Author: Ricardo Gibert
Date: 01:55:10 01/21/03
Go up one level in this thread
On January 21, 2003 at 00:14:53, Robert Hyatt wrote: >On January 20, 2003 at 23:00:55, Ricardo Gibert wrote: > >>On January 20, 2003 at 21:57:38, Robert Hyatt wrote: >> >>>On January 20, 2003 at 21:32:22, Ricardo Gibert wrote: >>> >>>>On January 20, 2003 at 20:44:21, Robert Hyatt wrote: >>>> >> >>[snip] >> >>>>>Round 6 Crafty vs Searcher >>>>> >>>>>A near disaster for the first game of the second day. The same d4 opening >>>>>led to a similar position, but things did not go very well here. First score >>>>>out of book was -.42, which was typical for every 1. d4 game crafty played as >>>>>white. But it was able to pull that up quickly normally. 10 moves out of book, >>>>>the score hadn't changed, showing that searcher was playing very well and with >>>>>a reasonable amount of understanding of the position. >>>>> >>>>>Finally by move 24, Crafty was back to a slightly + score, and this held until >>>>>it started dropping as it misjudged the queen/rook attacking in the center. At >>>>>move 32, the score was -.68 after 16 plies. at move 35, the score was -1.5, at >>>>>move 40 -2.0, -2.5 at move 50, -3 at move 60, and at this point Crafty dug >>>>>in its heels and pulled the score back to -2.3 where it stayed for a long >>>>>while. But it slowly traded pawns, and the score started swinting back. By >>>>>move 80, it was -2.0 again, -1.5 by move 85, -1.0 by move 95, and it finally >>>>>reported a draw score at move 102. Of all the games it played, this was a >>>>>really nice effort as it showed a lot of understanding about king rook and pawn >>>>>endings, something I have worked on a lot over the years. >>>> >>>> >>>>I find your last sentence surprising. Crafty was very lucky in this game that >>>>Searcherx did not play 62...Re8 62.Kxg3 Rb8 winning easily. >>>> >>>> >>> >>>Crafty says if you play Re8 it just plays Rb7 immediately, not Kg3. >>> >>>Score doesn't change much... >>> >>>It may be overlooking something, but it isn't going to let black get >>>the rook behind the pawn... >>> >>>Maybe we are at the wrong move? IE you have two move 62's above. >>>Do you mean 61. Re8? >> >> >>Yes. 61...Re8 62.Kxg3 Rb8 was my intention. >> >> >>> >>>Going back to move 61, and playing Re8 Kg3 Rb8 I get Rd4 and Rb4. White >>>loses one of the pawns on the h file, but only one. It isn't clear to me, >>>without a lot of study, how black makes progress. The pawn is blockaded, so >>>the black rook is stuck on the b file unless it gives check. The black >>>king can't abandon the kingside or white will eat the g pawn and the hpawn >>>should be enough to force the trade of the rook and a draw... >> >>[D]8/1p1R2pk/5p2/7P/7P/5Kn1/4r3/8 b - - 0 60 >> >>61...Re8 62.Kxg3 Rb8 63.Rd4 b5 64.Rb4 Kh6 65.Kg4 Rb7 66.Kh3 Kxh5 67.Kg3 g6 >>68.Kh3 f5 69.Kg3 Rb6 70.Kh3 Rc6 71.Rb3 Rc4 is a prosaic and convincing win. Here is yet another winning method that keeps the outside passer and instead makes a sham sac of a pawn to advance the outside passer 2 further squares, then tempos White into dropping his last h-pawn: Varying from the above with 70...f4 71.Rxf4 b4 72.Rf2 b3 73.Rb2 Rb5 74.Kg3 Rb4 75.Kf2 Kxh4 76.Kg2 g5 77.Kh2 g4 78.Kg2 g3 79.Kh1 Kh3 80.Rh2+ Kg4 81.Rb2 Kf3 82.Kg1 Ke3 83.Kg2 Kd3 winnning. I added this to illustrate the hopelessness of Whites position. After 61...Re8 62.Kxg3 Rb8 any decent endgame player should recognize this instantly as a completely lost position. >> >>The idea of getting the rook behind a passed pawn for a won rook ending also >>occurred in the CCT where both programs (one was Diep) missed it. If I remember >>correctly, Diep won anyways by a more circuitous and eventful route. >> >>In this same tournament, the other idea of abandoning the outside passer to get >>2 connected passers also occurred (as in the above augmented analysis). I can't >>remember the program names, but that game was particularly funny. The defending >>side tried to invite the other side to win in this way (!) and the other program >>resisted! I stopped watching the game at this point, so I don't know how it >>turned out. >> >>My point is that Rook endings are a major achilles heal for programs. Part of >>the problem is that the programmers really *do* have understand how to play Rook >>endings to be able to get their programs to play them well. The trouble is, >>learning to play them and then program it is very very difficult. I know of no >>program that has done this competently. For example, Ruffian too is blind to the >>move 61...Re8. > >Crafty was expecting this move, but as I said, the score was not a lot different >from what actually happened, at least to the depth I searched earlier tonight. > >It was expecting the same variation you gave, but the effect seemed to be the >same here. > >I won't try to debate whether Re8 is a winning move or not, yet, as it will >take some study. But I am not sure how to win after the response Crafty >gave, Rd4 and if b5 Rb5. Now the pawn is stopped, and the white king hold >the last h-pawn. Not that it isn't winnable, but I didn't see a way with a >pretty quick look... > > > >> >>The proper methods of treating the diagrammed position are stock methods that >>are part of every decent endgame players arsenal and so far programs remain >>clueless. >> > > >If you look at all the endings Crafty played, it at _least_ understands the >concept of "rooks belong behind passed pawns." And it is even smart enough >to know that doubling behind the pawn is pointless, which many programs will >do, thinking that if one behind the pawn is good, two is better. :) > > [snip]
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.