Woke up early this morning and !!!

[M ANSARI]

Turned to see if any interesting TCEC games are on and wouldn't you know it SF vs Houdini with this interesting position. Houdini always has a habit of rating a queen a little higher than its opponents and seems to have gone for an exchange that allows a queen vs. pieces, however looks pretty drawish initially as Knight recapture seems forced and the queen seems like it can harass the white King into a perpetual ... but

[d]8/7k/p2p2pp/P2RP3/1pP5/1P2R3/4Kn1P/7q w - - 0 48

Here SF played the amazing dxe6 !!! totally ignoring the Knight recapture ... which looks like it wins by force. Deep !!!



New game
[d]8/7k/p2P2pp/P2R4/1pP5/1P2R3/4Kn1P/7q b - - 0 1

Analysis by Stockfish 270913 64 SSE4.2:

48...Ne4 49.d7 Qxh2+ 50.Kd3 Nf2+ 51.Kd4 Qf4+ 52.Kc5 Qf8+ 53.Rd6 Qd8 54.Re8 Qc7+ 55.Kd5 Qb7+ 56.Kd4 Qa7+ 57.Ke5 Qe3+ 58.Kd5 Qf3+ 59.Kd4 Qf4+ 60.Kc5 Ne4+ 61.Rxe4
+- (7.19 --) Depth: 37/100 00:05:40 3887mN
(Doe, 09.11.2013)

[RJN]

Stunning game with the eval difference over 700 cp, and Houdini thinking it was basically a draw.

[Spliffjiffer]

in fact it was exactly an evaluation difference of 9,82 cp (0,09 to 9,91)

regards

[M ANSARI]

I think Houdini can see that exd6 wins in all lines, it is just that it needs more time than SF and maybe not enough time even in this very LTC format. At this level and with this hardware a tiny little oversight can cause an immediate and brutal KO. I was always impressed at how quick and deep SF can go in an endgame, and there is no doubt it paid off here.

Houdini 3 Pro x64 - Doe,J, Blitz 3m
[d]8/7k/p2p2pp/P2RP3/1pP5/1P2R3/4Kn1P/7q w - - 0 1

Analysis by Houdini 3 Pro x64:

48.exd6 Ne4 49.d7 Qxh2+ 50.Kd3 Nf2+ 51.Kd4 Qf4+ 52.Kc5 Qxe3+ 53.Kc6 Qe6+ 54.Rd6 Qe4+ 55.Kb6 Qe3+ 56.Kxa6 Ne4 57.d8Q Nxd6 58.Qc7+ Kg8 59.Qxd6 Qxb3 60.Qxg6+ Kf8 61.Qxh6+ Kf7 62.Qf4+ Ke6 63.Qe4+ Kf6 64.Kb5 Qa2 65.Qd4+ Kg5 66.Qc5+ Kh4 67.Qe7+ Kh3 68.Qxb4 Qe2 69.a6 Qe8+ 70.Ka5 Qd8+ 71.Ka4 Qd7+ 72.Qb5 Qd4 73.Qh5+ Kg3 74.Qg6+ Kh2 75.Kb5 Qb2+ 76.Kc6 Qa2 77.Qd6+ Kg2 78.Qd5+ Kh2 79.Kb6 Qb2+ 80.Kc7 Qg7+ 81.Qd7 Qa1 82.Qd6+ Kg2 83.c5
+- (4.01) Depth: 27/108 00:10:40 10517mN, tb=3455
(Doe, 09.11.2013)


Here is full game ... first quick look at the final position with Queen and Knight vs. Rook and some pawns, it is almost hard to believe that this is 1 - 0

[pgn][Event "TCEC Stage 4"]
[Site "TCEC"]
[Date "11.9.2013"]
[Round "?"]
[White "Stockfish latest dev"]
[Black "Houdini latest dev"]
[Result "1-0"]
[PlyCount "104"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. d3 b5 6. Bb3 Bc5 7. Nc3 d6 8. Nd5
h6 9. c3 Be6 10. Nxf6+ Qxf6 11. O-O Ba7 12. Be3 O-O 13. Bxa7 Rxa7 14. Bxe6 Qxe6
15. Re1 Ne7 16. a4 b4 17. d4 Rb8 18. d5 Qd7 19. c4 Rf8 20. Qb3 Ng6 21. a5 Rb7
22. Nd2 c5 23. dxc6 Qxc6 24. g3 Rd8 25. Qe3 Nf8 26. b3 Nd7 27. Nf3 Kh7 28. Rad1
Re8 29. Qd3 Re6 30. Qc2 Qc7 31. Qa2 Nc5 32. Nh4 Rf6 33. Ng2 Kh8 34. Ne3 Qd8 35.
Rd5 Rd7 36. Nf5 Kh7 37. Re2 g6 38. Nh4 Rf4 39. Nxg6 fxg6 40. gxf4 Qh4 41. fxe5
Qg4+ 42. Kf1 Qf3 43. Re3 Qh1+ 44. Ke2 Rf7 45. Qd2 Nxe4 46. Qe1 Rxf2+ 47. Qxf2
Nxf2 48. exd6 Ne4 49. d7 Qxh2+ 50. Kd3 Nf2+ 51. Kd4 Qf4+ 52. Kc5 Qxe3+ *[/pgn]

[Ryan Benitez]

I did not see the game but looking at the position the impressive part for SF is getting to the position not playing the move it did. Even the 4 year old Fruit instantly plays the same move so I must assume almost every other modern engine would as well.

[ccherng]

Houdini thought that 48..Nd1 was a draw after 49.Rxd1 but missed that Stockfish had 49.d7

48..Nd1 49.d7 Nc3 50.Rxc3 bxc3 51.Kd3 Qd1+ 52.Kxc3 Qc1+ 53.Kb4 Qe1+ 54.Kc5 Qxa5+ 55.Kc6 Qd8 56.Kb7

and the white king hides from the checks on b7 and the advancement of the c pawn will win.

[Milos]

Rf4 sac from Houdini was ok, but the big blunder came at 41...Qg4:
[d]8/3r3k/p2p2pp/P1nRP3/1pP1P2q/1P6/Q3RP1P/6K1 b - - 0 41

Probably safe move was Rf7. SF sees it pretty quickly and Houdini needs much more time.

[michiguel]

Houdini thought that 48..Nd1 was a draw after 49.Rxd1 but missed that Stockfish had 49.d7

48..Nd1 49.d7 Nc3 50.Rxc3 bxc3 51.Kd3 Qd1+ 52.Kxc3 Qc1+ 53.Kb4 Qe1+ 54.Kc5 Qxa5+ 55.Kc6 Qd8 56.Kb7

and the white king hides from the checks on b7 and the advancement of the c pawn will win.

Key variation

[pgn]
[Event "TCEC Stage 4"]
[Site "TCEC"]
[Date "11.9.2013"]
[Round "?"]
[White "Stockfish latest dev"]
[Black "Houdini latest dev"]
[Result "1-0"]
[SetUp "1"]
[FEN "8/7k/p2P2pp/P2R4/1pP5/1P2R3/4Kn1P/7q b - - 0 48"]

48... Ne4 49. d7 Nc3+
50. Rxc3 bxc3
51. Kd3 Qd1+
52. Kxc3 Qc1+
53. Kb4 Qe1+
54. Kc5 Qxa5+
55. Kc6 {Only move to win} Qd8
56. Kb7!! {wins! takes away the last check in a8, and c pawn cannot be stopped}
[/pgn]

[phenri]

Really nice party and a great demonstration of strength and talent from SF. I congratulate SF and his team for this fabulous engine.
Not to mention nTCEC for its good times offered to spectators.

[Spliffjiffer]

after some analysis i think that 41...Rf7 draws....best i found is:
[d]8/3r3k/p2p2pp/P1nRP3/1pP1P2q/1P6/Q3RP1P/6K1 b - - 0 41

41... Rf7 42. Qd2 Qg4+ 43. Kf1 Qh3+ 44. Ke1 Qg2 45. Kd1 Nxe4
46. Rxe4 Qxe4 47. exd6 Qb1+ 48. Qc1 Qxb3+ 49. Ke1 Qa4 50. Qd1 b3 51. d7 Qb4+ 52. Qd2 Re7+ 53. Kf1 Qxc4+ 54. Qd3 Qxd3+ 55. Rxd3 b2 56. Rd1 b1=Q 57. Rxb1 Rxd7 58. Rb6 Ra7 =