 10,966,000 -0.48 Ke6-f7 Bf4-e5+ Kf7-g8 Be5xd4 Nc6xd4 Kh1-h2 e7-e5 Rf1-a1 b7-b6 Kh2-h3 e5-e4 g4-g5 b3-b2 Ra1-b1 e4-e3 Kh3-g2 Nd4-e6 Kg2-f3 Kg8-h7 Rb1xb2 Ne6xg5+ Kf3xe3 Kh7-h6 Ke3-f4 Ng5-f7 Kf4-f5 Nf7-e5 Rb2-g2 b6-b5 Rg2-g8 b5-b4 Rg8-h8+ Kh6-g7 Rh8-d8 Ne5-c4 Kf5-g5 b4-b3 h5-h6+ Kg7-h7 Rd8-d7+ Kh7-h8 Rd7-b7 b3-b2 h6-h7 b2-b1Q Rb7xb1 Kh8xh7 Rb1-b7+ Kh7-g8 Kg5-f6 d6-d5 Rb7-c7 Nc4-e3 Kf6-e5 d5-d4 Ke5-e4 Ne3-g4 Ke4xd4 Kg8-f8 Kd4-d5 Kf8-e8

36/99 25:11:27 990,676,501,761 10,923,000 0.00 b5-b4 Rf1-c1 Be5-b2 Qf7-g6 Qe4xg6 h5xg6 Bb2-g7 Rc1-b1 e7-e5 Kg1-h2 Kd7-e6 g2-g4 d6-d5 Kh2-g2 b7-b6 Rb1-c1 Ke6-d7 Rc1-b1 Kd7-e6 Rb1-c1

37/83 25:01:50 445,644,337,352 4,945,000 +1.00 Bg5-h6 Bg7-d4+ Rf1-f2 Nc6-d8 Bh6-e3 Bd4-g7 Kg1-f1 Nd8-f7 Kf1-e2 b5-b4 Rf2-f1 Bg7-f6 Ke2-d3 Nf7-e5+ Kd3-c2 Ne5-g6 Kc2-b3 Ke6-f7 Be3-h6 Ng6-h8 Rf1-g1 Bf6-c3 Bh6-c1 Bc3-f6 Kb3xb4 Bf6-d4 Rg1-g2 Nh8-g6

35/87 27:51:59 1,030,636,169,299 10,273,000 -0.19 f5-f4 Qf7xh7 Qb4-e4 Qh7-f7 Be5xb2 Ra1-b1 Bb2-d4 Rf1-e1 Qe4-f5 Qf7xf5+ g6xf5 Bg5xf4 Bd4-c5 h3-h4 e7-e5 Bf4-e3 Nc6-b4 Kg1-f1 Kd7-e6 h4-h5 Nb4-c2 Be3xc5 b6xc5 Re1-c1 Nc2-b4 Rc1-c3 Ra8-h8 Rc3-h3 Rh8-h7 Kf1-e2 e5-e4 a4-a5 d6-d5 Rh3-c3 Nb4-d3

39/86 28:09:34 1,043,061,445,655 10,289,000 -0.35 Be5-h8 Rf1-b1 Ke6-f5 Bf4-d2 Kf5-g4 Kg1-g2 Bh8-e5 h5-h6 Kg4-h5 h6-h7 Kh5-g6 Bd2xb4 Kg6xh7 Bb4-d2 Kh7-g6 Rb1xb7 Be5-f6 g3-g4 Nc6-e5 Kg2-g3 d6-d5 Bd2-f4 Ne5-c4 Rb7-b5 e7-e5 Bf4-c1 d5-d4 Rb5-c5 Nc4-d6 Rc5-c6 Nd6-e4+ Kg3-f3 Ne4-g5+ Bc1xg5 Kg6xg5 Kf3-e4 Bf6-e7 Rc6-e6 Be7-f6

35/89 23:30:28 855,960,534,593 10,114,000 -0.10 Qe4-d4 Rf1-e1 Bg7-e5 Ra1-d1 Qd4xb2 Qb3-f7 Ra8xa4 Qf7xh7 Ra4-a2 Re1-f1 Qb2-b3 Qh7xg6 Qb3-e6 Qg6xe6+ Kd7xe6 Rd1-b1 Be5-d4 Rf1-e1+ Ke6-f7 Bg5-e3 Bd4-c5 Be3xc5 b6xc5 Rb1xb7 c5-c4 h2-h4 Ra2-a5 Kg1-f1 c4-c3 Re1-c1 c3-c2 Kf1-e1 Nc6-d4 h4-h5 Ra5-a3 Rb7-b4 Ra3-d3 h5-h6 Kf7-g6 Rb4-b7 Kg6xh6 Rb7xe7 Rd3-a3

38/113 39:35:45 1,531,443,204,846 10,743,000 -0.52 Nc6-e5 Bf4xe5 d6xe5 Kh1-g2 b4-b3 g3-g4 b3-b2 Kg2-f3 Ke6-f6 h5-h6 Kf6-g6 Rf1-h1 Kg6-h7 g4-g5 Kh7-g6 Kf3-e4 b7-b6 Ke4-d3 b6-b5 h6-h7 b2-b1B+ Rh1xb1 Kg6xh7 Rb1xb5 Kh7-g6 Rb5-b1 Kg6xg5 Rb1-f1 Kg5-g6 Kd3-e4 Kg6-g5 Rf1-f7 Kg5-g6 Rf7-f8 Kg6-g5 Rf8-h8 e7-e6 Rh8-b8 Kg5-f6 Rb8-b7 Bd4-c5 Rb7-c7 Bc5-d4 Rc7-h7 Kf6-g5 Rh7-h1 Kg5-g6 Rh1-c1 Kg6-g5 Rc1-h1 Kg5-g6

Updated 2010-10-12

This time I looked at White moves with Rybka 4.
New conclusions:

Move 38: Rd1

38. Bc1 also wins as noted below.
info depth 33 score cp 405 time 161437729 nodes 69430236420 nps 430074 pv f1d1 d5e4

Move 27: h4

This is further analysis of White's move 27, as below
info depth 29 score cp 114 time 204542896 nodes 96154838662 nps 470096 pv h3h4 e4f5 b3b1 f5e5 b1d3 b6b5 d3f3 b5b4 g5f4 e5b5 f4e3 d4f6 f1b1 b5d3 b1d1 d3c2 d1c1

After 27 h4: ... Qf5

info depth 26 score cp -128 time 24876982 nodes 10841695723 nps 435812 pv e4f5 b3b1 f5e5 b1d3 b6b5 d3f3 b5b4 g5f4 e5b5 f4e3 d4c3 f3g4 d7c7 h4h5 b4b3 h5h6 b3b2 g4e4 b5e5 e4d3 c6b4 d3c4 c7d7 h6h7 b7b5 c4h4 e7e6 e3d2 e5e2 d2c3 e2f1

After 27 h4 Qf5: 28. Qb1

info depth 28 score cp 125 time 26127923 nodes 11170615318 nps 427535 pv b3b1 f5e5 b1d3 b6b5 d3f3 b5b4 g5f4 e5b5 f4e3 d4c3 f3g4 d7c7 h4h5 b4b3 h5h6 b3b2 g4e4 b5b3 f1b1 b7b5 g2g4 c3e5 h6h7 b3c4 f2f3 e5h8 g1g2 c7d7 e3f4 e7e6 b1e1

Move 37: g5

w37 info depth 31 score cp 391 time 117929799 nodes 45405532258 nps 385021 pv g4g5 e7e6

Move 11: Nd5

As in game - the game continuation is predicted up to move 15, when Rybka expects 15... b5
info depth 31 score cp 13 time 117888615 nodes 56443400261 nps 478785 pv c3d5 e6e4 d5c7 e8d7 c7a8 e4c4 a8b6 a7b6 e2c3 b6b5 f1e1 h7h5 h2h3 b5b4 c3a4 d7c7 c1d2 f6d5 d1f3 c7b8 f3f7 g7d4 d2g5 c6e5 f7e6 e5d3 e1e2 d3f2 e2f2 b7b5 a1f1

Move 15: Nc3

As in game.
info depth 30 score cp 12 time 132673952 nodes 57461243079 nps 433101 pv e2c3 b6b5 f1e1 h7h5 c1e3 b5b4 c3a4 f6d5 d1d5 c4d5 a4b6 d7e6 b6d5 e6d5 a1d1 d5c4 b2b3 c4b5 d1d5 b5a6 e1c1 h8d8 a2a4 e7e6 d5b5 d8d7 g1f1 d6d5 e3c5 g7b2 c1d1

Move 16: Be3

Instead of 16. a4 in the game
info depth 32 score cp 19 time 126232718 nodes 52477514699 nps 415720 pv c1e3 b6b5 a2a3 d7e8 f1e1 e8f8 h2h3 a8b8 a1c1 b5b4 c3e2 c4d5 d1a4 b4a3 b2a3 d5a5 a4a5 c6a5 e1d1 f6e8 c1c2 e7e6 e3d2 a5c6 d1b1 c6d8

Move 17: Nxe4

As in game.
info depth 31 score cp 0 time 122582186 nodes 53353364759 nps 435245 pv c3e4 c4e4

Move 18: Qb3

As in game.
info depth 32 score cp 0 time 193846640 nodes 83861192251 nps 432616 pv d1b3 g7d4

Move 19: Bd2

info depth 32 score cp 56 time 118987985 nodes 49573360775 nps 416624 pv c1d2 c6d4

Move 21: h3

As in game.
info depth 30 score cp 29 time 105244063 nodes 53926843256 nps 512397 pv h2h3 f5f4 f7h7 b4e4 h7f7 e5b2 a1b1 b2d4 f1e1 e4f5 f7f5 g6f5 g5f4 a8a4 f4e3 d4c5 e3c5 b6c5 b1b7 d7c8 b7b1 a4h4 f2f3 e7e5 g1f2 c8d7 b1b7 d7e6 b7b6 e6d5

Move 25: Qf7

As in game.
info depth 30 score cp 91 time 207197456 nodes 89752409782 nps 433173 pv g6f7 b2d4 f7b3 d4c5 b3d1 e4e6 h3h4 e6g6 g5e3 d7c7 d1f3 g6f7 f1c1 c6e5 f3f4 f7h7 f4g5 c7d7 c1d1 e5f7

Move 28: f3

Different from game, but the evaluation is still =
info depth 31 score cp 0 time 207840773 nodes 76391122670 nps 367546 pv f2f3 e4d3 f1e1

Move 29: h5

As in game.
info depth 29 score cp 19 time 67892292 nodes 26473950705 nps 389940 pv h4h5 b5b4 f1c1 e5b2 f7g6 e4e6 c1b1 b2e5 b1e1 e6g4 g5h6 g4h4 g6f5 d7e8 e1c1 e5d4 c1f1 d4e5 f5g6 e8d7 h6g5 h4g4 f2f3 g4e6 f1b1 b7b5 g6e6 d7e6 g1f2 e6f5

Move 30: Qf5+

As in game.
info depth 29 score cp 124 time 115749857 nodes 35816151259 nps 309427 pv f7f5 c4e6 g2g4 f4g3 f5e6 d7e6 f2g3 b5b4 g1f2 b4b3 g5d2 e6f5 f2e2 f5g4 f1h1 b3b2 h5h6 g4f5 e2d3 f5g6 g3g4 e5h8 d3c2 c6d8 g4g5 d8f7 h1h3 h8e5 h3h4 g6f5 h4h1

Move 31: Qxe6+

As in game.
w31 info depth 32 score cp 118 time 98659042 nodes 28118405549 nps 285005 pv f5e6 d7e6 g2g3 f4g3

Move 32: g3

As in game.
info depth 34 score cp 115 time 176963229 nodes 47467543300 nps 268233 pv g2g3 f4g3 f2g3 e5g3 h5h6 g3e5 h6h7 e5g7 g5h6 g7d4 f1f2 b5b4 h6c1 c6d8 g1f1 d8f7 f2f3 d4f6 c1d2 b4b3 f3b3 e6f5 b3g3 d6d5 g3g8 d5d4 d2b4 d4d3 f1e1

Move 34: Bf4

As in game (Rybka chooses 34. Kf2 up until ply 31, the score changes from +107 to +105).
info depth 33 score cp 104 time 128008134 nodes 53099114256 nps 414810 pv g5f4 e5h8

Move 35: Kh1

As in game (Rybka chooses 35. Kg2 up until ply 28, the score changes from +156 to +142).
info depth 31 score cp 142 time 75866110 nodes 30655562899 nps 404074 pv g1h1 b4b3

Move 36: g4

As in game.
info depth 32 score cp 138 time 139010230 nodes 38417810611 nps 276366 pv g3g4 d4c3 g4g5 c6d4 g5g6 d4f5 h5h6 b3b2 f1d1 e6f6 g6g7 f6f7 h1g2 c3f6 f4d2 f7g8 g2f2 d6d5 f2e2 f5d4 e2d3 e7e5 d3c3 e5e4 d1h1 b2b1q h6h7

After 35... Ne5

info depth 36 score cp 131 time 131036332 nodes 44903329450 nps 342678 pv f4e5 d6e5

After 36... Bh8

info depth 29 score cp 295 lowerbound time 38061686 nodes 14894175409 nps 391316 pv g4g5 c6d4 f4d2 d4f5 h1g2 b3b2 g2h3 h8e5 h3g4 f5g3 f1b1 g3e4 d2f4 e5c3 h5h6 e6f7 g4f5 e4c5 g5g6 f7g8 f4g5 e7e6 f5f4 c3e5 f4e3 d6d5 g5e7 c5d7 e3e2 d7b6 b1f1

After 36... Bc3

Hmmm ... maybe 35 ... Ne5 was a good idea!
info depth 31 score cp 141 time 24528382 nodes 11078079230 nps 451643 pv g4g5 c6d4

After 36... Kf7

info depth 32 score cp 343 time 16815364 nodes 7453057194 nps 443229 pv f4e5 f7g8

After 35... Ne5 36 Bxe5 de

info depth 35 score cp 126 time 17600782 nodes 10289399942 nps 584599 pv f1b1 d4c3

Updated 2010-05-31

I used Rybka 4 to analyse eleven different Black moves on eleven Dell Vostro 430s (Core i7 860s at 2.8 GHz). Each program ran for about 44 hours.

Move 15: ... b5.

As recommended by Kasparov.
depth 30 score cp -12 time 104769649 nodes 46910990572 nps 447753 pv b6b5 f1e1 h8c8 a2a3 c6d4 c1g5 h7h6 g5e3 d4b3 a1b1 c8c6 d1f3 c4g4 f3g4 f6g4 c3b5 g7e5 g2g3 c6c2 h2h3 g4e3 e1e3 b3d4 b5c3 e7e6

Move 16: ... Ne4

as in game.
depth 31 score cp 0 time 110774751 nodes 45377177231 nps 409634 pv f6e4 c3e4

Move 18: ... Bd4

With a draw.
depth 31 score cp 0 time 139421183 nodes 56483370094 nps 405127 pv g7d4 b3f7 c6e5 f7f4 e4f4 c1f4 d4b2 f4e5 b2e5 a1a2 e5d4 f1b1 e7e5 g1f1 d7e6 f1e2 d6d5 f2f3 h7h5 e2d3 g6g5

Move 19: ... Qd4

depth 30 score cp -26 time 47411060 nodes 24022915990 nps 506694 pv e4d4 f1e1 g7e5 h2h3 a8f8 e1e2 d4b4 b3b4 c6b4 a1d1 b4c6 g5e3 f8a8 f2f4 e5f6 b2b3 a8a6 g2g4 c6a5 d1d3 a5c6 g1g2 c6b4 d3d1 e7e6 e2d2 b4d5 g2f3 b6b5 a4b5 a6a5

Move 21: ... f4

As Kasparov feared.
depth 30 score cp -33 time 93006222 nodes 46366206229 nps 498528 pv f5f4 f7h7 b4e4 h7f7

Move 25: ... Bd4

As in game.
depth 28 score cp -93 time 134286411 nodes 67078963921 nps 499521 pv b2d4 f7b3 d4c5 b3d1 d7e8 f1e1 e4c4 d1f3 c4f7 h3h4 b6b5 h4h5 e8f8 h5h6 f8g8 f3e2 b5b4 g5e7 g8h7 e7g5 b4b3 e1b1 c5d4 e2d3 b3b2 g5d2 f7f6 g1f1 c6d8 d2e3

Move 26: ... Bc5

As recommended by three analysts.
depth 29 score cp -93 time 127144684 nodes 56916615703 nps 447652 pv d4c5 b3d1 e4e6 h3h4 e6g6 g5e3 d7c7 d1f3 c6e5 f3e2 g6f6 e3f4 e5c6 f4g5 f6f7 h4h5 e7e6 f1e1 c6d4 e2d1 e6e5 g5e3 b6b5 e3d4 c5d4 d1f3 b5b4 g2g4

Move 29: ... b4.

depth 30 score cp 0 time 52403779 nodes 30028879291 nps 573028 pv b5b4 f1c1 e5b2 c1d1 e4e2 d1b1 b2d4 b1f1 d4e5 g5f4 e5f4 f7f5 d7c7 f5f4 e2h5 f1b1 h5b5 g2g4 b5d3 b1c1 c7b8 f4e3 d3g6 f2f3 e7e5 c1c6 b7c6 e3b6 b8c8

Move 32: ... fxg3.

depth 33 score cp -117 time 92687402 nodes 37012224219 nps 399323 pv f4g3 f2g3

Move 34: ... Bh8.

depth 33 score cp -107 time 107915611 nodes 45924932091 nps 425563 pv e5h8 f1b1 e6f5 g1g2 f5g4 f4d2 h8f6 h5h6 g4h5 g3g4 h5g4 h6h7 c6d8 d2c1 d8f7 c1b2 e7e5 b2c1 g4f5 b1b4 f5g6 b4b7 e5e4 g2g3 f6g7 b7e7 f7e5

Move 36: ... Bc3.

depth 31 score cp -109 time 142117318 nodes 64125849690 nps 451217 pv d4c3 h5h6 c3d4 h6h7 d4h8 f4d2 b3b2 g4g5 b7b5 g5g6 b2b1q f1b1 e6f6 b1c1 c6e5 c1c8 e5g6 d2c3 e7e5 c3d2 f6g7 c8c7 g7f6 h1g2 g6f4 d2f4 e5f4 g2f3 f6g6 f3f4 d6d5

The drop from -33 on move 21 to -93 on move 25 indicates that Rybka thinks 21 ... Rxa4 was an error compared to 21 ... f4. After 21 ... f4, Kasparov recommended 22. Rac1 in his book. After that, Rybka recommends a line leading to a complex draw: 22... Qe4 23. Rxc6 bxc6 24. Qxe7+ Kc8 25. Qe8+ Kb7 26. Qd7+ Ka6 27. b4 Qc4 28. Qc7 Qxb4 29. Qxc6 Rb8 30. Rfc1 Ka7 31. Kh2 Qd2 32. f3 Rb7 33. Rc2 Qe1 34. Bd8 Qg3+ with perpetual check.

Similarly, the increase from -93 on move 26 to 0 on move 29 indicates that Rybka thinks 26 ... f4 was an error, and also that 27. Qf7 was an error compared to 27. h4. The PV on White's move 27 is 27. h4 Qf5 28. Qb1 Qf7 29. Qe4 Be5 30. Rb1 Kc7 31. Qa4 Bd4 32. Rc1 Bb2 33. Rxc6 bxc6 34. Qa7+ Kc8 35. Qxb6 Be5 36. Qc6+ ... (+117).

I also analysed the moves with Stockfish 1.7.1 (to be updated).

Updated 2009-07-13

I greatly enjoyed participating in this game. The lines were very deep and during the game, I spent several months almost full time analysing with the help of Crafty using spare CPU time on computers in the mathematics department at the University of Queensland.
Now, ten years later, the combined power of all of those computers is just a joke, compared to the best chess software (Rybka 3) on an average desktop computer, such as the one I'm using now (Dell Vostro 420, Core 2 Quad 9400 2.66 GHz, 4 GB RAM, 1 TB HD). I'm not a highly rated player, but I still enjoy going over this game.
I'm leaving the updated analysis (Crafty stuff) at the bottom for posterity, but for now I'd just like to write about two of the interesting variations I found with Rybka.

Variation 1

The first is Black's move 29. Rybka 3 seemed to favour 29 ... b4. However Kasparov wrote in his book that White has better chances after 30. h6 Qc2 31. Bxf4 Nd8 32. Qh5 Bxf4 33. h7 Be5 34. Qxe5 dxe5 35. h8=Q Nc6 36. Qh3+.

Kasparov gave the variation 36... Kc7 37. Qe3 and wrote that White is ready to activate the rook on c1.
However, Rybka 3 suggests 36 ... e6 is the only move that draws. The main variation is 36... e6 37. Qe3 Nd4, then if 38. Qxe5 Ne2+ 39. Kh1 b3, or 38. Re1 Qf5 39. Rd1 Qc2.

Variation 2

The second is White's move 38. Rybka 3 favoured 38. Bc1. Ken Regan wrote
After 38.Bc1, Black plays 38...b5! Saving the back b-pawn is very important. This passed pawn is outside the square of the White King. Now: A) 39.g6 Ne7 40.Rf7 Nf5! 41.h6 Nxh6 42.Bxh6 b2 43.Rf1 Kc4 44.Rb1 (44.g7 Bxg7 45.Bxg7 e5, illustrates the foundation of Black's drawing plan) 44...Kc3 45.Bc1 bxc1=Q+ 46.Rxc1+ Kb3 47.Rc7 (???) b4 48.g7 Bxg7 49.Rxg7 Kc2, with a draw - this variation illustrates the importance of saving Black's back b-pawn - it is the pawn that is significant (for Black) in the R v Ps endgames that sometimes appear in these variations.
However, Rybka 3 points out the variation
38. Bc1 b5 39.g6 Ne7 40.Rf7 Nf5 41.h6 Nxh6 42.Bxh6 b2 43.Rf1 Kc4 44.Rb1 Kc3 45.Bc1 bxc1=Q+ 46.Rxc1+

46... Kb3 47. Kg2 and 48. Kf3
when White wins. White has KRP, black has KBPPP, eight pieces in total, and I should download all the 6 piece tablebases that could arise from this position. However, other people from analyzed this position and confirmed the win. Ken Regan doesn't reply to his email and hasn't updated his page in a long time.

Archive material from 2002 and 2005


Crafty 19.19 with hash=384M, hashp=24M running on a Pentium 4-2.4Ghz produced:
Move 15 (23 May 2005)

20-> 13884:28 0.89 15. ... Rd8 16. Be3 b5 17. Rc1 b4 18. Na4 Qb5 19. Rc5 Qa6 20. b3 Ke8 21. Rc2 Kf8 22. Bb6 Rd7 23. Re1 Kg8 24. Qf3 d5 25. Bc5

Move 16 (15 June 2005)

Crafty 19.19 with hash=384M, hashp=12M running on a Pentium 4-2.4Ghz produced the same as in 2002:
20-> 5321:47 0.61 16. ... Ke8 17. Be3 Kf8 18. Bxb6 Qb4 19. Be3 Qxb2 20. Bd2 Qb6 21. Rb1 Qc7 22. Bg5 Qd7 23. Re1 Kg8 24. Qf3 h6 25. Bd2 Ng4 26. Re4

Move 18 (19 Jul 2005)

same setup as before:
20-> 5096:17 0.11 18. ... Nd4 19. Qxf7 Nc2 20. Bd2 Nxa1 21. Re1 Qh4 22. Qd5 Rf8 23. Qxb7+ Ke8 24. Qc6+ Kf7 25. Qd5+ Ke8 26. Qb5+ Kf7 27. Rxa1 Bd4 28. Be3 Kg7 29. Qb4 e5 30. Qxd6 Bxb2


The 2002 settings were hash=24M, hashp=10M. The computers used were dual Pentium III 800Mhz computers, and the software was Crafty 18.12.

The experiment shows that computers still don't play the best moves; see the commentary below.

Move 15 (October 2002)

(3) 19 20115:48 0.34 15. ... e6 16. Be3 Ra8 17. a3 Nd5 18. Nxd5 Qxd5 19. Qe2 Qb3 20. Rab1 Ke8 21. Rfd1 d5 22. Rd3 Qc4 23. Qd1 Kf8 24. Bxb6 Kg8 25. Rc1

15...e6 was not suggested by any analyst at the time of the game. After the game it was suggested in the SmartChess Online analysis.

Interestingly, Krush and Georgiev (2001) and van Wely (2000) all played 15...b5. Other games from 2000 and 2001, as mentioned on the commentary page, continued 15...e6.

Move 16 (October 2002)

(3) 20 11707:55 0.03 16. ... Ke8 17. Be3 Kf8 18. Bxb6 Qb4 19. Be3 Qxb2 20. Bd2 Qb6 21. Rb1 Qc7 22. Bg5 Qd7 23. Re1 Kg8 24. Qf3 h6 25. Bd2

16...Ke8 was examined by Krush but she recommended 16...Ne4 instead. The SmartChess Online analysis suggested meeting 16...Ke8 by 17.Re1, as did the GM Chess School.

Move 18 (October 2002)

(3) 20 10167:59 0.00 18. ... e6 19. Qxb6 Nd4 20. Kh1 Qc6 21. Qxc6+ Nxc6 22. Ra3 d5 23. Rf3 Ne5 24. Rb3 Ke8 25. Rxb7 Rxa4 26. Bg5 Nd3 27. b3 Ra2 28. Rb8+ Kd7 29. Rb7+ Ke8

18...e6 was Danny King's recommendation. Krush said that White will stand better with
"22...c5 23.Bd2 Nc2 24.Rf3 Ke8 25.b3 Nd4 26.Rd3 Rb8 27.Bc3! and White has an advantage; "
Still, it's impressive that crafty saw the next three moves in the PV. 18...e6 is not mentioned in Smartchess's post-game analysis, but we discussed it a lot at the time.

Move 19 (October 2002)

(4) 19-> 22740:33 0.47 19. ... Be5 20. Rac1 Rxa4 21. Bd2 Rd4 22. Bc3 Rc4 23. Rce1 Qd5 24. Qxb6 Nd4 25. Bxd4 Bxd4 26. Qb3 Ke8 27. Re2 Kf8 28. Rfe1 e5 29. Qg3

19...Be5 was recommended by the Computer Chess Team and "IM2429" but not by any analysts.

Move 26 (October 2002)

(3) 21 20876:03 0.45 26. ... Bc5 27. Qd1 Ne5 28. Re1 Bxf2+ 29. Kxf2 Nd3+ 30. Kf1 Qxe1+ 31. Qxe1 Nxe1 32. Kxe1 e5 33. Kd2 Ke6 34. h4 d5 35. Bd8 b5 36. h5 Kf7 37. h6 Kg6 38. Bg5 Kh7 39. Kd3

Kasparov preferred 26... f4 and recommended 27. Qd1 after 26... Bc5. Krush preferred 27. Qb1 on 26... Bc5 but Kasparov did not think this was as good as 27. Qd1. The GM School looked only at 27. Be3 and 27. h4.
The problem with the computer's PV is that the endgame after white's move 32 is winning for White, though even after 23 ply and 1536M of hash on another computer (in 2002) Crafty still couldn't see the win for White.

Move 33 (October 2002)

(3) 23 20760:55 0.43 33. ... b4 34. Bf4 Bd4+ 35. Kh1 b3 36. g4 Kd5 37. g5 e5 38. Bc1 b2 39. Bxb2 Bxb2 40. h6 Ne7 41. Rf6 e4 42. h7 Bxf6 43. gxf6 Ng6 44. f7 e3 45. h8=Q Nxh8 46. f8=Q

Obviously 33... b4 should lead to a draw. In the PV, as in the game, 37... e5 is incorrect but 37... e6 draws.