click for larger view

Analysis by Houdini 4 Pro w32:

click for larger view

Analysis by Houdini 4 Pro w32:

As I see it, the pawn sac is a way to play for a win while keeping a draw in hand.

Play to avoid exchanges unless he understands that he has to force exchanges. We should avoid exchanges and try to keep the two bishops as long as possible.

This concept requires him to have a human plan greater that the forty ply engine ideas.

If I was playing black my first objective would be the exchange of dark square bishops by playing Ph5 and Ph4

1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 Bc5 5. c3 O-O 6. O-O d6 7. Nbd2 Ne7 8. d4 exd4 9. cxd4 Bb6 10. Re1 Bg4 11. h3 Bh5 12. a3 Bg6 13. Ba4 d5 14. e5 Ne4 15. Nxe4 dxe4 16. Nh4 Qxd4 17. Qxd4 Bxd4 18. Bg5 Nc6 19. Nxg6 fxg6 20. Rxe4 Bxf2+ 21. Kh2 Rf5 22. Bd2 Re8 23. Bc3 h5 24. Rd1 h4 25. e6 Bg3+

click for larger view

Analysis by Rybka : depth 35 ply

1. = (0.00): 26.Kg1 Bf2+ 27.Kh1 Bg3 28.Bc2 Ref8 29.Kg1 Bf2+ 30.Kh1 Bg3 31.Kg1 Bf2+ 32.Kh1 Bg3 33.Kg1 Bf2+ 34.Kh1 Bg3 35.Kg1 Bf2+ 36.Kh1 Bg3 37.Kg1 Bf2+ 38.Kh1 Bg3 39.Kg1 Bf2+ 40.Kh1 Bg3 41.Kg1

2. = (0.00): 26.Kh1 Ref8 27.Kg1 Ne7 28.Rg4 Bf2+ 29.Kh2 Bg3+ 30.Kg1 Bf2+ 31.Kh2 Bg3+ 32.Kg1 Bf2+ 33.Kh2 Bg3+ 34.Kg1 Bf2+ 35.Kh2 Bg3+ 36.Kg1 Bf2+ 37.Kh2 Bg3+ 38.Kg1 Bf2+ 39.Kh2 Bg3+ 40.Kg1 Bf2+ 41.Kh2

This perp threat forces off the black square bishops with 25 Be1 unless the team decides that the perp is prettier.

<cormier: peterfritz ... 17...Bxd4 18. Bg5 Nc6 19. Nxg6 fxg6 20. Rxe4 Bxf2+ 21. Kh2 Rf5 22. Bd2 Bb6 23. Bb3+ Kf8 24. Be6 Rd8 25. Bf4 Ke7 26. Bxf5 gxf5 27. Ree1 Ke6 28. Rad1 Rd4 29. Rxd4 Bxd4 30. b4 a6 31. Kg3 Bb2 32. Bc1 Bc3 33. Rd1 Bd4 34. h4 Bc3 35. h5 Bd4 36. h6 g6 37. Bf4 Bb2 38. Rd3 Ba1 39. Kh4 Bd4 40. Rd2 Bc3 41. Rc2 Bd4 42. Kh3 Ba7 43. g4 fxg4+ 44. Kxg4 Bb6 45. Kg3 Kf5 46. Re2 Ke6 47. Rd2 Bd4 48. Rc2 Bb6 49. Kg4 Bd4 50. Kf3 Bxe5 51. Bxe5 Kxe5 52. Kg4 Kd6 53. Kg5 Ne7 54. Rd2+ Nd5 55. Re2 Kd7 56. a4 c6 57. a5 Ne7 58. Kf6 Nf5 59. Rh2 Ke8 60. Rh1 1-0 >
[Event "Lang 240m+30s"]
[Site "Dort"]
[Date "2014.09.05"]
[Round "1"]
[White "Komodo 8 64-bit121416"]
[Black "Stockfish PA_GTB 240514 64 SS"]
[Result "1-0"]
[ECO "C65"]
[PlyCount "157"]
[EventDate "2014.??.??"]

Carlsen a chicken?! Interesting. Well, last game of the tournament. Nobody wanted to play seriously. A pity, yes. Congrats to Caruana, way above the field, and what field!>

Are you sure? A little wishful thinking can ---.

24. Be6 Rf2 25. Bc3

(25. Be1 Rxb2 26. Rf4+ Ke7 27. Bd5 Rf8 28. Bh4+ g5 29. Rxf8 Kxf8 30. Rf1+ Ke8 31. Bf7+ Kf8 32. Bc4+ Ke7 33. Bxg5+ Ke8 34. Bf7+ Kf8 35. Bh4 Rd2 36. Bb3+ Ke8 37. Bf7+)

25... Rd8 26. Be1 Rxb2 27. Bh4 Rdd2 28. Rf1+ Ke8 29. Bf7+ Kd7 30. e6+ Kc8 31. Rg4 Ne5 32. e7 Nxg4+ 33. Kh1 Re2 34. hxg4 Bf2 35. Bxf2 Rxe7 36. Bc5 Ree2 37. Bc4 Re4 38. Bd5 Re8 39. Bf7 Rd8 40. Be6+ Kb8 41. Be7 Re8 42. Bd7 Rg8 43. Be6 Rh8 44. Rf7 Rb1+ 45. Kh2 Re1 46. Bd7 Rd1 47. Bg5 Ra1 48. Rxg7 Rxa3 49. Bf6 Re3 50. Rf7 Rg8 51. Rxh7 c6 52. Bg5 Re4 53. Bd2 Rf8 54. Rg7 Rd4 55. Bc3 Rd3 56. Be5+ Ka8 57. Bc7 Rh8+ 58. Kg1 Rf8 59. Be6 b5 60. Rxg6 b4 61. Be5 Re3 62. Bg7 Rd8 63. Kf2 Ra3 64. Bf5 Rd2+ 65. Kg1 Re3 66. Bh6 Re1+ 67. Kh2 Rd8 68. Rxc6 Kb7 69. Rc4 a5 70. Bg7 Rf1 71. Bf6 Re8 72. Rc5 Rf8 73. Bd4 Rd1 74. Bf2 Re8 75. Kg3 Re2 76. g5 Rd8 77. g6 a4 78. Kf3 Rde8 79. Rb5+ Kc6 80. Rb6+ Kc7 81. Rxb4 a3 82. Ra4 a2 83. g7 Rxf2+ 84. Kxf2 Rg8 85. Rxa2 Kb6 86. Rb2+ Ka7 87. Rd2 Rxg7 88. Rd7+ Rxd7 89. Bxd7 Kb6 90. Kg3 Kc5 91. Bf5 Kd6 92. Kf4 Ke7 93. g4 Ke8 94. g5 Ke7 95. Ke5 Kf8 96. g6 Kg7 97. Ke6 Kf8 98. Kf6 Ke8 99. g7 Kd8 100. g8=Q+ Kc7 101. Ke5 Kb6 102. Qb8+ Ka5 103. Bd3 Ka4 104. Bc4 Ka5 105. Qb5#

I do! Here is a comparison with 1.d4 instead of e4:

1. d4 Nf6 2. c4 e6 3. Nf3 b6 4. g3 Ba6 5. b3 d5 6. Bg2 Nbd7 7. O-O c5 8. cxd5 Nxd5 9. Bb2 cxd4 10. Nxd4 Rc8 11. Bxd5 exd5 12. Nf5 Nf6 13. Qd4 Rc6 14. Na3 Bc8 15. Qe3+ Re6 16. Qg5 Rg8 17. Nb5 Rxe2 18. Rac1 Be6 19. Qf4 Ne4 20. Nc7+ Kd7 (21. Rfd1 Qb8 22. Nxd5)

click for larger view

I also knew that I could not expect less than an engaging answer. As <kwid> pointed out that endgame is very instructive and "should serve as a learning experience for all of us; it also illustrates the shortcomings of engines in the endgames".

Carlsen - Aronian. Position after 45...Kh7

click for larger view

In the diagram position, instead of 46.h6? (allowing Black to play 46...Rb3 followed by ...Rb5), Carlsen could have won the game by playing 46.Kc2!, restricting Black's rook from getting to the fifth rank:

46. Kc2 Kg7 47. h4 Kh7 48. h6 (Black is in Zugzwang) Kh8 49. Ra7 Kg8 50. h5 Kh8 51. a6 (Black is again in zugzwang) Ra5 (or any other rook move) 52. Kxc3 reaching a theoretically won position.

After 46.h6 (loosing a precious tempo to restrict the black rook) Black draws with 3 pawns down using the Vanèura defense:

46.h6 Rb3 47.Kc2 Rb5 48.Kxc3 Rf5

click for larger view

This is the so called Vanèura defense (discovered by Josef Vanèura in 1924): Black's rook is capable of both attacking the a-pawn and delivering lateral checks if the white king advances too far.

This position is a theoretical draw, but engines assess it as won for White!

Excellent play by both players after 48.Kxc3.

Stockfish 5 (Syzygy) 60 [+5.19] [+5.04] >

Yes however none of that matters if the engine is able to find the correct move and as long as the eval doesn't cause the engine to lose when it could have drawn.

click for larger view

Analysis by Houdini 4 Pro w32:

What a great question, not just in this specific context but in general!. Yes, I thought that the reason why tablebase probing might have been considered slower than calculations was because of the disk access required. But these tests were probably done some time ago, when computers had less memory than they typically do now. And with larger memories come larger disk caches, so while there might have been some slowdown when the pertinent tablebases were initially loaded, if the disk cache is large enough then all subsequent tablebase probings would be done with the data already in the disk cache and there would not be any need for additional disk access. Or at least they would be greatly reduced.

I also thought that because Syzygy tablebases are much more compact than either Gaviota or Nalimov tablebases that there would be a much greater chance that the tablebase data would be in the disk cache with the Syzygy tablebase than with the others. And, indeed, the Komodo times were significantly greater (148%) with the Syzygy tablebases disabled than with them enabled. But the Stockfish times, while still improved with Syzygy tablebases enabled, was not significantly different (53% faster) than Houdini's times with Gaviota tablebases enabled and disabled (48% faster), so this is not conclusive.

Of course, this is <one> position and the results are not necessarily generally applicable. And my timing was not 100% rigorous. For example, I did not reboot between analysis runs although I did take care to minimize disk cache effects by not running analyses with engines using the same tablebase one after another. And I occasionally snuck in and checked my e-mail while the analyses was going on, although this should have a minimal effect in even the shortest run.

I did a more thorough comparison test about 2 years ago with 4 positions that arose in the Team White vs Team Black, 2012 game and 5 different engines, with tabled bases either enabled or disabled, and the tablebases stored on either the hard disk or a flash drive. Again the results were not consistent; for some positions and some engines the time-to-search-depth was greater when the tablebases were enabled than when they were not, and for different engines and the same position it was the reverse. And sometimes the times were less when the tablebases were on the hard drive and sometimes they were less when the tablebases were on the flash drive.