Random Posts

Monday, June 13, 2011

Engines and Material Imbalances

      As you well know, one area where engines have a hard time is evaluating a position with an unusual material imbalance.  Of course, this can also be true for humans because while there are a lot of “rules” for the value of the pieces in general, an exact evaluation cannot be made on general principles; specific considerations of the position must be taken into account.  In addition to approximate counting the value of the pieces, you also have to take into account such things as P-structure, mobility, K-safety and a host of other features.
      In recent years, engines have shown much improvement in making these evaluations. However, you still cannot always take those evaluations at face value.  Back in the old days of chess engines, say 5 years ago, things were a lot less clear.  For example in the following position taken from CC GM Robin Smith’s book, Modern Chess Analysis which was published in 2004, some evaluations were:

Junior-White 2-1/2 P’s up
Hiarcs-slight advantage to Black
Fritz, Rebel and Chessmaster-equal or slight advantage to Black
Chess Tiger and Shredder-slight advantage to White
Nimzo-White up by 1.0 P’s

       Which engine was right?  The programs were giving Black a huge plus for his 3 connected, passed P’s.  This may be an advantage in the ending, but this position is not an ending and the truth is White has about a one Pawn advantage here.  Today let Houdini analyze this position and it quickly spots the correct 32.d7 and gives White about a one Pawn edge and in only a couple of minutes is 100% sure it’s the correct move.
      Fritz 12, Firebird and Spike all also quickly (unlike the engines of Smith's day. He often had to let them run for hours!) concluded that 32.d7 was the best move, only varying a little in their evaluations.  I should note that I’d read Spike was primarily a positional engine so I tried it out in a short blitz match against Houdini.  In CC positional play is what you want because no engine is going to fall for faulty tactics.  Houdini won by +1 -0 =4.  Spike’s loss came as a result of a tactical error that it probably wouldn’t have made if it had not moved almost instantly.  Still, for the time being, I’m sticking with Houdini for my analysis.
      In a recent Lechenicher SchachServer event I actually had 3 games in which there were material imbalances.  In two games I had a Q vs. 2 R’s which were both drawn.  In the third game we arrived at the following position with White (me) to make move 56.  The engines showed I had a 2-3 Pawn advantage but after several shootouts and trying several candidate moves I was unable to devise any winning plan so I decided to reduce the material to the basics: a R+P vs. B+2P’s because Black’s extra R always seemed to offer him counterplay. 

Thus after 56.Rg6 Rxg6+ 57.hxg6+ Kxg6 my plan was to transfer the K to c5 and sac the R for a B and P.  After this I hoped to capture the a-Pawn then advance it.  Unfortunately this does not work because Black’s K is able to reach the queening square with only one move to spare and I could not figure out a way to pick up the extra tempo I needed.  58.Kf2 Now I supposedly have a one P advantage. The K passes behind the R so as to keep the Black K cut off. 58… Kg5 59.Ke3 Bd5 60.Rf8 Bc4 61.Kd4 Kg4 62.Kc5 Here I looked hard to find a mate with the R&K but his B was always stymied that idea. 62...Kh4 63.Kb4 Kg3 64.Rf5 Kg4 65.Rxb5 Bxb5 66.Kxb5 arriving at the following drawn position:

     This is a draw because Black’s K is inside the square.  If it was outside it would be a win. After spending a lot of time analyzing this position with several different engines and ending up with gobs of analysis stored with the game I was unable to discover any method of forcing a win in the initial position. All this shows what I’ve said before is correct (as are most things I say…that’s a joke):  Titled CC players who manage to routinely defeat players who rely solely on engine use and even manage to outplay other titled players have to be pretty strong players or they would never be able to discover ways to defeat the engines of today.  On LSS, I looked up the names of several of their top rated players and discovered most didn’t have OTB ratings that I could find, but many did have CC titles with the ICCF.  Their #2 ranked player is OTB Grandmaster Alex Fier of Brazil but he never finished any of his games and lost them all by forfeit.  Too bad he didn’t stick with it because it would have been interesting to see how well an OTB GM fared in this kind of chess.

No comments:

Post a Comment