GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-55)

Discussion of computer chess matches and engine tournaments.

Moderators: Harvey Williamson, bob, hgm

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Terry McCracken
Posts: 16106
Joined: Wed Aug 01, 2007 2:16 am
Location: Canada

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by Terry McCracken » Sun Aug 22, 2010 3:31 pm

Don wrote:
Terry McCracken wrote:
bob wrote:
Don wrote:
bob wrote:
Don wrote:
Terry McCracken wrote:
lkaufman wrote:
Steve B wrote:
lkaufman wrote:
Once again Connie blocks + by advancing her P with...

52..f5

[d] 8/p1pQ4/1p4k1/2r2pn1/8/3B4/P7/6K1 w - - 0 53

Steve
I play 53.Kf2, giving the option of going to g3 or e3. Now Black could consider the maneuver ...N-e4-d6 to centralize the knight, but Connie won't play this as I could win a pawn by taking on e4 and checking on e6, although I wouldn't do this as the resultant ending looks drawn.
If the f-pawn falls and if you remove both the a7 pawn and the a2 pawn you have a tablebase win. I don't think the Connie can draw, regardless.
But those kinds of wins are very difficult.

[d]8/2p3k1/1p6/2r5/4Q3/8/5K2/8 b - - 0 56

White to Move Win in 37 Moves

Black to move Loss in 48 Moves
How does white win that? White king can not cross 4th rank with that rook fortress holding it out. No way to win the c-pawn with the rook supporting it. I don't see how to force the black king out with just a queen??? Is this really a tablebase win??
This could still be a win. The plan would be to "pseudo stalemate" the king. In other words set things up so that the king has no legal moves and the rook or a pawn is forced to move - breaking the fortress or letting the king in.

I don't really know whether this is an actual win or not, but if it's been checked via database then it must be - and that would be the only possible way to proceed anyway.
After I looked at it a bit, you are probably right. White can force the king to the corner and zug him. If the king was on the other side of the board it is not so easy...
Well I know I'm right and it's technique based on Zugzwang and yes it's due to King position etc., but as Don said it's hard in practise but there's no rush here. Anyway the Connie will play worse and Larry will soon be picking off pawns.
Bob was not responding to you. Bob was responding to me he said, "you are probably right."
I know that Don.
Terry McCracken

Terry McCracken
Posts: 16106
Joined: Wed Aug 01, 2007 2:16 am
Location: Canada

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by Terry McCracken » Sun Aug 22, 2010 3:33 pm

Terry McCracken wrote:
Steve B wrote:
lkaufman wrote: I play 53.Kf2, giving the option of going to g3 or e3. Now Black could consider the maneuver ...N-e4-d6 to centralize the knight, but Connie won't play this as I could win a pawn by taking on e4 and checking on e6, although I wouldn't do this as the resultant ending looks drawn.
Connie continues to seek the safety of the dark squares with..
53..Kf6

[d] 8/p1pQ4/1p3k2/2r2pn1/8/3B4/P4K2/8 w - - 0 54

The Queen Of Darkness Sends Her Regards
Steve
I anticipate 53. Qd8+..Kg6 54. Qe7! and Black is in real trouble.
I think the Connie will snap here.
Terry McCracken

User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 2:27 pm

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by Don » Sun Aug 22, 2010 3:39 pm

Terry McCracken wrote:
Don wrote:
Terry McCracken wrote:
bob wrote:
Don wrote:
bob wrote:
Don wrote:
Terry McCracken wrote:
lkaufman wrote:
Steve B wrote:
lkaufman wrote:
Once again Connie blocks + by advancing her P with...

52..f5

[d] 8/p1pQ4/1p4k1/2r2pn1/8/3B4/P7/6K1 w - - 0 53

Steve
I play 53.Kf2, giving the option of going to g3 or e3. Now Black could consider the maneuver ...N-e4-d6 to centralize the knight, but Connie won't play this as I could win a pawn by taking on e4 and checking on e6, although I wouldn't do this as the resultant ending looks drawn.
If the f-pawn falls and if you remove both the a7 pawn and the a2 pawn you have a tablebase win. I don't think the Connie can draw, regardless.
But those kinds of wins are very difficult.

[d]8/2p3k1/1p6/2r5/4Q3/8/5K2/8 b - - 0 56

White to Move Win in 37 Moves

Black to move Loss in 48 Moves
How does white win that? White king can not cross 4th rank with that rook fortress holding it out. No way to win the c-pawn with the rook supporting it. I don't see how to force the black king out with just a queen??? Is this really a tablebase win??
This could still be a win. The plan would be to "pseudo stalemate" the king. In other words set things up so that the king has no legal moves and the rook or a pawn is forced to move - breaking the fortress or letting the king in.

I don't really know whether this is an actual win or not, but if it's been checked via database then it must be - and that would be the only possible way to proceed anyway.
After I looked at it a bit, you are probably right. White can force the king to the corner and zug him. If the king was on the other side of the board it is not so easy...
Well I know I'm right and it's technique based on Zugzwang and yes it's due to King position etc., but as Don said it's hard in practise but there's no rush here. Anyway the Connie will play worse and Larry will soon be picking off pawns.
Bob was not responding to you. Bob was responding to me he said, "you are probably right."
I know that Don.
Ok. It just sounded so bizarre - let me paraphrase:

bob: I can't see how white can win.
don: I think it's possible.
bob: you may be right.
YOU: I know I'm right.

It was like you were assuming my identity.

Terry McCracken
Posts: 16106
Joined: Wed Aug 01, 2007 2:16 am
Location: Canada

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by Terry McCracken » Sun Aug 22, 2010 3:47 pm

Don wrote:
Terry McCracken wrote:
Don wrote:
Terry McCracken wrote:
bob wrote:
Don wrote:
bob wrote:
Don wrote:
Terry McCracken wrote:
lkaufman wrote:
Steve B wrote:
lkaufman wrote:
Once again Connie blocks + by advancing her P with...

52..f5

[d] 8/p1pQ4/1p4k1/2r2pn1/8/3B4/P7/6K1 w - - 0 53

Steve
I play 53.Kf2, giving the option of going to g3 or e3. Now Black could consider the maneuver ...N-e4-d6 to centralize the knight, but Connie won't play this as I could win a pawn by taking on e4 and checking on e6, although I wouldn't do this as the resultant ending looks drawn.
If the f-pawn falls and if you remove both the a7 pawn and the a2 pawn you have a tablebase win. I don't think the Connie can draw, regardless.
But those kinds of wins are very difficult.

[d]8/2p3k1/1p6/2r5/4Q3/8/5K2/8 b - - 0 56

White to Move Win in 37 Moves

Black to move Loss in 48 Moves
How does white win that? White king can not cross 4th rank with that rook fortress holding it out. No way to win the c-pawn with the rook supporting it. I don't see how to force the black king out with just a queen??? Is this really a tablebase win??
This could still be a win. The plan would be to "pseudo stalemate" the king. In other words set things up so that the king has no legal moves and the rook or a pawn is forced to move - breaking the fortress or letting the king in.

I don't really know whether this is an actual win or not, but if it's been checked via database then it must be - and that would be the only possible way to proceed anyway.
After I looked at it a bit, you are probably right. White can force the king to the corner and zug him. If the king was on the other side of the board it is not so easy...
Well I know I'm right and it's technique based on Zugzwang and yes it's due to King position etc., but as Don said it's hard in practise but there's no rush here. Anyway the Connie will play worse and Larry will soon be picking off pawns.
Bob was not responding to you. Bob was responding to me he said, "you are probably right."
I know that Don.
Ok. It just sounded so bizarre - let me paraphrase:

bob: I can't see how white can win.
don: I think it's possible.
bob: you may be right.
YOU: I know I'm right.

It was like you were assuming my identity.
Sorry Don, that wasn't the intention. I use flat view and I just picked up at the last post, I knew Bob was talking to you but I really didn't feel like repling to a post that appeared insulting with question marks everywhere.

I think programmers rely too much on tablebases and need to teach these techniques to machines. They're fast enough now to do so.

People are starting to rely on them too much as well and in a day it when it wasn't possible people like Capablanca showed how it was done and later Fischer.
Terry McCracken

lkaufman
Posts: 4339
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by lkaufman » Sun Aug 22, 2010 3:50 pm

Steve B wrote:
Connie continues to seek the safety of the dark squares with..
53..Kf6

[d] 8/p1pQ4/1p3k2/2r2pn1/8/3B4/P4K2/8 w - - 0 54

The Queen Of Darkness Sends Her Regards
Steve
I play 54Qd8+. It's desirable to prevent defenses based on ...Re5 and/or ...Ne6. Regarding the tablebase win in the queen vs. rook and two pawns, it's quite surprising, but believable due to zugzwang. If the defending king were close to the pawns it should be a fortress draw. I was not aware that any ending of queen vs. rook and two connected passed pawns could be won when nothing was lose and there were no threats, but I learned something new today.

User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 2:27 pm

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by Don » Sun Aug 22, 2010 4:46 pm

Terry McCracken wrote: Sorry Don, that wasn't the intention. I use flat view and I just picked up at the last post, I knew Bob was talking to you but I really didn't feel like repling to a post that appeared insulting with question marks everywhere.

I think programmers rely too much on tablebases and need to teach these techniques to machines. They're fast enough now to do so.

People are starting to rely on them too as well and in a day it when it wasn't possible people like Capablanca showed how it was done and later Fischer.
A project I'm highly interested in is replacing bitbases with something much more compact. It can even be lossy since that can often give a huge additionally savings in space.

One idea I have is based on extracting some set of features and applying a naive bayes classifier against this data. The classification is win/loss/draw. It's ok with me if some knowledge engineering is required because the goal is to have a compact chess program that does not require dragging along gigabytes of databases. I would want such a thing to play nearly perfect - but the requirement of absolutely perfect is relaxed. The goal would be that it would still play perfectly with a modest search depth and even at end nodes would be very nearly perfect.

Such a project is extremely non-trivial. A compressed bitbase, which such a system would be designed to replace, is already an extremely efficient data structure.

One idea in this regard is to construct a set of rules which classify some ending with high success, and then use something like a bloom filter (or a compressed bit base) to identify incorrectly classified cases.

I think this is a very difficult problem in general, especially if you want something that is at least an order of magnitude more compact than a bitbase.

Terry McCracken
Posts: 16106
Joined: Wed Aug 01, 2007 2:16 am
Location: Canada

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by Terry McCracken » Sun Aug 22, 2010 4:57 pm

Don wrote:
Terry McCracken wrote: Sorry Don, that wasn't the intention. I use flat view and I just picked up at the last post, I knew Bob was talking to you but I really didn't feel like repling to a post that appeared insulting with question marks everywhere.

I think programmers rely too much on tablebases and need to teach these techniques to machines. They're fast enough now to do so.

People are starting to rely on them too as well and in a day it when it wasn't possible people like Capablanca showed how it was done and later Fischer.
A project I'm highly interested in is replacing bitbases with something much more compact. It can even be lossy since that can often give a huge additionally savings in space.

One idea I have is based on extracting some set of features and applying a naive bayes classifier against this data. The classification is win/loss/draw. It's ok with me if some knowledge engineering is required because the goal is to have a compact chess program that does not require dragging along gigabytes of databases. I would want such a thing to play nearly perfect - but the requirement of absolutely perfect is relaxed. The goal would be that it would still play perfectly with a modest search depth and even at end nodes would be very nearly perfect.

Such a project is extremely non-trivial. A compressed bitbase, which such a system would be designed to replace, is already an extremely efficient data structure.

One idea in this regard is to construct a set of rules which classify some ending with high success, and then use something like a bloom filter (or a compressed bit base) to identify incorrectly classified cases.

I think this is a very difficult problem in general, especially if you want something that is at least an order of magnitude more compact than a bitbase.
I'm not a programmer myself...to my regret but I think it's a very adventuresome idea. I hope you pursue it and that others follow.

Best Regards,
Terry McCracken

lkaufman
Posts: 4339
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by lkaufman » Sun Aug 22, 2010 5:33 pm

Don wrote:
Terry McCracken wrote:
lkaufman wrote:
Steve B wrote:
lkaufman wrote:
Once again Connie blocks + by advancing her P with...

52..f5

[d] 8/p1pQ4/1p4k1/2r2pn1/8/3B4/P7/6K1 w - - 0 53

Steve
I play 53.Kf2, giving the option of going to g3 or e3. Now Black could consider the maneuver ...N-e4-d6 to centralize the knight, but Connie won't play this as I could win a pawn by taking on e4 and checking on e6, although I wouldn't do this as the resultant ending looks drawn.
If the f-pawn falls and if you remove both the a7 pawn and the a2 pawn you have a tablebase win. I don't think the Connie can draw, regardless.
But those kinds of wins are very difficult.

[d]8/2p3k1/1p6/2r5/4Q3/8/5K2/8 b - - 0 56

White to Move Win in 37 Moves

Black to move Loss in 48 Moves
By an incredible coincidence today the ending of queen vs. rook and two pawns (not connected, but seemingly safe) arose in the top level game between Svidler and Howell. Svidler played it amazingly well, creating multiple zugwangs, and won the game.

Terry McCracken
Posts: 16106
Joined: Wed Aug 01, 2007 2:16 am
Location: Canada

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by Terry McCracken » Sun Aug 22, 2010 5:39 pm

lkaufman wrote:
Don wrote:
Terry McCracken wrote:
lkaufman wrote:
Steve B wrote:
lkaufman wrote:
Once again Connie blocks + by advancing her P with...

52..f5

[d] 8/p1pQ4/1p4k1/2r2pn1/8/3B4/P7/6K1 w - - 0 53

Steve
I play 53.Kf2, giving the option of going to g3 or e3. Now Black could consider the maneuver ...N-e4-d6 to centralize the knight, but Connie won't play this as I could win a pawn by taking on e4 and checking on e6, although I wouldn't do this as the resultant ending looks drawn.
If the f-pawn falls and if you remove both the a7 pawn and the a2 pawn you have a tablebase win. I don't think the Connie can draw, regardless.
But those kinds of wins are very difficult.

[d]8/2p3k1/1p6/2r5/4Q3/8/5K2/8 b - - 0 56

White to Move Win in 37 Moves

Black to move Loss in 48 Moves
By an incredible coincidence today the ending of queen vs. rook and two pawns (not connected, but seemingly safe) arose in the top level game between Svidler and Howell. Svidler played it amazingly well, creating multiple zugwangs, and won the game.
Wow! Larry, could you be so kind to post that game?

Thanks!

Best,
Terry McCracken

lkaufman
Posts: 4339
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: GM Kaufman v.Novag Constellation 3.6 QR Odds (Moves 41-O

Post by lkaufman » Sun Aug 22, 2010 6:08 pm

[Event "5th NH Chess Tournament"]
[Site "Amsterdam NED"]
[Date "2010.08.22"]
[Round "10"]
[White "Svidler, P."]
[Black "Howell, D."]
[Result "1-0"]
[ECO "A15"]
[WhiteElo "2734"]
[BlackElo "2616"]
[PlyCount "213"]
[EventDate "2010.08.12"]

1. Nf3 Nf6 2. c4 g6 3. Nc3 d5 4. Qa4+ Bd7 5. Qb3 dxc4 6. Qxc4 a6 7. e4 b5 8.
Qe2 Bc8 9. d4 Bg7 10. g3 c5 11. dxc5 O-O 12. Bg2 Be6 13. e5 Nfd7 14. O-O Nxc5
15. Rd1 Nbd7 16. Nd4 Bc4 17. Qe3 Nxe5 18. Nc6 Nxc6 19. Rxd8 Raxd8 20. Qxc5 Bxc3
21. Bxc6 Rd1+ 22. Kg2 Bf6 23. Bb7 Bf1+ 24. Kf3 Rfd8 25. Be4 Re1 26. Qc2 Bg7 27.
g4 h5 28. gxh5 f5 29. Bb7 gxh5 30. Qc7 Be2+ 31. Kg3 Rg1+ 32. Bg2 Rd3+ 33. Be3
Rxa1 34. Qc8+ Kf7 35. Qxf5+ Bf6 36. h4 Rxe3+ 37. fxe3 Bg4 38. Qh7+ Bg7 39. Be4
Kf8 40. Bg6 Rxa2 41. Bxh5 Bxh5 42. Qxh5 Rxb2 43. Qf5+ Bf6 44. e4 Rc2 45. h5
Rc3+ 46. Kg2 Rc2+ 47. Kf3 Rc3+ 48. Ke2 b4 49. h6 Rc2+ 50. Ke3 Rh2 51. Qg6 b3
52. h7 Bg7 53. e5 Rh6 54. Qf5+ Ke8 55. Kd2 e6 56. Qe4 Kf7 57. Kc3 Rh5 58. Kxb3
Rxe5 59. Qh4 Rd5 60. h8=Q Bxh8 61. Qxh8 a5 62. Kc4 Rf5 63. Kd4 Rd5+ 64. Ke4 Rf5
65. Qh7+ Ke8 66. Qg7 Rd5 67. Qf6 Kd7 68. Qf7+ Kd6 69. Qe8 Re5+ 70. Kf4 Rf5+ 71.
Kg4 Ke5 72. Qd8 a4 73. Qa5+ Kf6 74. Qxa4 Ke7 75. Qa7+ Kd6 76. Qb8+ Kd7 77. Qb7+
Kd6 78. Qc8 Ke7 79. Qc7+ Ke8 80. Qd6 Kf7 81. Qd7+ Kf6 82. Qe8 Rg5+ 83. Kh4 Rf5
84. Qf8+ Kg6 85. Qe7 Re5 86. Kg4 Re4+ 87. Kf3 Re1 88. Qh4 Re5 89. Qe7 Re1 90.
Kf4 Rf1+ 91. Ke5 Re1+ 92. Kd6 Rd1+ 93. Kc6 Re1 94. Qh4 Re5 95. Kd6 Rh5 96. Qe4+
Kg7 97. Kxe6 Rh6+ 98. Ke7 Rg6 99. Qd4+ Kg8 100. Qe5 Rg1 101. Qd5+ Kh7 102. Qd3+
Kg8 103. Kf6 Rg7 104. Qd5+ Kh7 105. Qh1+ Kg8 106. Qh5 Ra7 107. Qd5+ 1-0

Post Reply