Author: Ricardo Gibert
Date: 20:51:50 08/05/99
Go up one level in this thread
On August 05, 1999 at 22:56:22, Robert Hyatt wrote: >On August 05, 1999 at 15:17:52, Ricardo Gibert wrote: > >>On August 05, 1999 at 10:42:09, Robert Hyatt wrote: >> >>>On August 05, 1999 at 00:47:24, Ricardo Gibert wrote: >>> >>>>On August 05, 1999 at 00:44:35, Ricardo Gibert wrote: >>>> >>>>>On August 04, 1999 at 23:29:12, Robert Hyatt wrote: >>>>> >>>>>>On August 04, 1999 at 21:04:00, Ricardo Gibert wrote: >>>>>> >>>>>>>On August 04, 1999 at 20:08:41, Robert Hyatt wrote: >>>>>>> >>>>>>>>On August 04, 1999 at 17:32:51, Ricardo Gibert wrote: >>>>>>>> >>>>>>>>>On August 04, 1999 at 16:30:38, Robert Hyatt wrote: >>>>>>>>> >>>>>>>>>>On August 04, 1999 at 14:40:31, KarinsDad wrote: >>>>>>>>>> >>>>>>>>>>>On August 04, 1999 at 14:09:18, Bruce Moreland wrote: >>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>>On August 04, 1999 at 12:16:52, Dann Corbit wrote: >>>>>>>>>>>> >>>>>>>>>>>>>As a 'for instance': >>>>>>>>>>>>> >>>>>>>>>>>>>Suppose that on promotion, a program sees that it can promote to a knight >>>>>>>>>>>>>instead of a queen, and get a king fork, taking a bishop, followed by a queen >>>>>>>>>>>>>fork, taking the other bishop. In such a case, it might evaluate: >>>>>>>>>>>>> -pawn+knight+bishop+bishop+two_bishop_bonus+(minor positional goo) >>>>>>>>>>>>>verses >>>>>>>>>>>>> -pawn+queen >>>>>>>>>>>>>and get something a fraction more valuable than a queen. But down the road I >>>>>>>>>>>>>would rather have the queen than a knight and remove the two bishops. >>>>>>>>>>>>> >>>>>>>>>>>>>How do programs deal with this? >>>>>>>>>>>> >>>>>>>>>>>>You are really saying you'd rather have a queen against two bishops than be a >>>>>>>>>>>>knight up, right? >>>>>>>>>>>> >>>>>>>>>>>>bruce >>>>>>>>>>> >>>>>>>>>>>Actually, assuming an equal game, it is a preference of being up a queen for a >>>>>>>>>>>pawn as opposed to being up a knight and two bishops for a pawn. >>>>>>>>>>> >>>>>>>>>>>Of course, decisions like these are always based off of the actual position, but >>>>>>>>>>>here is a comment Kasparov made just the other day on Ponomariov - Al Modiakhi >>>>>>>>>>>in round 1 of the championship: >>>>>>>>>>> >>>>>>>>>>>"Looking at Ponomariov's 7.Be3 with 8.Bb6 I have sensed chess of the very >>>>>>>>>>>distant future. With my limited knowledge of the game I would consider 3 minor >>>>>>>>>>>pieces in such position much better than Queen+pawn". >>>>>>>>>>> >>>>>>>>>>>So, there are obviously positions where having 3 minors is better than having >>>>>>>>>>>the queen. >>>>>>>>>>> >>>>>>>>>>>KarinsDad :) >>>>>>>>>> >>>>>>>>>> >>>>>>>>>>I think that in almost _all_ cases, three minors pieces are better than a >>>>>>>>>>single queen.. and most games I have seen where this happens are wins for the >>>>>>>>>>three minors. I don't like two minors and 3 pawns vs a queen however, unless >>>>>>>>>>maybe if the pawns are all on the 6th rank or farther along. :) >>>>>>>>> >>>>>>>>>I felt pretty sure about this too, but I analyzed with Jack Peters a particular >>>>>>>>>position and came away with idea that it was a lot closer than I thought and >>>>>>>>>that subjective factors have a big impact. I still prefer the 3 pieces, but now >>>>>>>>>I am more careful. The 3 pieces don't organize themselves very quickly, but the >>>>>>>>>Q is relatively much faster in this respect. It depends very much on the >>>>>>>>>position. >>>>>>>>> >>>>>>>>>Related to this topic is R+P+P vs B+N. You have indicated that you set this as >>>>>>>>>equal. At first I thought this was a mistake, but then I thought about it and >>>>>>>>>this may be right for chess playing programs, since they are not sufficiently >>>>>>>>>effective in getting the rook into play. As a human, I use R+P < B+N < R+P+P. I >>>>>>>>>believe this is the "normal" evaluation. R+P+P = B+N is a practical choice for >>>>>>>>>computers. Of course subjective factors must be considered. >>>>>>>> >>>>>>>> >>>>>>>>I have done that for a long time... but IM Larry Kaufman published an article >>>>>>>>in Chess Life (not about computers particularly) about such material >>>>>>>>imbalancess, and he concluded after looking at a lot of games, that two pieces >>>>>>>>are about equal to R+P+P. Of course, if the two pawns are on the 7th, that >>>>>>>>goes out the door, naturally... or if they are connected passers that can roll >>>>>>>>quicker than the two pieces + the king can mobilize to win them... >>>>>>> >>>>>>>That is what he may have said, but he is going against the opinion of a lot of >>>>>>>strong players all the same. In the endgame for example, Reuben fine states that >>>>>>>R+P vs 2 minors is a draw, with more pawns on both sides it is still normally >>>>>>>drawn, while R+P+P vs 2 minors always win. His use of "always" pre-supposes a >>>>>>>"normal" position. >>>>>>> >>>>>>>It's true, before the endgame, comes the middlegame, so the side with 2 minors >>>>>>>MUST play very energetically to avoid loss. The side with the 2 minors is in >>>>>>>trouble. It is very possible to get compensation, since it is often not so easy >>>>>>>to activate the rook in the middle game. >>>>>>> >>>>>>>You should change your eval for the endgame at least. >>>>>> >>>>>> >>>>>>I think this is just another case where Fine is wrong. Larry's results were >>>>>>based on studying games between top players, and looking at games with a >>>>>>specific material imbalance, and then determining the win/lose/draw ratio for >>>>>>each side. He concluded, based on actual GM play, that two minors vs a rook and >>>>>>pawn is a significant advantage... and that this advantage holds until it >>>>>>becomes a rook and two pawns vs the minor, then it becomes 'equal'. >>>>> >>>>>That R+P+P has an advantage over B+N is undeniable. >>>> >>>>In the endgame of course. >>>>> >>> >>> >>> >>>the problem is that such positions first arise in the middlegame. But >>>even in an endgame, it is not terribly hard to sacrifice one piece for >>>the two pawns since that side can attack a square one more time than the >>>opponent can. And KR vs KB or KR vs KN is a draw. >>> >> >>In the endggame then, you are admitting R+P+P > B+N. Yes? > >No. I am saying that generally RPP vs BN is generally a draw. From my >observation of Crafty in these endings, and in the analysis Larry did in >his Chess Life article. If the two pawns are connected, the two pieces can >generally win one of them, or one of the pieces can be exchanged for both, >and any kr vs kn is generally drawn except for rare positions, ditto for >kr vs kb. > >So KBN vs KRP is likely a draw, if the game reaches that simple state. But >with lots of other pawns (and even a few other pieces) the BN is far better. > >KBN vs KRPP is also likely a draw, except for those cases where one king gets >cut off from the action by the rook... and even then the pawns probably have >to be on opposite sides of the board for realistic chances of winning. I >don't remember seeing any KRPP vs KBN or KBB or KNN that were anything but >drawn... > > >> >>It is NOT so easy to sac a piece for BOTH pawns. That is not really how you draw >>this endgame. When the pawns are split it is not easy to arrange to do that. In >>this case the defending side must be able to recognize when he can sac a piece >>for ONE pawn and still draw. R+P vs N or B is quite drawable under certain >>circumstances. See Basic Endgames by Balashov and Prandsetter (an excellent book >>on theoretical endgames) for how to draw SOME R+P vs N and R+P vs B endings. >>It's HARD. >> >>When the pawns are united the defending side has a really miserable time. In the >>book Endgame Strategy, Shereshevsky analyzes an ending with the pawns being >>united. FIDE awarded Lasker a medal for best defensive game of the year for >>drawing this ending. The superior side made a couple of mistakes and it was >>still impressive that the defending side drew that ending. It's HARD. >> > > >Once KQ vs KR was considered hard. We now know that computers can solve this >easily with a second per move of search, no tablebases needed. I've seen many >KRPP vs KBN and KNN endings, and I don't ever remember a result other than a >draw when the opponent was competent (ie two computers playing)... > Actually, it was considered "easy" until computers showed it was not so simple. By "easy" I mean any master was expected to be able to do it. When computers first started using EGTBs, one (belle?) played against GMs Miles and Browne and embarassed them with draws. Computers showed it was harder than expected. Browne succeeded (barely) on a 2nd try after "booking up". A GM finds this ending hard. For computers it is easy even without EGTBs as you point out. The position that had humans surprised were "the 3rd rank defense" and to a lesser extent, "the 4th rank defense", both of which were previously unknown. In both, the defending rook separates from the defending K and rounding it up was not as easy as expected. When computers play the side with the Q, it looks easy, but the problem is there are a lot of false trails that confuse humans into drawing by the 50 move rule. I learned quite a bit about this ending when I assisted a friend in his project to play this ending flawlessly. It isn't necessary to learn how to play it flawlessly to win, but part of his aim was to be able to embarass strong players just like the computers were doing. > > > > >>Virtually any type of imbalance in the middlegame signifies that one side or the >>other prefers an endgame and the other side prefers a middlegame. Trust me on >>this. With the material imbalance of R+P+P vs B+N, the side with the extra >>minors is VERY dangerous in the middlegame. In the endgame, the pawns grow >>stronger and it is the side with the extra pawns that is dangerous. I'm sure >>Kaufman would tell you that the side with R+P+P, generally, is seeking >>exchanges, while the side with B+B seeks to generate piece play in the >>middlegame. >> > >I wouldn't argue... the BN side is probably winning in the middlegame, >because of the extra attacking piece. But in the endgame, there is little >real hope of losing with two minors vs a rook and pawn or even two pawns. >Unless we decide GM players don't know how to play, because they generally >don't win this ending (with KRPP vs KMM) according to the games Larry examined. > >Again, I don't remember the total number of games for this particular ending >that he looked at, but I think he had over 150 games for every specific ending >he studied, else he tossed it out as 'insufficient data'... > > > > > >>> >>> >>> >>> >>>>>> >>>>>>Again, based on previous games GM vs GM without computers in the loop... That >>>>>>wasn't really news as most everyone considers B+N vs R+P to be bad for the R+P >>>>>>side (which is why such trades on f2/f7 are not played very often in high-level >>>>>>chess games.) He found other interesting things as well and published them in >>>>>>his article in CL.
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.