Engines and Pawn Breaks

     If you analyze with engines you have noticed that various engines don't always agree on the best move. They are all good at spotting tactics, but what if no tactics are available? In endings and even in the middlegame engines sometimes enter positions that are lost because they can overvalue factors like space and material. Larry Kaufman has said engines play at around 2300 or 2400 in these situations; their super-human strength comes from spotting tactics. One area where all the engines seem to struggle is an understanding of P-breaks. 
     Stockfish is good at finding deep tactical moves and for calculating endgames, but it can it can miss some things because of its pruning; it investigates some lines deeply and others not so much. Houdini is pretty good at blitz time controls, but not for deep analysis. Shredder is good in purely positional positions. Komodo is also good for making good positional valuations. It is classical in its approach and is probably the best when it comes to finding P-breaks. 
     What is a Pawn break? According to Dan Heisman, it is a P-move that breaks up the opponent's fixed P-chain by attacking the opponent's Pawns with the Pawn is making the break. C.J.S. Purdy advised that when examining a position you should always look at them because sometimes a P-break can give surprising results. 
     The question we face when making P-breaks is when we should play one. In the following position taken from Panchanathan vs. Novikov, Dallas UTD, 2004, black's P-break 14...g5! was very significant because the key e5 square is what really matters. Black's P-sac gained control of e5 and then white's K was exposed to a dangerous attack. Once the black N landed on e5 square the rest was easy. 
 

Black to move

     Being curious as to how my engines would play in a situation where a P-break was nearly decisive they were allowed to examine the position for 3 minutes on a single core. Note: Giraffe's evaluation in no way compares to those of other engine's!! For example, after 1.e4 e5 it evaluates the position at +/- and as being 0.87 in white's favor. 
     Of the moves suggested by the engines, six selected the correct move, but their evaluations varied considerably.  Of course, there is always the possibility that had any of the engines not selecting 14...g5 been allowed to think much longer they might have changed their move selection.  That's because all of moves were at least examined by all of the engines and there was not that much difference in the score of their top candidates.

Engines selecting 14...g5
SmarThink - (-/+) (-0.97) 
Komodo 8 - (=/+) (-0.62) 
Stockfish 7 - (=/+) (-0.37) 
Gull - (=/+) (-0.32) 
Houdini 1.5 -  (=/+) (-0.26) 
Shredder Classic 4 Kamikaze  (=) (-0.16) 

Engines selecting 14...Nc5
Rybka 2.3.2 - (=) (0.11) 
Fritz 12 -  (=) (0.03) 

Engine selecting 14...h5
Crafty - (=) (0.12) 

Engines selecting 14...e5
Zappa 1.1 - (=) (0.23) 
Giraffe - (-/+) (-5.62) 

Engines selecting 14...O-O
Shredder Classic 4 Solid - (=) (0.12) 
Shredder Classic 4 Beancounter -  (=) (0.07) 
Shredder Classic 4 Gambit =  (=) (-0.16) 

     Using Komodo how did the other candidate moves fare in engine evaluation? It's evaluations are as follows: 

14...e5 (=) (0.12) 
14...Nc5 (=) (0.01) 
14...h5 (=) (-0.14) 
14...O-O (=) (-0.07) 

     Clearly the strongest engines were up to the task of finding the correct P-break even though they did not agree on the score, but I was rather surprised at SmarThink's evaluation which in this case was probably the closest to what would a human grandmaster's opinion would be even though in engine vs. engine play it does not fare well against Stockfish or Komodo. 
     But, what about a position that involves Pawn chains and a Pawn break? How do the engines handle those positions? I'll look at that in the next post.