Author: Robert Hyatt
Date: 07:17:04 01/03/98
Go up one level in this thread
On January 03, 1998 at 01:55:11, Amir Ban wrote: >On January 02, 1998 at 17:39:22, Robert Hyatt wrote: > >>On January 02, 1998 at 16:50:29, Amir Ban wrote: >> >>> >>>The first game Comet - Junior in CSCC has a challenging position for >>>computers, fit for some test-suite material. >>> >>>On the 43rd move, black is an exchange up, and can win another piece to >>>be a rook up, but this loses. >>> >>>Black manages to avoid Rxd3 on the 45th move, but plays it on the 47th >>>move. Which program manages not to play Rxd3 both times ? Is there a way >>>to win here ? It's not even clear to me that black can avoid losing. >>>White can play 45. b6 or maybe even h3 and black is in some trouble. >>>Also interesting is that white can save the piece by 45. c7 but that may >>>be the only way to lose :) >>> >>>How to evaluate black's 43... e4 ? Maybe one exclamation mark for >>>playing it, two exclamation marks for NOT playing it ? >>> >>>Strangely enough, white got a decisive advantage but then failed to win. >>> >>>Amir >>> >>> >>> >> >> >>I ran this and at move 45 crafty never considers Rxd3. But it does >>believe that Qc5 gives black a significant advantage (almost +2). First >>I'll give it's analysis when it is only allowed to consider Rxd3, then >>I'll give complete analysis: >> >>(these are run on my notebook with a .75mb hash, so this can >>be searched *far* quicker with bigger hash and a P6200 or >>something...) >> depth time score variation (1) >> 7 0.25 -2.87 Rxd3 Rxd3 Qxd3 Qxd3 Rxd3 b6 Rd1+ Kg2 >> Rd2+ Kh3 Rd8 Kg4 >> 7-> 0.27 -2.87 Rxd3 Rxd3 Qxd3 Qxd3 Rxd3 b6 Rd1+ Kg2 >> Rd2+ Kh3 Rd8 Kg4 >> 8 1.08 -- Rxd3 >> 8 1.16 -3.03 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Kg7 b7 >> Rd1+ Kg2 Rd2+ Kh3 Rd8 >> 8-> 1.18 -3.03 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Kg7 b7 >> Rd1+ Kg2 Rd2+ Kh3 Rd8 >> 9 1.59 -3.16 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Rd1+ Kg2 >> Rd2+ Kh3 Rd5 b7 Rh5+ Kg4 Rb5 >> 9-> 1.61 -3.16 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Rd1+ Kg2 >> Rd2+ Kh3 Rd5 b7 Rh5+ Kg4 Rb5 >> >> 10 4.82 -3.18 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Rd1+ Kg2 >> Rd2+ Kh3 Rd5 b7 Rh5+ Kg4 Rg5+ Kf4 >>Rb5 >> 10-> 4.84 -3.18 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Rd1+ Kg2 >> Rd2+ Kh3 Rd5 b7 Rh5+ Kg4 Rg5+ Kf4 >>Rb5 >> 11 5.86 -- Rxd3 >> 11 13.71 -4.43 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Kh7 b7 >> Rb3 c7 Rxb7 c8=Q Rb1+ Kg2 Rb2+ Kh3 >> Rd2 Qc3 >> 11-> 13.75 -4.43 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Kh7 b7 >> Rb3 c7 Rxb7 c8=Q Rb1+ Kg2 Rb2+ Kh3 >> Rd2 Qc3 >> 12 45.43 -4.43 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Kh7 b7 >> Rb3 c7 Rxb7 c8=Q Rb1+ Kg2 Rb2+ Kh3 >> Rd2 Qc3 Re2 >> 12-> 45.47 -4.43 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Kh7 b7 >> Rb3 c7 Rxb7 c8=Q Rb1+ Kg2 Rb2+ Kh3 >> Rd2 Qc3 Re2 >> 13 2:23 -4.46 Rxd3 Qxd3 Qxd3 Rxd3 Rxd3 b6 Kh7 b7 >> Rb3 c7 Rxb7 c8=Q Rb1+ Kg2 Rb2+ Kh3 >> h5 Qf8 f5 Qd6 >> >> >>I then went on to move 47, and found the same thing >>as before... .07 seconds into the search Crafty finds that >>Rxd3 loses badly and that move never comes back up again. >>Is it possible you have a bad bug to play that?? >> > >Huh ? Your PV at ply 7 shows black a rook up, yet -2.87 down ? Looks >like you evaluate the two passed pawns as worth almost a queen, unless I >am missing something. Of course, you had to do that to solve WAC #2. >It's quite correct in this position, and maybe the only way to handle >it. This would be very risky if applied too generally. What is the >formulation ? > >Amir No. Two passed pawns are worth a rook, *if* there is less than a queen left for the opponent, *and* the opponent's king can't get to the queening square of either pawn, *and* neither pawn is blockaded. I didn't do this to solve wac2, because wac2 is tactically solvable without any real problem (I believe it takes 11 plies for crafty with this eval heuristic turned off, and about 25-30 seconds on a P6). I did this because I had lots of such endgames on ICC and against IM/GM players, crafty was entering endgames material up, but with zero chance of winning because of the two connected passers. I looked at the position at the end of the PV. One white pawn is on the 7th, the other on the 6th. Neither is blockaded, the black king is too far away, and the rook is totally ineffective here. All done in evaluation... In any case, you did ask if *any* program would avoid Rxd3. :) The answer is "yes".. :)
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.