Getting the most out of your computer

Robin Smith: Modern Chess Analysis

Gambit Publications, 176 B5 pages
UK £15.99 – US $24.95

The last 15 years have seen a profound change in the chess world; the rise of powerful personal computers has given every player the chance to have a grandmaster-strength second. But how many players really use computers to their best effect? Probably very few. There are many pitfalls to using computers since, unlike a human GM, the computer equivalent has some remarkable blind spots which can easily lead an unwary user astray.

The author of this book may not be a familiar name to over-the-board players, but he has achieved considerable success in correspondence chess and recently qualified for the correspondence GM title. Using computers is perfectly legal in correspondence chess, and Smith’s extensive experience in this area makes him an ideal person to write this book. He describes how best to use computers in various types of chess analysis, including tactical positions, the opening and the endgame. He discusses the strengths and weaknesses of chess engines in general, and goes on to identify the idiosyncrasies of particular engines.

Note that you won’t find detailed instructions of the ‘now press Shift-F7’ type in this book. The discussion is all about how engines analyse, and how to get the most out of them. The specific keystrokes required to accomplish the tasks will obviously vary from program to program, and even from one version of a program to another, and wisely Smith does not get involved in this level of detail. Topics covered include interactive analysis, using multiple engines, running engine tournaments, using tablebases and blunderchecking.

Perhaps the greatest surprise is how computer analysis can be of help even in positions where there are no immediate tactics. Take the following excerpt from the book:

Tal – Vogt, Tallinn 1981
1 e4 c5 2 Nf3 e6 3 d4 cxd4 4 Nxd4 Nf6 5 Nc3 d6 6 Be2 Be7 7 0-0 0-0 8 f4 Nc6 9 Be3 e5 10 Nb3 exf4 11 Rxf4 Be6 12 Qe1 Nd7 13 Rf1 Nde5 14 Nd5 Bg5 15 Qd2 Bxe3+ 16 Nxe3 Qh4. Drazen Marovic, in Understanding Pawn Play in Chess, gives this move a question mark, stating that Tal thought it was the “crucial error”. 17 Nf5 Bxf5 18 exf5 Qf6 19 Rad1 Rad8 20 c4.

Marovic states: “Black’s sally to h4 has finished badly. The d6-pawn has been blocked as a lasting weakness, and the d5-square is in White’s control.” When I showed this position to chess programs, they universally said it is equal. Hiarcs and Junior prefer White by about 0.2 pawns, while Fritz, Shredder and Chess Tiger prefer Black by the same amount.

At first I thought, “Aha! An example of a pawn-structure that programs don’t understand”. Isolated d-pawns are notoriously difficult, for both humans and computers, since in some cases they are a weakness and endgame liability, while in middlegames they are often strong, both providing support to piece outposts and constantly threatening to break open the position via their advance.

Yet here the d-pawn is without a doubt weak. At d6 it is not far enough advanced to support any particularly dangerous knight outposts. In addition, Black’s position is somewhat cramped, which should make it impossible to organize his pieces in support of advancing the pawn. Surely the pawn is weak, and thus Black’s position must be worse.

Yet programs are all quite consistent and stable in their assessment of equal. So I ran a 7-engine, 42-game tournament. Black scored 61%, to White’s mere 39! OK, perhaps there were a lot of endgame flukes. So I looked at the games. It is important when running engine tournaments to check the games to see why each side got the results they had.

Black was not winning in the endgame via flukes, but in the middlegame, by finding counterplay. And the counterplay took many forms. First, while the d6-pawn is certainly a liability, White’s pawns are not all solid either. The pawn on f5 requires constant protection, while the pawn on b2 can also sometimes become a target.

Secondly, Black could in many of the games support and play ...d5, ridding himself of the weakling at d6, and as is typical of isolated queen’s pawn positions, creating havoc in the process. Black’s knights can quickly be redeployed in support of such a push; for example, ...Ne7 or ...Nd7-b6. White’s bishop is also an awkward piece, since it cannot go immediately to its natural long diagonal via the f3-square, without Black being able to exchange it off. The white knight has an even harder time entering the game, since Nd4 allows Black to play ...d5 immediately, and other squares for White’s knight are not appealing.

20...Rfe8 21 Rf2 h6 22 Qf4 Qg5?!

Here is the true start of Black’s troubles. With 22...Ne7 Black would already be threatening to rid himself of the d6-pawn, while the target pawn at f5 is also still a problem for White. Another 7-engine, 42-game tournament from this position again indicated that Black is perfectly OK, in fact scoring 63% to White’s mere 37%.

23 g3 f6? This is also a mistake, creating light-square weaknesses at e6 and g6.

24 h4 Qxf4 25 gxf4 Nf7 26 Bf3. Most top programs quickly find White’s last three moves, and in spite of White’s doubled, isolated f-pawns, have no trouble seeing that it is White who now stands better.

26...Re3 27 Kg2 Kf8 28 Rfd2 Rde8 29 Kf2 h5 30 c5 d5 31 Rxd5 Ne7 32 Kxe3 Nxd5+ 33 Kf2 Nxf4 34 Rd4 Nh3+ 35 Kg2 Re3 36 Bxb7 Nh6 37 c6 1-0.

Click here to replay and download this game

The chapter on endgame analysis is also intriguing; Smith shows how a combination of a deep search plus tablebase access can prove a powerful tool. The following study was used on one website as an example of how computers don’t understand fortresses, and indeed considerable fun was had at the expense of the programs which couldn’t see the ‘draw’. However, as Smith points out, the reason for this is that the fortress can in this case be broken down and the ‘draw’ doesn’t actually exist. This is how the author explains it:

Zakhodiakin ‘64’, 1929

White to play and win

This ending study is featured on the website of Valentin Albillo as an example of how computers cannot understand certain types of positions. After the forced sequence of moves 1 g5+ Kh7 2 Bf7 c2 3 Kh5 c1Q 4 g6+ Kh8 5 Kg4

the black king has no moves and is perpetually caged in by the bishop and g-pawns.

Here is what the website says about the position: “The queen alone can neither mate the white king, nor separate it from the pawn, so stalemating the king to force the bishop to move is also impossible. Black can only give check after check, without accomplishing anything. A draw. However, most chess programs, if not all, cannot recognize this.”

Sure enough, if you set up the position after White’s 5th move, computers all show a large advantage for Black, since the queen is worth far more than the bishop. The 50 moves required for a draw are so far away that no computer can see that far.

But if you set up the program Hiarcs and let it think for a while, strange things starts to happen. Instead of Hiarcs’s evaluation of Black’s advantage starting to fall over time, it starts to grow. Ply after ply, the evaluation continues to climb, initially showing about a five pawn advantage and steadily climbing to nearly seven pawns after a few minutes.

We learned previously that in drawn positions the position evaluation will usually decrease towards zero, yet here it is increasing. This is often a telling sign that things are not all as they appear. OK this is interesting. What will happen if we let the computer think about it for much longer? Hiarcs announces a checkmate! It turns out that White can be forced to give up the pawn, after which the king is stalemated, forcing the bishop to leave its post as well. The analysis can be continued as follows:

5...Qe3! 6 f4

We can verify that other moves also lose for White:

6...Qd3 7 Kh4 Qf3 8 Kg5 Qh3 9 f5 Qf3 10 Bc4 Qe3+ 11 Kh5 Qe4 12 Be6 Qf4.

This position is certainly not a draw! Either Black will win the f-pawn, or his king will escape its prison, as White’s king has no moves and the bishop cannot both guard the f-pawn and control g8.

Hiarcs' reaction to the above position after a few seconds (!) of thought.

Click here to replay and download this game

In summary, then, Modern Chess Analysis is an essential guide to the effective use of computers for chess analysis. Gambit Publications inform me that this book is currently being translated into German, with a scheduled publication date of February 2005.

Links

• Gambit Publications home page

You can buy the book from directly from Gambit, from the British Chess Magazine Shop, the London Chess Center, US Chess4Less or Schach Niggemann