*
The Campbell Report
Hard Chess
with USCF Senior Master Mark Morss
*

July 1999
Computerized Analysis; Rubinstein's Defense to the Max Lange

What, yet another tooth-gnashing and shirt-rending over the vexed ethical question of the computer's use to find moves in correspondence chess? No, I will not use this pulpit to sermonize upon that subject, however much I would enjoy conjuring images of the Lake of Fire that awaits the unrepentant sinner. I will offer something more practical: how to use the computer to do effective chess analysis. The Devil, you say! But we can keep our souls safe by considering the topic from a particular angle: how to use the computer to prepare openings variations, before our games ever begin. Thus we behold the death-striking image of the demon, but safely reflected, as if in a polished shield. But do not ask, pilgrim, what to do if you want to prepare a part of your openings repertoire that could be reached from the current position of an ongoing game (say, after 1. d4)!

The use of chess engines to help analyze sharp variations is one of many ways computers have made a big impact on chess. I believe that virtually every strong player who is doing independent openings research is using a chess engine to help sort out tactical issues. The principal service that a chess engine provides is that it sees tactical possibilities extremely quickly. A strong player, studying a position, will sooner or later see the same tactical shots the computer sees, but the time he takes to recognize the possibilities impedes the progress of his or her analysis. Since chess analysis almost invariably requires consideration of many lines and sub-lines, the time taken to see tactics can be a significant impediment to productivity. Thus, much more can be done using computers than can be done without them. And the results will be free of rank blunders, which the analysis even of strong players not always is. The increase in productivity conferred by using a computer, if my reasoning is correct, diminishes as one's tactical abilities increase. Yet I am sure that IMs and even GMs benefit from using chess engines to help with their researches.

One implication of my point about enhanced productivity is that machines shouldn't be allowed to think about a position for very long. Relative to a human, the machine's comparative advantage is its quickness. A chess engine is unlikely to discover much after the first minute of calculation that a player would not discover from looking at the position for the same length of time. It is true that the quality, such as it is, of the machine's analysis increases with the time that it thinks, but not enough to warrant the extra use of time.

It is no use having a machine to help with analysis if one doesn't understand how good analysis is done. We used to have a little circle of chess masters that would meet, here in Columbus, to analyze difficult openings variations such as 1. d4 d5 2. c4 e6 3. Nc3 c6 4. e4 dxe4 5. Nxe4 Bb4+ 6. Bd2 Qxd4 7. Bxb4 Qxe4+. I noticed during our meetings that some players had a tendency to chase specific moves deeper and deeper into the game, as if an idea at move eleven could be justified by the outcome of a single strand of analysis terminating at move twenty-eight (or sometimes, so it seemed, never terminating). The obvious problem with this mode of analysis is that much can happen between move eleven and move twenty-eight. The deeper a position is, relative to the position being evaluated, the less likely it is to be relevant to the evaluation (the exception that proves the rule is a position that is deep but arises from a very forcing set of moves). Good chess analysis therefore goes broad before it goes deep, considering carefully at each stage the possible alternatives and the objective demands of the position. "We are too deep!" I was often saying during the meetings of our analytical group.

When one decides to follow a certain path, one should keep in mind what the alternative, unexplored paths looked like based on one's preliminary judgments. One should generally abandon, provisionally, exploration of a path as soon as it looks worse than an earlier, as-yet-unexplored alternative. If one has the self-discipline, it is best to proceed only a limited distance down any given analytical path, form a preliminary judgment or simply suspend one's judgment, then go back and follow another path a similarly short distance. When one has surveyed many paths in this limited way, it may be time to go deeper along some or all of them, but by a similar process of halting and retracing. One will thus avoid a great deal of wasted time that comes from too much consideration of excessively deep positions. A benefit of working in this way is that parallel themes and motifs will be seen in different variations, and these will inform each other.

A chess engine beckons us to an entirely different style of work. The machine is ever-ready with the next move, or set of exactly two alternative moves, and our inclination is to play the next and the next and the next, until we are much too deep. Variations produced with the aid of computers therefore tend to chase specific possibilities too far, and to fail to consider good chess principles at each stage. A case in point is the book that my chessfriend Bruce Monson produced about the Belgrade Gambit -- a brilliant work that I recommend to everyone. In this original and imaginative book, Monson uses a lot of computerized analysis. I think this is why one sometimes finds that he has pursued funny, specific moves of the sort that a computer would select, while not considering simple, principled moves that would more likely be chosen by a strong player, like Monson, if he encountered the same position over-the-board. A work of such comprehensiveness and depth as Monson's, concerning such a difficult system as the Belgrade, would have been difficult to produce without computerized assistance. But the example shows that even very good players can be led into unchesslike pathways by the wiles of chess engines.

Another temptation offered by chess engines is to trust their evaluations of positions, which are mostly worthless. When using a chess engine to do analysis, one should either ignore the output of the scoring function or regard it as approximately indicative of the material balance, nothing more. Rather than just checking the machine's score and moving on, it is important to try to grasp intellectually each position that appears on the screen. One reason for this is that chess engines lack imagination. They will see amazing tactics that resolve within their horizon (every machine considers only a certain number of moves deeper into the game), but they will miss surprisingly obvious ideas that do not resolve that soon. Chess engines are notoriously unable to deal with the notion of inevitability (such as the king which must be mated on g8 because there is an unexchangeable enemy bishop on f6, no possible defense of g7, and an enemy queen that will inevitably reach h6). Long-range sacrifices are mostly beyond their ken. When a tactical idea occurs to one that the machine does not see, one should nudge the machine in that direction and help it along to the eventual result. Sometimes in those cases, the stupidity of the machine will be exasperating, and that is just one drawback of using chess engines.

I have found just one useful way to use a chess engine's scoring function. If you are down material and the machine says the score is about even (other than a forced perpetual), then you have the advantage! Trust me, it is true.

Another reason for keeping in mental touch with the screen position is that the moves to be considered next should generally be determined by the player, not by the machine. The machine is too much of a materialist to trust it with the selection of candidate moves. The player should rely on his or her own judgment; machines have none.

Some readers may want a recommendation of a good chess engine. I am familiar only with two, Fritz 5 and Hiarcs 6. I use them both, but between the two, I prefer Hiarcs. Its thinking is more like that of a human, and so it makes a better analytical companion. Also, it seems to know more about chess in general than Fritz does. I understand that Hiarcs 7.32 works with Windows. Good. Hiarcs 6 is DOS-based (with a correspondingly ugly user interface) and causes problems with Windows 98 on my Sony laptop. One problem with Hiarcs is that it over-values rooks relative to minor pieces, so it always likes to play with a rook and pawn against knight and bishop -- often a questionable proposition. Also, it suffers from the mistaken belief that rook, bishop and pawn are generally adequate compensation for a queen. I mention these problems of material evaluation because machines are self-avowed materialists: what else are they good for, if not to win material? Having said all that, Hiarcs is still my best choice.

Apropos of using chess engines to analyze sharp opening systems, a reader sent me an e-mail questioning my evaluation of an important position in my discussion of the Classical Defense to the Spanish - Part 1 (Variation 1): 1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. O-O Bc5 5. Nxe5 Nxe5 6. d4 a6 7. Be2 Nxe5 8. dxc5 Nxc5 9. Qd4 d6 10. f4 Ng6 11. Qxg7 Bf5.

Diagram 1
Position after 11...Bf5

Here I said, "This position offers Black interesting counterplay." No less an authority than Nunn's Chess Openings rules the position unclear, so I hoped I would be on firm ground. My correspondent looked for something better than the 12. Qd4 and 12. Nc3 suggested in my writeup. He shared, first, 12. Re1 Ne6 13. Bd3 (if 13. Bf3, I would say 13...Qe7 14. Qc3 O-O-O remains unclear) 13...Bxd3 14. cxd3 Qe7 15. Qc3 Rg8 16. f5 Nh4 17. fxe6 Ne6+ 18. Kh1 Nxe1. Now 19. Qxe1 fxe6 leaves Black with a rook and pawn versus two minor pieces rather far from the ending, but also with a good files for his rooks and a potentially mobile mass of queenside pawns.

The reader then turned the diagrammed position over to Fritz, and it came up with a much more challenging idea: 12. g4!? Ne6 13. Qc3 (13. Qh6 may also be good) 13...Rg8 14. gxf5 Nxf4+ 15. Kh1 Nxe2 (15...Qg5 doesn't seem to help) 16. Qe3 N2d4 17. fxe6 Nxe6 and it is hard to believe that Black's two pawns, plus White's slightly insecure king, add up to compensation for the missing piece.

It seems to me that, instead of 12Ne6, Black should play 12Qe7 13. gxf5 Qxe2 14. fxg6 Qg4+ 15. Kh1 O-O-O. This looks more like chess than the Fritz analysis submitted by my correspondent. Then 16. Nc3 (16. Be3 hxg6 17. Nc3 Rdg8 18. Qf6 Nd7 and Black wins) 16hxg6 17. Qd4 (17. Rf2 Rh3 18. Bd2 Rdh8 and 17. Be3 Rdg8 18. Qd4 Rxh2+! 19. Kxh2 Ne6 20. Qf6 Ng7 look very good for Black) 17Rh3 18. Be3 (18. Qg1 is satisfactory for Black after 18Qh5 19. f5 Rdh8 20. Bf4 g5 21. Bg3 Qg4) 18Ne6! 19. Qd2 Ng7! and by hurrying his knight to f5 with attendant pressure down the h-file, Black obtains what I believe is adequate compensation for his missing piece.

Diagram 4
Position after 19...Ng7

One plausible idea for play from the diagram is 20. Bd4 Nf5 21. Qg2 Qh5 22. Rfd1 c5 23. Bf6 Ne3 24. Rd5! Nxd5 25. Nxd5 Re8 26. Ne7+ Kd7! with a complicated situation. For example 27. c4 Qf3 28. Re1 Qxg2+ 29. Kxg2 Rd3 30. Re2 Rd4! or 27. Rf1 d5 28. Re1 Qf3 29. f5 Qxf5 (but not 27. Qxb7+? Ke6). Of course, there are other possibilities, and I invite readers to submit their analyses of the diagrammed position. If I receive any, I will treat the question in a future column. Certainly, Fritz's 12. g4!? is a critical test of Black's game after 11...Bf5, and my correspondent (who has not replied to my request to use his or her name) is to be thanked for pointing this out.

The Max Lange Attack is a maze of highly specific variations seemingly ununited by any principles of chess other than the primal urge to mate, and is therefore a fertile ground for computer-assisted investigation. Of course, after 1. e4 e5 2. Nf3 Nc6 3. Bc4 Nf6 4. d4 exd4 5. O-O, Black can play the safe and sound 5...Nxe4, avoiding the Max Lange. This was discussed at some length in the March 1999 edition of this column. But when I am eager for the full point, I prefer to play 5...Bc5, agreeing to the Max and its almost unfathomable complications.

The main line of the Max then goes 6. e5 d5 7. exf6 dxc4 8. Re1+ Be6 9. Ng5 Qd5 10. Nc3 Qf5 11. Ne4. Early deviations by White are decidedly suboptimal. My standard try here has been Rubinstein's Defense: 11...Bf8!, and I have been well satisfied. The move 11...O-O-O has been theory's favorite for the past five or six decades (someday all systems will be old enough for remarks like that to be made about them), but I mistrust the position after 12. g4 Qe5 13. Nxe6 fxe6 14. fxg7 Rg8 15. Bh6! as first played by Marshall against Tarrasch in Hamburg 1910.

Diagram 2
Position after 11...Bf8

Rubinstein's move takes a much more straightforward approach to the position. Black simply defends his g-pawn and trusts that his mobile queenside pawns will provide the advantage in the ending that will hopefully be forthcoming. White's only good continuation is 12. Nxf7 Kxf7 13. Ng5+ Kg8 14. g4 Qg6. Now White can opt for one of the most complicated lines chess has to offer: 15. Rxe6 gxf6 16. Qf3 Kg7 with almost utter obscurity and excellent winning chances for each player. A sample of what can happen is given in Baffo-Morss, below. Or White can try to clarify the situation with 15. fxg7 Bd5! (another Rubinstein idea, emphasizing the weakness of White's kingside) 16. gxh8=Q+ Kxh8 and the question is, does Black have compensation for the exchange? Koltanowski and recent authors advocating the Max have answered "no," based on 17. Bf4 (the older try, 17. Nh3, leads to an excellent game for Black). In various antique simultaneous games, Koltanowski encountered 17...Bd6 and 17...Bc5, and did very well. But Black here has not only kingside play, but also potential play on the queenside, and for that reason and to better control the center, I think he should play 17...Bg7!

Diagram 3
Position after 17...Bg7

I like Black's chances in this as-yet-untested position, and some of my specific ideas will be found in the notes to Baffo-Morss.

The second game, Soricelli-Morss, demonstrates what happens to White if he answers 11...Bf8 with 12. g4 instead of 12. Nxf7.

Game 1. Baffo - Morss, USCF-93RT21

Game 2. Soricelli - Morss, USCF-92CM96

Download Games
for a zipped file of both games (with commentary)
in new ChessBase (CBH) and PGN formats.

Next month: The Staunton Gambit .

Copyright © 1999 by Mark F. Morss

Back Home

Webmaster: J. Franklin Campbell