The Campbell Report
Correspondence Chess
Computer CC Challenge Match

Nimzo 7.32 versus Steve Ham

White: Nimzo 7.32
Black: Steve Ham
Computer CC Challenge Match

1.e4, c5 2. Nf3, d6 3.d4, cxd4 4.Nxd4,Nf6 5.Nc3, g6 6.Be3, Bg7 7.f3, Nc6 8.Qd2, O-O 9.Bc4, Bd7 10.h4, ...

After 10.h4

I've been playing the Dragon Sicilian for most of my chess playing life. You may question whether it suits me since my style is more technical than tactical, more long-range planning than short-term executions. Well, I can't supply a logical explanation other than this: psychologically I don't like passive positions; this is even more true when playing Black, hence the reasons I play the Dragon Sicilian and the Gruenfeld Defenses. Also, the fact that I continue to score highly with these systems encourages me to defend with aggression.

White is playing the Modern Yugoslav Attack against my Dragon. Originally the Yugoslav Attack involved 9. 0-0-0, but later it was believed that White was better served to first prevent Black from playing the freeing ... d5. Nimzo's opening book is well conceived since White's 10th move is the most precise. A popular but faulty alternative is 10. 0-0-0, which immediately declares where White's King can be found. This information allows Black a number of King hunting lines, including 10... Qb8 which is currently unfashionable and even condemned by opening theory.

However, I never follow fashion and have developed my own ideas about opening theory. I believe that the attack on White's King after this move is very difficult to face. Instead Nimzo's text move initiates White's attack upon Black's castled King. If Black now launches a Queenside assault with 10... Qb8, he's attacking thin air because White's King isn't in the line of fire. White can use the tempo saved to pry open the h-file; White can free his other Rook with Kf2 to continue the attack while Black's attack misfires on the other end of the board.

10. ... Rc8 11.Bb3, Ne5 12.0-0-0, Nc4 13.Bxc4, Rxc4 14.h5, Nxh5 15.g4, Nf6 16.Bh6, ...

After 16.Bh6

White has plenty of alternatives at this juncture. For example, 16 e5, 16 Nd5, 16 Nde2, 16 Nb3, 16 Kb1, 16 Qh2, 16 b3, and 16 Rdg1. The move played by Nimzo 7.32 is the most popular/fashionable line presently and can thus be considered today's mainline. White sacrifices a pawn in order to exchange off Black's powerful dark squared Bishop.

Now here's a philosophical point to ponder... given that your high rated opponent favors a certain line and is thus assumed to be well versed/experienced with it, does it make sense to play directly into that highly theoretical line? I personally would think twice about doing this if I were White. All things being equal, unless I had a lot of experience battling Dragons or had a theoretical improvement up my sleeve, I'd look to enter some less analyzed side line of the Yugoslav Dragon, or better yet, avoid the Yugoslav Dragon lines altogether. Still, Nimzo 7.32 has no fear factor so we will likely have a battle that will test opening theory and the skills of the combatants.

16. ..., Nxe4

Black's reply is tactically forced because passive play is punished here: 16... Bh8?? 17 e5, Nxg4 18 fxg4, Bxe5 19 Nf5, Re8 20 Nd5, Rxg4 21 Bg7, Bxf5 22 Bxe5, h5 23 Bc3, Qc8 24 Rhe1, Re4 25 Rxe4 followed by Qd4 wins.

17.Qe3, ...

White's reply too is forced since 17 Nxe4?, Rxd4 18 Qh2, Be5 19 Qh4, Rxd1+ 20 Kxd1, f5 21 Bxf8, Kxf8 22 Qxh7, Qb6 gives Black a clear advantage.

17. ... Rxc3

Black loses after 17... Nf6?? 18Bxg7, Kxg7 19 Qh6+, Kh8 20 Ne4 since Black's Knight on f6 will be exchanged, leading to an eventual checkmate.

Similarly, 17... Nxc3?? 18 Bxg7, Kxg7 19 Qh6+, Kf6 and here many books erroneously quote Rolf Schwartz with 20 g5+ winning. Frankly I don't see any White advantage at all after 20... Ke5 21 Rhe1+, Kd5 22 Ne6+, Nxd1 23 Nxd8, Rxd8 24 Rxd1+, Kc6 25 Qxh7, Be6 when an approximately equal position is reached after Black plays ... Kd7 and ... Rc8. Instead of 20 g5+??, White should play 20 Qh4, Kg7 (20... Ke5?? 21 bxc3, Rxc3 22 Nc6+, Bxc6 23 Rhe1+ checkmates) 21 Qxh7+, Kf6 22 g5+, Ke5 23 Rhe1, Kd5 24 Ne6, Nxd1 25 Nxd8, Nf2 26 rxe7, Bf5 27 c3, Nd3+ 28 Kd2 when White gains a tempo on Schwartz's line and now can demonstrate a win.

This is just one of many examples where published opening theory is wrong. Thus it's important to analyze everything yourself; don't trust anything until you've checked it. Also, even if you believe the analysis is correct, can you trust the assessment? Often I don't. Therefore it's more important to have your own ideas about the openings than to merely follow others.

18.bxc3, Nf6 19.Bxg7, Kxg7 20.Rh2, ...

After 20.Rh2

I'm really impressed with Nimzo 7.32's opening book so far. It has already gone 20 moves deep into this line and maybe will continue even deeper. A quick search of the 2 million game database of ChessLab shows this position was reached in 68 games since 1971. To date, White has an overwhelming score of 31 wins, 24 draws, versus only 13 losses. Thus the Nimzo 7.32 team were wise to have selected this line for it. I just read on the Computer Chess Resource Center web site that several chess computer experts independently stated that Nimzo 7.32 has the best opening book. Don Maddox, President of ChessBase USA, clarified this by writing that its opening book isn't necessarily the biggest, but it's very well tuned for that chess engine.

Nimzo 7.32's last move threatens to treble heavy pieces on the h-file. More common ideas are 20 Qh6, Kh8 21 Ne2, Rg8=or 20Ne2 (White plans Ng3 and g5), Qa5 21 g5, Nh5 which is drawish. Thus Black may want to consider 21... Ng8!? if he's playing for a win. Although the text move is not a theoretical main line (some older books wrongly question this move), it has amassed overwhelming success as seen from the data base survey.

20. ..., Rg8

I have never been an advocate of selecting moves based upon raw database game scores/results. Instead, I try to find the best moves first, and only later compare my choice to what theory says and last to what the databases imply. Black has several choices here, such as 20... Qc7 as advocated by several Dragon experts such as Sapi and A. Schneider. But I think White is better after 20... Qc7?! (not !, as Sapi and Schneider advocate) 21 Ne2, Qc4 22 Rd3, Qxa2 23 Qh6+, Kg8 24 Nf4 when the threats of g5 and then Nd5 give White an edge. Other Dragon experts such as GM Gufeld and IM Golubev have independently advocated 20... Rh8, which I think does provide equality. However, I successfully played a similar position a couple years ago and favored the text idea then. Black plans the odd looking ... Kh8 (or even ... Kf8) and ... Rg7 when Black seems to have an impregnable fortress, freeing the remaining forces for an assault upon White's weakened Queenside. Then the Rook is subsequently freed via ... f6, which should result in gxf6, exf6 and then after Black's light squared Bishop relocates, the Rook swings over to c7 with powerful effect. This line too is blessed by opening theory. I'll admit my alarm at the overwhelming success White had in ChessLab after his 20th move, so I was curious what the data says after my reply. The 2 million game database now says that 26 games were on record with White winning 12, drawing 4, and losing 10 times. This is much more satisfactory. The low draw rate suggests this line is razor sharp and the 10 Black wins in this line compares very well to the 13 total Black wins after White's 20th move! Thus the data suggests this is the only line that tests White. I can now sleep at night feeling vindicated by both theory and history.

21.Ne2, Kh8

Although this move has the blessing of most opening books, it was only played in two games in the ChessLab database, a win and a draw for Black. Perhaps the minimal frequency of this move in practice is the reason that Nimzo 7.32 is no longer in it its opening book. Other acceptable lines involve: 21... Bc6, 21... Be6, and 21... Kf8. Instead 21... Qa5?? 22 g5, Nh5 23 Qxe7, Be6 (23..'Bc8 24 Rxh5 wins, Short-Ernst, Subotica 1987) 24 Rxh5, gxh5 25 Nf4, Kh8 26 Qf6+, Rg7 27 Rd5 wins.

22.g5, ...

After calculating for 12 hours and 23 minutes Nimzo dislodges my Knight. My limited perception of computers is that they are pawn grabbers, so I expected 22 Qxa7. My limited perception of humans is that since they played 21 Ne2, they would follow that idea up with the logical 22 Ng3 which prevents Black from ... Nh5 after a subsequent 23 g5.

22. ..., Nh5

Black's alternative was 22Ne8 23 Qa7!?, e5 24 f4 when all three candidates I examined led to a clear White advantage. Hence 24Qc7 25 fxe5 (25 Qa3 gives White an edge), dxe5 26 Nf4 is very strong for White, as are 24Be6 (24…exf4 25 Qd4+, Rg8 26 Qxf4 gives White a clear advantage) 25 Qxb7, Rg7 (25…Qa5?? 26 Rxh7+, Kxh7 27 Qh1+ leads to checkmate) 26 Qc6, giving White a clear advantage. After the text move, I assess the position as unclear, but White has to be very careful. This position is typical of many Yugoslav Dragons; White is safe if he doesn't push too hard for victory and simply “battens down the hatches”. Instead, if White tries to throw a punch, then an even harder counter-punch often hits him. This has to be psychologically difficult for White humans to cope with since the very reason they played the Yugoslav Attack is to “sac, sac, and mate”. That mentality seemed justified in the decades from the 1950's-80's, but loyal Dragon advocates have now found ways to help their Dragon grow a new head after one got lopped off.

23. Rdh1, …

After 12 hours and 15/32 ply, Nimzo 7.32 finds what I think is the best move, threatening 24 Ng3. Now, Black must play carefully to avoid getting an inferior position. Nimzo 7.32 evaluated the position as 0.07 pawns in its favor, very much in line with my assessment of unclear. This striving for the initiative is very impressive. I was surprised when the computer played this because I assumed chess engines were so materialistic that it would grab material with 23 Qxa7. It rejected that move due to 23... b5 24 Kb1, which it evaluated as 0.03 pawns in my favor. Instead, I would likely have played 23... Bc6 24 Qe3, e5!?.

23…, Rg7

As mentioned, White's last move restricted Black's options. I saw only three candidate moves. Of those, the other two were unsatisfactory. For example, 23…Bf5 24 Ng3, Qa5 25 Qxe7, Qxc3 (25…Qa3+ 26 Kd2, Rc8 27 Qe3, gives White a clear advantage) 26 Qxb7, Bxc2 27 Qb2, Qxb2 28 Kxb2 gives White a clear advantage. Worse yet, 23…Qa5 24 Ng3 (24 Qxe7, Be6 25 Ng3, Qa3+ 26 Kd2, Rc8 27 Ne4, Qa5 =), Qa3+?? (24…Bf5 transposes to 23…Bf5) 25 Kd1, Qxa2 26 Qxe7, Bc6 27 Nxh5, Bxf3+ 28 Kc1, Bxh5 29 Qf6+, Rg7 30 Re1 wins for White. Hence the text move, which was anticipated by my 20th move, seems forced.

24. Qxa7, …

Nimzo 7.32 calculated for 11 hours and 30 minutes to a depth of 15/32 ply and assessed the position as 0.09 pawns in Black's favor. In short, it thinks it has equality. I guess Nimzo 7.32 couldn't resist its computer "instincts" to grab material. After all, Black has a preponderance of pawns on the Kingside that will start marching toward promotion squares in the endgame. So with the removal of Black's a-pawn, White has hopes of rushing it's a-pawn forward at some point. This is double-edged though because Black now has an open file from which to mount a future assault upon White's King. Although this is a matter of personal style, I would have tried to activate the White Knight before Black prevents this with …e5. So the best line I can find for White is 24 Nf4, Ng3 25 Rg1, Nf5 26 Qxa7, Bc6 27 Rh3, e5 28 Nd3, b5 (28…f6 29 fxf6, Qxf6 30 f4, e4 31 Nb4, Qe6 is unclear) 29 Nb4, Bd7 when White has good prospects for equality. Nimzo also analyzed 24 f4?!, a6 25 c4, and now missed that simply 25…Be6 gives Black the advantage.

24..., e5!

Black now has the initiative. The text move prevents White's Knight from developing to either d4 or f4, while starting mobilization of Black's Kingside pawn majority. The alternatives, which fail to stop White's Knight are punished as follows: 24... Bc6? 25 Nd4, Bd5 (25... e5 26 Nxc6, bxc6 27 Rd2, d5 28 c4, Qxg5 29 cxd5, cxd5 30 Qa8+, Rg8 31 Qxd5 gives White an edge) 26 c4, Bxc4 27 Nf5, gxf5 28 Rxh5, Qc8 (28... b6 29 f4 is better for White) 29 Qd4, Kg8 30 Qxg7+, Kxg7 31 Rxh7, Kg6 32 f4, Qf8 33 R1h6+, Qxh6 34 Rxh6+, Kg7 35 a4 when White has the advantage. Also 24... Be6? 25 Nd4 (25 Kb2?!, Bd5 26 Qe3, Qa5 [26... e5!?] 27 Qxe7, Qxa2+ 28 Kc1, Bxf3 29 Rxh5, Bxh5 30 Qd8+ draws by perpetual check), Bd5 transposes into the 24... Bc6 line.

25. Qxb7, ...

Nimzo 7.32 calculated 9 hours and 40 minutes to a depth of 15/32 ply, assessing the position as 0.09 pawns in my favor. The text move is the only logical follow through to its previous move. Again, White clears a path for the advancement of its Queenside pawns, in anticipation of Black doing the same on the Kingside.

25..., Qxg5+

After 25.Qxb7

Black now had to decide whether it was better to first remove White's g-pawn with check to prepare for the eventual advance of Black's Kingside majority (the text move plan), or first prevent the advance of White's passed pawn and potential passers on the Queenside by threatening the a-pawn. The answer to this question is reached only via deep calculation of both options. Here is my analysis of the rejected line. After 25... Be6, White has three candidate lines:
  • 1) 26 Kb1?, f5 27 Qb4, Ra7 28 a3 (28 a4, Qd2 gives Black a clear advantage), Ra6 when Black has a clear advantage.
  • 2) 26 Rd1?!, Bxa2 27 Qb4, Qxg5+ 28 Kb2, Be6 when Black has an edge
  • 3) 26 f4!, exf4 (26... Bxa6 27 Qb4, Be6=)27 a4, Qxg5 28 Qb8, Rg8 29 Qxd6, Rd8 (29... f3 30 Qf4!, Qe7 31 Qd4+, Rg7 32 Nf4 is unclear) 30 Rxh5, gxh5 31 Qxf4, Qxf4+ 32 Nxf4 when White has equalized.

The text line forces White to make immediate decisions regarding conflicting goals, King safety and Queenside pawn advancement. This decision would burden most humans who seek a meaningful long-range plan. Let's see how the calculating machine fares.

26. Kb2!, ...

Nimzo 7.32 solved the strategic problem that I gave it on the last move correctly. It calculated for 12 hours to a depth of 16/34 ply and assessed the position as 0.06 pawns in my favor. Originally I thought Black has an edge here, but this assessment is now seen to be wrong, as White is equal. Initially I thought White should have continued with 26 f4, Qe7 (26…exf4? 27 Qxd7, f3+ 28 Kb2, fxe2 29 Rxe2 when White has a clear advantage) 27 Rd1, Nd6 (27…exf4!? 28 Qb8+, Be8 29 Qxd6, Qe3+ 30 Kb2 is unclear) 28 Qb4, Ne8 29 fxe5, Qxe5 30 Qf4, Qa5 31 Kb2, Be6 32 a3, g5 33 Qb4 when the position is unclear to me, although subjectively I prefer Black. Black's Kingside pawn mass will be difficult to stop, once it begins its advance..

26..., Qd8 27. Rd1, ...

Nimzo 7.32 calculated for 13 hours and 40 minutes to a depth of 15/32 ply to find what I believe to be the only playable move for White. Interestingly, Nimzo assessed the position as 0.02 pawns in its favor. Nimzo 7.32's top two rejected candidate moves were 27 Kc1 and 27 Ka1. I think both moves are truly silly. Why move the King again? Another option was 27 Ra1?, as it prepares to advance the passed a-pawn. However, 27…f5! (27…Be6 28 a4, f5 retains Black's edge) 28 Qd5, Qb6+ 29 Qb3, Qc6 30 Rf1 (20 Rf2?, Be6 31 Qxe6, Qb6+ when Black wins), Bc8 31 Qa3, Rb7+ 32 Kc1, Nf6 when Black has a clear advantage. This is just an example of how careful White must be in order to avoid quickly getting a bad position.

27..., Be6

I considered no other move since the text move is an integral part of my long-term plans. This multi-purpose move:

  1. targets White's a-pawn in conjunction with ... Qa5
  2. threatens the White Queen after ... f5
  3. prepares ... d5
  4. in moving the Bishop the d-pawn is now protected

Now White has a lot of candidate moves to consider. A human would now try to find a long-range plan for White. Let's see how our tactically inclined computer solves this problem.

28. c4, ...

Bravo! Nimzo 7.32 selected the best move from the field of candidates. This is impressive in that it had to calculate that its pawn offer was proper; something that many humans find difficult to qualify. The chess engine calculated for 19 hours and 25 minutes to a depth of 15/32 ply. Nimzo 7.32 assessed the position as only 0.14 pawns in my favor. However, since it previously favored itself, the trend is thus encouraging.

Nimzo 7.32's second candidate line was 26 Qc6, d4 29 c4, d4 30 Ka1, Bf5 31 Nc3, f6 32 Nd4, Ra7 33 Rb1, which it assessed as giving Black the edge. Instead, I think the position is unclear. My plan was 29... Qb8+ 30 Ka1, dxc4 31 Rb1, Qd8 when Black does indeed have the edge. Like me, Nimzo 7.32 rejected the logical looking 28 a4, Rg8 29 c4, Bxc4 30 Nc3, Nf4 31 Ne4, d5 32 Qxf7, Qb8+ 33 Ka1 when it says Black is better. I think so too after 33... Rg7. However, a summary of my notes reads, 28 a4??, f5! 29 Qb4, d5 30 c4, d4 31 a5, Rc7 32 Qb6, Bxc4 33 Nxd4, exd4 34 Rxd4, Qe7 35 a6, Nf6 wins. Of course White's moves aren't forced but the other lines I examined led to the same conclusion.

28..., Bxc4

I think Black had only two legitimate options to consider: the text move and 29... f5. The later move allows White to force Black into a draw by repetition. Since I want to win this game, the text move is the only winning attempt. Here's a summary of my analysis of the alternative:

28... f5!? 29 Qc6 (29 Qb4), Rd7 (29... Qb8+) 30 Nc3 (White plans 31 c5), Qe7 31 Nb5, d5 and now White can force Black to accept a draw by perpetual check upon White! For example, 32 cxd5, Bxd5 33 Rxd5, Qb4+ 34 Ka1, Qe1+ draws by perpetual check. I can't find anything better for Black here. The same conclusion is reached after 32 Rhd2, Nf4 33 cxd5, Bxd5 34 Rxd5, Nxd5 35 Rxd5, Qb4+ drawing again. Since I don't like the fact that White can decide Black's destiny, I only glanced at 32 Qc8+, Bg8 (32... Rd8?? 33 Qc7 wins the e-pawn) 33 c3 (this prevents Black's perpetual checking scheme, and is thus a winning attempt), Rd8 34 Qc6, Nf4 35 Rhd2, e4 36 fxe4, fxe4 37 Qb6, Nd3+ when Black is better. Again, this last line would be attractive if it were not for the fact that White can force Black into taking the draw in the other two lines.

After the text move, the position is very difficult for me to assess since it is so unbalanced and so very sharp. The potential for heavy piece activity in middlegame situations exists, when instead I'd like to play for an endgame. Then my wall of connected passed pawns advances to victory on the Kingside, as long as I retain a piece to prevent White from queening on a8. I've won many Dragon Sicilians via this motif. If I've done my homework correctly, White will discover that Black is better here. Still, this is the type of wild position where calculating skill is paramount, so it's quite possible that I missed something big while my calculating opponent has it all mapped out. Let's see what happens.

29. Qb4, ...

This move seems forced. Nimzo 7.32 calculated for 17 hours and 10 minutes to a depth of 17/35 ply and assessed the position as being 0.27 pawns in its favor. This is discouraging because it favored me slightly on the last move. Perhaps it thought 28... f5 was the superior continuation. Regardless, Nimzo 7.32 and I agree that White's alternatives were inferior. For example, Nimzo's next best line was 29 Nc3, f6 (I think 29... f5 30 Qb4, Bg8 31 Rxd6, Qg5 is unclear) 30 Qb4, Bg8 31 Rxd6, Qa8 32 Rf2 (I think 32 Rhd2, Qxf3 33 Rd7, Qa8 gives Black an edge), Qa7 33 Rb6, Rc7 34 a4 (Nimzo ends its analysis with the assessment of 0.21 pawns in Black's favor), Nf4 when Black is clearly better due to the threats of ... h5-4 etc. Nimzo 7.32 also analyzed 29 Qc6?, Qb8+ 30 Ka1, d5 31 Nc3, f5?! (I think 31... Nf4+ gives Black a clear advantage) 32 Nxd5?! (White's best line is 32 Rb1, Qd8 when Black has a small advantage), Bxa2 33 Nb6, Ra7 32 Kb2, Rb7 when Black has a large advantage.

29..., Qc7

This is the only line I considered, although 29... Qc8 can transpose. The plan is to meet 30 Qxd6 with 30... Qb7. In so doing, Black keeps the pressure on White's Queenside and inhibits the advance of White's pawns while threatening the pawn on f3. It thus seems to me that Black has the initiative and just needs time to mobilize the connected Kingside pawns for a decisive advance. The line that I believe will get played is pretty wild and difficult for me to assess since it falls into the boundary between middlegame and endgame motifs. Fasten your seatbelt because it will be a bumpy ride ahead.

30. Nc3!, ...

Surprise! In activating this once passive piece, Nimzo 7.32 deviated from the complex line I wrote about in my last comment. Nimzo 7.32 calculated for 19 hours and 30 minutes to a depth of 16/34 ply and assessed the position as 0.20 pawns in its favor, meaning the game is equal. The line I originally thought best for White is summarized with 30 Qxd6, Qb7+ 31 Ka1, f6 (an even wilder line is 31... Qxf3!? 32 Qc7, Bxe2 33 Rd8+, Rg8 34 Qxe5+, Nf6 35 Rxg8+, Kxg8 36 Rxe2, h5 37 Re3, which is totally unclear to me) 32 Rxh5!, gxh5 33 Qxf6 (White plans Qd8+), Qe7 34 Rd8+, Bg8 35 Qxe7, Rxe7 36 Ng3, h4 37 Nf5, Rc7 38 Nxh4, Rxc2 39 Kb1, Rf2 40 Nf5, Rxf3 41 Nh6, Rf1+ when it seems White can force a draw. Instead Nimzo 7.32's line deviated with 31 Kc1?!, Qxf3 32 Qxe5, f6 33 Rd8+ and now it missed Black's continuation of 33... Bg8 34 Qe6, Qa3+ 35 Kd1, Qb2 36 a4, h6! (Black breaks the pin with ... Kh7) when Black is better.

After 30. Nc3!

Nimzo 7.32 correctly rejected 30 Rxd6??, but for the wrong reason. After 30... Rg8 31 Rb6, Bd5 32 Rf2, Rc8 33 c3, Qa7 34 a4, Nf6 35 Rf1 it says that Black is better by only 0.19 pawns (meaning equality), but 35... Bc4 wins outright. This may be another example of the horizon effect.

30..., Rg8

No other line is worth considering. The tempo White spent on its last move allows Black time to activate this once passive piece with the threat of 31... Rb8. Black has the better game.

31. Qxd6, ...

Nimzo 7.32 spent 19 hours and 30 minutes in searching to a depth of 16/34 ply. Now it really likes its position since it upgraded its evaluation to 0.31 pawns in its favor, the equivalent of an edge. Regardless, we both agree that the alternatives clearly favor Black. For example, Nimzo's second choice was 31 Kc1, when I would have replied with 31... Be6 which gives a Black edge, rather than the silly line that Nimzo continued with after 31... Qc5?!. Nimzo's 3rd best selection was 31 Ka1, Be6 32 Rxd6, Rb8 33 Qa3, Qb7 34 Rhd2, Kg7 35 Rd8, Rd8 36 Qf8??, which loses. The less said about that line, the better. Actually the only alternative I considered was 31 a4 (31 Rxd6??, Rb8 wins), Rc8 when Black has the edge.

31..., Qb7+!?

This is Black's only line to pressure White., Instead 31... Rb8+ 32 Ka1, Qxd6 33 Rxd6, Ng3 34 Nd5, h5 (34... Nf5!?) is unclear. However it felt a little dubious to allow the Queen exchange so quickly after 31... Rb8+. Instead, the text move forces both combatants to perform incredibly complex and deep calculations. This position is so complex for me that I'll probably never know whether my 31st move was correct or not.

32. Ka1, ...

Now White threatens 33 Rxh5, gxh5 34 Qf6+, Rg7 35 Rd8 checkmate. The alternative is 32 Kc1?, Qxf3 33 Qxe5+, f6 34 Qd4, Rc8 (This position differs from the text line since White's a-pawn in now en prise.) 35 a4, Be6 when Black has the edge. Nimzo 7.32 calculated for 19 hours and 45 minutes to a depth of 17/35 ply. It now assesses the position as being 0.53 pawns in its favor. Previously I'd been critical of Nimzo 7.32's assessments, which seemed biased in its favor. However, I now agree that White does indeed have an advantage here, although the amount of advantage is difficult to gauge. It is probably merely an edge, but could also be totally won!

I find this position to be exceptionally complex, since both sides conspired in unbalancing a formerly equal position. Both sides have passed pawns on opposite wings. While White's pawns presently have piece support and are more mobile, Black's pawn mass is connected. While White will soon advance his passed pawns, Black can stop them with either a blockade or a piece sacrifice. However, once Black's slow moving pawn mass advances, the fact that the pawns are connected discourages attempts at blockade. If White sacrifices material to stop a pawn, another pawn will take its place with a recapture. Therefore, this is really a tortoise and hare race, which is extremely difficult to assess. While fatal risks are there for both sides, the White position is easier to play.

32..., Qxf3

This was my original plan, but after checking and rechecking my notes, I kept jumping back and forth between the text move and 32... f6. It seemed every time I looked deeper or considered a new line, I changed my opinion regarding the best move. Finally, after many hours of research, I concluded that my original plan, which clears a path for my pawn-wall to advance, is both sharpest and best. Instead, I now believe that with best play 32... f6?? loses. Here's a summary of my analysis.

32…f6?? can now be met by the immediate assault 33 Rxh5??, gxh5 34 Qxf6+, Qg7 (34…Rg7 35 Rd8+, Bg8 36 Nd5, h4 37 Qxe5, Qf7 38 c4, h3 39 Rb8! [39 c5, h2 40 Qxh2, Qxf3 41 Qe5, Qf1+ is equal], h5 40 Rb6 gives White a clear advantage) 35 Qc6 (35 Qxg7+, Kxg7 36 Ne4, Be2 37 Rg1+, Kh8 gives Black a clear advantage), Bf7 36 Ne4 (36 Ne2?? [White plans Rg1], e4+! [36…, Bxa2 37 Kxa2, Qa7+ 38 Kb3, Rb8+ 39 Kc3, Qa5+ 40 Kc4, Qa2+! draws by perpetual check] 37 c3, e3! 38 Rg1, Qe5 39 Rxg8+, Kxg8 40 a4, Qd5 wins for Black), h4 gives Black an edge.

Clearly White must instead respond with 33 Rh4!, Bf7 34 Nd5 (34 Rxh5??, Qxf3 35 Rd3, Qf1+! 36 Rd1, Qf3 37 Rd3 draws by repetition), Bxd5 (34... Qc8 35 Rxh5, gxh5 36 Qxf6+, Rg7 37 Qxe5, Qxc2 38 Rg1, Bg6 39 Ne7 [White plans Nf5] wins) 35 Qxd5 and now Black has a choice of plans. One involves allowing a Queen exchange and the other avoids it. The former line continues with 35... Rb8 36 Qxb7, Rxb7 37 Rb1, Rc7 38 Kb2!! (The natural looking alternative of 38 c4, Nf4 39 Rb4, h5 40 a4, Ne2! 41 Kb2, Nd4 42 Rh3, Rc5 43 Kc3, Ra5 44 Rb8+, Kg7 45 Rb7+, Kh6 46 Kb4, Nc6+ is unclear), Nf4 39 a4, h5 40 a5, Rb7 41 Ka2, Ra7 42 Rb5, Ne2 43 Rc4, Nd4 44 Rbc5, Ra6 45 Rb4, Kg7 46 Rb6, Ra7 47 a6, Kh6 48 Ra5, Nxc2 49 Rb7, Ra8 50 a7, h4 51 Rb8, Rxa7 52 Rxa7, f5 53 Kb2, Nd4 54 Rh8+, Kg5 55 Ra3 wins. The other line continues with 35... Qa7 36 Rb4, Rg7 (36... Ng7 [36... Nf4? 37 Rb7 wins] 37 Rb7, Qa8 38 Qb3, h5 39 Qf7 wins) 37 Rdb1!, h6 38 Rb8+, Kh7 39 R8b7, Qa3 40 Rd1 and White wins since Black is tied up in a sort of zugzwang position. I think you now can understand how sharp and complex this position is. The text move sharpens matters even further.

33. Qxe5+, f6

These last two moves are the "only moves". Nimzo calculated for 19 hours to a depth of 16/34 ply, at an average speed of 377 kilonodes/second. It assessed the position as 0.57 pawns in its favor.

34. Qd4, Rc8

Again, I think these are the "only moves". This time, Nimzo 7.32 calculated for 21 hours to a depth of 15/32 ply, at an average speed of 381 kilonodes/second. It now assesses the position as being 0.77 pawns in its favor. I find it interesting that Nimzo calculated for a longer period of time and at a faster speed than on its previous move, yet wasn. t able to search as deeply.

35. Ne4, ...

The assessment gap widens as Nimzo 7.32 now believes it is better by 0.83 pawns. It reached this conclusion after calculating for 18 hours and 30 minutes. The chess engine reached a search depth of 15/32 ply at an average speed of 385 kilonodes/second. I expected the text move, which I too believe best, but also thought 35 Rhd2 was playable, although the resulting complications were quite unclear.

35..., Ra8

One thing I've learned from playing the Black side of Dragon Sicilians most of my chess life is that there is a very fine line between success and failure. Even in such sharp positions, innocent looking moves can evolve into lines that totally destroy the evaluation of positions. I think this may be an example; Black must chose between two related looking paths that have unrelated motifs. One path seems successful, leading to unclear positions, while the other path apparently gets ground down into an inferior position.

Black's primary task here is to activate his pieces so that they effectively blockade White's passed pawns while preparing the mobilization of Black's pawn mass. This is complex because there are so many heavy pieces still on the board with lots of open lines available. Black shouldn't construct a rigid fortress to blockade White's pawns because White's superior mobility would then enable him to mount a successful mating attack. Meanwhile, Black's Knight is the least active piece on the board. Thus it seems that Black should examine two motifs: one involving the trade of Queens and the other avoiding that exchange. I concluded that the Queen exchange option (my text move anticipates this) was better than the plan that avoids the exchange. This was a difficult decision since the resulting positions were very complex. While certain lines were technically correct, tactical flaws totally destroyed them... and vice versa. So here's a summary of my analysis of the rejected option.

35... Qa3 36 c3, Rc6 37 Qd8+, Bg8 38 Rb2, Ra6 39 c4, Ra8 40 Qd4, Qa6 41 Nd6, Qa5 42 Qd2, Qc5 (42... Qe5 43 Qd4, Qe6 44 Rb7, g5 45 Rdb1, Nf4 46 Rh1! gives White an edge) 43 Rb7, Qe5+ 44 Qb2 gives White an edge.

Instead, 41..., Qa3?! 42 Qd2 (42 Rf1, Rf8 43 c5, Qa6 44 Rc1 gives White an edge), now leads to three options for Black.

  1. 42... Qf3 43 Rdb1, Qg3 44 Rb7 (44 Rb8?, Qe5+ is unclear), Qe5+ 45 R1b2, f5 46 Nf7+, Bxf7 47 Rxf7 gives White a clear advantage.
  2. 42... Qa6 43 Rdb1, Qa7 44 Rb7, Qc5 45 R1b5, Qg1+ 46 Kb2 (White plans Nf7+) gives White a clear advantage.
  3. 42... Qg3 43 c5, Qe5 44 Qd4, Qe6 (44... Qe7?? 45 c6, Qc7 46 Rc1, Be6 47 Rc5 [White threatens Rxh5] wins) 45 c6, Rxa2+ 46 Kb1, Ra6 47 c7, Rc6 48 c8/Q, Rxc8 49 Nxc8, Qxc8 50 Rb6 gives White a clear advantage.

So in a best case scenario for Black, the above lines allow a White edge. However after my text move, if I calculated and assessed the resulting positions correctly (that is a very big if), Black obtains positions that I can only assess as unclear. At least, I couldn't find any way for White to establish any dominance. Let's see how things evolve.

36. c3, Bb3

After 36. ... Bb3

Nimzo 7.32 calculated for 16 hours and 45 minutes to a depth of 15/32 ply, at an average speed of 385 kilonodes/second. The chess engine continued to upgrade its assessment, now believing it is better by 0.88 pawns.

37. Rb1!, ...

So this is the move that Nimzo 7.32 was "banking on" to justify it's very high assessment of its position! Nimzo 7.32 calculated for 19 hours and 45 minutes to a depth of 16/34 ply, at an average speed of 387 kilonodes/second. It now assesses the position as 0.92 pawns in its favor. Hmm... Nimzo's evaluation continues to grow in its favor. Sure, these types of positions favor computers over humans in OTB chess and probably in correspondence chess as well. Yes, the position is extremely complex; heavy pieces remain on the board with lots of open space to maneuver in. Yes, Black's Knight on the rim is inactive and his King can potentially be targeted. However, in many alternative lines for White, that Knight comes into play and Black's Kingside pawn mass advances with devastating effect, forcing White to sac pieces to try to stop them.

Because of the complexity of this position, it was difficult for me to anticipate White's plan. Instead, expecting White to pressure Black's Kingside, I originally thought that the only line that really challenged Black was 37 Rg1, Qf5 38 Rf2, Qd5 (this follows the plan of exchanging Queens referenced in my earlier commentary) 39 Qxd5! (39 Nxf6, Qxd4 40 cxd4, Nxf6 41 Rxf6, Rxa2+ 42 Kb1, Rd2=), Bxd5 40 Nd6, f5! 41 c4, Ra6 42 c5 (42 Nxf5 [42 Rd1!? is also unclear], Bxc4 43 Ne3 is unclear), Nf6 43 Rd2, Bc6 44 Kb2, Nd7 (44... Ra5?! 45 Nxf5 [White threatens Rd6], Bd5 46 Nd6, Rxc5 [46... Rxa2?? 47 Kc3, Rxd2 48 Kxd2, Nd7 49 Ke3, Kg7 50 Kd4, Bc6 51 Ra1, h5 52 Ra6 wins] 47 a4, Kg7 48 Ka3 [White plans Kb4] gives White a small edge) is totally unclear. If Black can mobilize his Knight so his pawn mass can advance, then Black is better, but that's a big if. However, I now see that Nimzo 7.32 did indeed play the best move. I think White will now leave a Rook on the Kingside to hamper a Black pawn advance, and then when the time is right, will quickly double Rooks on the b-file in and effort to either penetrate Black's position or to support the advance of White's pawns.

37..., Bg8

It's very difficult to anticipate White's moves since we have a wide-open board with plenty of heavy pieces remaining. Possibly White will double Rooks on the b-file, but if done immediately that can ease pressure on Black's Kingside, allowing Black to advance his pawn mass. Instead, I think White will keep a Rook on the Kingside and target Black's Queen with 38 Rf2. Let's see what Nimzo 7.32 comes up with.

38.Rf2, Qg4

Black's last move seems forced since 38… Qf5? 39 Rb6 loses. Black's Queen is headed for e6, where it threatens checkmate. On its last move, Nimzo 7.32 calculated for 16 hours and 45 minutes to a depth of 16/34 ply, at an average speed of 382 kilonodes/second. It kept its evaluation static at 0.92 pawns in its favor. Now that I've had more time to examine this position, I do indeed believe White has a clear advantage here. It hurts to admit that I'm faring poorly with my Dragon Sicilian, a defense to which I've strong sentimental attachments; one that's scored extremely well for me versus strong human opposition. My thoughts recall a book I read about 30 years ago, "Flowers For Algernon". Charlie, the protagonist, is a mentally retarded man who is being tested in a laboratory. He must solve a maze in competition with a laboratory rat, Algernon. The rat solves the puzzle faster, so while it's eating its cheese, Charlie starts to cry. When the doctor tries to comfort him, he cries, "Well, how would you feel if you were dumber than a mouse!" I now feel like Charlie. Here, a machine has played Dragon positions to date better than I have. That hurts! Still, there's a complex battle ahead and I'll keep fighting. I've gone on record stating that even the most powerful chess engines have not demonstrated to me that they have the ability to defeat a strong master, unless that master tries too hard to unbalance the position to win. Well, in hindsight, I may have created such a scenario. I've shown that Black could have accepted a drawing line at the 28th move. We can now see that my 31st move was too sharp; the unclear alternative is better. Also, my 35th move choice is now seen to be inferior. Still, I've no regrets since it's very much in my style to try to push positions out of balance in order to try to win. This involves great risk and with that comes an eventual loss. In the long run, I've won a lot more than I've lost so I don't plan to change. Let's see what happens next.

39. a4, Qe6

Nimzo 7.32 calculated for 18 hours and 30 minutes to a depth of 16/34 ply, at an average speed of 386 kilonodes/second. It played the only line to challenge Black. Instead, 38 Rfb2?, Qe6 39 a4, Ng7 allows Black's pawns to roll toward promotion. If White's Rook moves off the 2nd rank, then White gets checkmated. Nimzo 7.32 now upgraded its assessment to a 1.10 pawn advantage for itself.

40. Rb4, …

Nimzo 7.32 shows the correct technique. Premature removal of the Rook from the Kingside results in 40 Rfb2?, Ng7 when Black's pawns are ready to roll forward, creating an unclear position. Nimzo calculated for 18 hours and 20 minutes to a depth of 16/34 ply, at an average speed of 386 kilonodes/second. It now favors its position by 1.19 pawns& a clear advantage.

40. … , Ra6

After 40. … Ra6

Black's Queen and Rook tandem form a barrier to slow the advance of White's a-pawn. As such it seems forced. The only other logical candidate move I considered loses by force. For example, 40…Rc8? 41 Nxf6, Nxf6 42 Qxf6+, Qxf6+ 43 Rxf6, Rxc3. At first glance, Black looks OK here, but look a little further. Now 44 Kb2, Rc7 (44…Kg7? 45 Rxg6+ wins) 45 a5, Bd5 (45…Kg7? 46 Rd6 wins as 47…Bd6 is prevented. Black's Bishop can't get to the a8-h1 diagonal in time to attack the a8 queening square.) 46 Rd6, Bf3 47 Rbd4, g5 48 Rd7, Rxd7 49 Rxd7, g4 50 Rd3, h5 51 a6, Kg7 52 Kc3, h4 53 a7 (White threatens Rxf3), Ba8 54 Rd8, Bc6 55 Rd7+, Kf8 56 Rc7, Bf3 57 Rc4 (White threatens 58 Rf4+ and 59 Rxf3), Ke7 58 Rxg4 wins.

While matters are looking pretty grim for the human, I'd still feel relatively confident of doing well if my opponent were human, since there are plenty of tactical traps for White to fall into and lose or at least allow Black to equalize. However, Nimzo 7.32 has performed admirably in this game to date, and surely won't miss any short-term tactical shots for either side. There are almost no long-range strategic concepts available for Black to turn the game from a calculating tactical battle into a strategic planning battle, an area where humans generally outplay computers. However, even if such lines were available to me, I've pledged to continue to play in the style that I would against a human. As such, I'd look to keep things sharp/complex and hope he or she misses a tactical shot here or there.

41. c4?, …

Nimzo 7.32 calculated for 19 hours to a depth of 15/32 ply, at an average speed of 387 kilonodes/second. It now favors its position by 1.07 pawns, a small retraction of the higher assessments it gave on its last two moves. I think White had 2 primary candidate moves to evaluate, the text and 41 Rb8! In fact, the natural looking text move was the first one I prepared for, since it advances a passed pawn and blocks Black's checkmate threats, thus freeing the Rook on f2 to move off the second rank. However, I subsequently found 41 Rb8! to be an even stronger move and thus had been worried about meeting it. Its point is to probe Black's defense prior to any pawn advances. Therefore, the majority of my time was spent on finding a reply to 41 Rb8! I eventually found three candidates; none of them satisfactory.

The weakest defense is 41…Qc6?? 42 Nc5, Ra7 43 Qc4, Rg7 44 a5, Qc7 45 Rfb2, Ng3 46 a6, h5 47 R2b7 when White wins.

The next best line is 41…Qe7?! 42 Rfb2, h6!? (42…Qa3+? 43 Kb1, Qxa4 44 Rxg8+, Kxg8 45 Qd8+, Kf7 46 Rb7+ wins) 43 Qc4, Qe6 44 Qxe6, Rxe6 45 Nc5, Re1+ (45…Rc6? 46 R2b5, Kg7 47 R8b7+, Bf7 48 a5 wins) 46 Rb1, Re3 (46…Re2 47 Rd1 also gives White a large advantage.) 47 c4, f5 48 Kb2, Kg7 49 Rb3 when White has a large advantage.

The best Black defense seems to be 41Rc6 42 Re2 (42 Ra8?, Rb6 is unclear), Rc4 43 Qa7, Rc8 44 Rxc8, Qxc8 45 Qd4 (45 a5??, Qc4 gives Black the advantage, so White still needs to be careful too!), Qe6 46 Rf2, Ng7 47 Rb2, f5 48 Nf6 gives White a clear advantage.

So what's the matter with the natural looking text move? I think Black now seems to equalize. In fact, there are plenty of opportunities now for White to lose! At least, that's what I think at the present time. I just hope Nimzo 7.32 doesn't have some rude surprise in store for me. This position is very complex and thus really difficult for me to analyze with much confidence.

41. …, Qc6

Black had four candidate lines to examine here. My first choice was 41…Qe7? 42 c5, h6 (See 41…h6 for an explanation of this move. Instead, 42…Re6 43 Rc2, Re5 44 c6, Rd6 45 Qc3, Rd1+ 46 Kb2, Qe6 47 Rc4, Rd8 48 c7, Rd8 49 Qd4 gives White a large advantage.) 43 Rfb2, Ng7 44 Rb6, Ne6 45 Qf6+, Qxf6 46 Nxf6, Rxa4 47 Kb1, Bf7 48 Rb7, Kg7 49 Ne8+, Kf8 50 Nd6, Rf4 51 c6, Rf1+ 52 Ka2 gives White at least a clear advantage.

The last line I examined was 41Rc6?? 42 c5, Rc7 (42…h6?? 43 Rb6 wins) 43 Nxf6, Nxf6 44 Rb6, Qe1+ 45 Kb2, Qe7 46 Qxf6+, Qxf6+ 47 Rfxf6, Rxc5 48 Rb5, Rc4 49 a5, Ra4 50 Rd6, Kg7 51 Rb7+, Kh8 52 a6, which wins for White.

As you may have noticed, … h6 figured into the above lines. It serves to create an escape valve for Black's King and also prepares … g5, which is presently impossible due to Nxg5. Therefore 41… h6!? 42 c5, (42 Nc5, Qe1+ 43 Rb1, Rd6 44 Qxd6, Qxf2 45 Nd3, Qd2 46 Rc1 [46 c5??, Qa2+ checkmate!], g5 is unclear), Qc6 43 Nd6, Ra7 (43… Ra5?? 44 Rxf6, Nxf6 45 Qxf6+, Kh7 46 Qe7+, Kh8 47 Nf7+ leads to checkmate) 44 Rfb2, Qh1+ 45 Rb1, Qf3 seems about equal.

The text move is very similar to the 41… h6!? line but tries to do without that move for the time being. After all, if White's pieces move away in response to Black's moves, there are lines where Black can play … g5 without the preparatory … h6 move. If needed, Black can play … h6 and can transpose back into this line. Another reason for choosing 41… Qc6 is that the very first line I analyzed led to a Black victory in a wild position. Maybe Nimzo 7.32 will play into that line too. Anyway, the best I can find for White after the text move is equality. But then again, I could be missing something big.

42. c5, …

Nimzo 7.32 finds the most logical defense for its a-pawn. Gee, it's so nice writing about White having to defend now, after being pushed around myself for the past few moves! Since passed pawns should advance, in principle, then the text move is more logical than the alternative of 42 Nc5 (blocking the advance of its own passed c-pawn), Ra8 43 Rfb2, g5 44 Rb8, Qh1+ 45 Rb1 (45 Ka2?, Rxb8 46 Rxb8, Qh2+ 47 Qb2, Qf4 48 Qb4, Kg7 49 Nb3, Bxc4 50 a5, g4 51 Rb7+, Kg6 52 Qc3, Qf1+ gives Black a clear advantage), Qf3 (45…Qh3 =), is probably equal. Nimzo 7.32 calculated for 18 hours and 15 minutes to a depth of 16/34 ply, at an average speed of 387 kilonodes/second. However, to my great dismay, it has upgraded its assessment now to 1.20 pawns in its favor. Gosh, one of us is really confused about this complex position. I just hope it's not me. I still think Black has equality.

42. …, h6

I elected to transpose back into the line I presented above for 41…h6. This was an easy choice since the only other candidate moves I considered looked terrible. For example, 42…Bd5? is meaningless after 43 Nc3, handing a clear advantage back to White. Worse yet, tricky Nimzo 7.32 set another trap for me, since 42…Ra5?? looks playable initially after 43 Rxf6??, Nxf6 44 Nxf6, Qxc5 45 Qxc5, Rxc5 46 Rb8 (46 Rb5?, Rc6 47 Nxg8, Kxg8 48 Kb2, h5 49 a5, Ra6 50 kc3, h4 wins for Black), Kg7 47 Nxg8, Ra5! 48 Kb2, Rxa4 49 Kc3, h5, which gives Black a large advantage. However, 43 Nxf6, Nxf6 44 Qxf6+, Qxf6+ 45 Rxf6, Rxc5 46 Rb7, Ra5 47 Rf8, Rxa4+ 48 Kb1, h4 49 Rbb8 wins for White. Even if Black can find an improvement to the last line, 43 Rd2, h6 44 Qd7, Qxd7 45 Rxd7, f5 46 c6, fxe4 47 c7, Rc5 48 Rb8 also wins. Wow! That looks pretty grim indeed for Black. So it seems strange to believe that a quiet move like 42…h6 can provide salvation in such a sharp position. Well, one thing is certaint; that at least one of us is mistaken about this position. Let's find out whom. I just hope to see a repeat of Ham-Nimzo 7.32, where the same chess program favored it position for a long time, only to later agree with me that it is in serious trouble. The difference there though is that game is largely strategic, while this one is tailor made for computer chess programs.

43. Kb1!, …

I was perplexed by Nimzo 7.32's highly favorable White assessments over the past couple moves. Since I found satisfactory Black defenses that are within Nimzo's search horizon, I questioned whether I missed something big. Therefore I searched for White improvements but kept finding strong Black lines. Then an idea popped into my mind. What if White offers material with the only goal of promoting the c-pawn? In this case, White can sac his a-pawn to draw Black's Queen and Rook tandem away from their defenses. In order to be successful, Black can't be allowed to capture the a-pawn with check, so the King must first move. Thus I discovered 43 Kb1! (maybe !! is appropriate) prior to receipt of Nimzo's move. Therefore the formal receipt of this fiendishly clever move ruined my day. Nimzo 7.32 calculated for 19 hours and 50 minutes to a depth of 16/34 ply, with an average speed of 391 kilonodes/second. It now favors its position by a large amount, 1.57 pawns. Drat!

43. …, g5

When I first discovered 43 Kb1!, and saw how strong it is, I subsequently found an adequate defense, and so smugly slept easily at night. Then, I found a White improvement on the 49th move that busts my defense. Although I'm still searching for a Black defense, the thematic text move figures into all my lines. Black prepares to advance the Kingside pawn mass and activate the Bishop with …Bh7+.

A big mystery exists though (aside from the mystery of whether an adequate defense exists). The mystery is that the strength of White's line doesn't manifest itself until after Nimzo 7.32's 17-ply search horizon has been exceeded. Perhaps it found the solution in its 34 ply extension. Let's see.

44. Rb6, …

An interesting, innocent, and serendipitous error occurred on the part of the computer user. Originally this position was accidentally fed to Fritz 6a, which calculated 5 hours and 45 minutes to a depth of 17/17 ply. It chose 44 Rh2 and favored White by 1.63 pawns. This position is too complex for me to come to any conclusions regarding best lines and assessments. I didn't thoroughly investigate 44 Rh2 because the announcement that the wrong chess program made this move caused me to stop my investigations until a move became official. However, I suspect that 44 Rh2? may offer Black a defense. For example, 44…Qe8 (44…Bf7?? 45 Qd6, Kg7 46 Qxc6, Rxc6 47 Rb7 [White plans Nd6], Kf8 48 Rd2 [48 Rxf7+??, Kxf7 49 Rxh5, f5 50 Ng3, Kg6 51 Rh2, Rxc5 favors Black with an edge], Ng7 49 Rdd7, Bg6 50 Rxg7, Bxe4+ 51 Kb2, Rxc5 52 Rbf7+, Ke8 53 Rxf6, Rf5 54 Rxf5, Bxf5 55 a5, Kf8 56 Rc7, Bd3 57 Kc3 wins) 45 Nd6, Qe1+ 46 Kc2, Bh7 47 Ne4 (47 Kb2?, Qb1+ 48 Ka3, Qc1+ probably draws), Ra8 (47…Bxe4? 48 Qxe4, Qxe4 49 Rxe4, Nf4 50 Re7, Rxa4 51 Rxh6+, Kg8 52 Kb2 [52 Rxf6??, Rc4+ 53 Kd2, Rxc5 is unclear but Black has apparent drawing chances] wins) 48 Rxh5 (48 Kb2!?), Qe2+ 49 Kb3, Qf3+ 50 Nc3, Qxh5 51 Qxf6+, Kg8 52 Rb7, Qg6 53 Qxg6, Bxg6 54 c6, Bf5 55 c7 still favors White but I didn't see any clear wins for White.

The line that worried me and to which I could find no defense for Black was 44 Nd6, (blocking Black's heavy pieces from defending f6 and preventing 45…Bf7 as a reply to 45 Rh2. More importantly, it covers the potential queening square. A summary of my analysis continues with 44…Ra7 (44…Ra5? 45 Rxf6, Nxf6 46 Qxf6, Kh7 47 Rb7+ wins) 45 Rh2, Bh7+ 46 Kc1, Bg6 47 Rb6!, Qf3! (the alternatives lose quickly, such as 47…Qxa4 48 Qxa4, Rxa4 49 Rb8+, Kh7 50 Rb7+, Ng7 51 c6 wins, and 47…Rxa4 48 Rxh5, Ra1+ [48…Rxd4? 49 Rxh6+, Kg7 50 Rxg6+ wins] 49 Qxa1+, Qxc5+ 50 Kd2, Qf2+ 51 Kc3, Qe3+ 52 Kc4, Qe6+ 53 Kb4, Bxh5 54 Rb8+, Kg7 55 Rb7+ wins). The point of Black's last move is the threat of …Qa3+. Now 48 Kb2, Qa8 (48…Qf4? 49 Qxf4, Nxf4 50 c6, Rc7 51 Rb8+, Kg7 52 Rb7 wins, as does 48…Kg7? 49 c6, Rc7 50 Nb5).

Now if White continues with the logical looking 49 Rxh5??,??, Qg2+ 50 Ka3, Bxh5 (50…Qf3+?? 51 Rb3, Qxh5 52 c6 wins) 51 Qxf6+, Kh7 52 Nf5, Qf3+ 53 Rb3, Qf4 54 Qxh6+, Kg8 55 Rb8+, Qxb8 56 Qxg5+, Kf8 57 Qh6+, Kg8 58 Qxh5, Qf4 draws. However, I then found 49 c6!, Rxa4 50 c7, Ra2+ 51 Kb3, Rb3+ 52 Kc4, Ra4+ 53 Rb4, Qa6+ 54 Kd5, which wins for White. Needless to say this discovery ruined my late evening analysis session and I got no sleep that night. Still, if the computer chess engine saw all of that on its last move, then I wouldn't mind admitting my inferiority to its skills and I'd bow in resignation to my superior.

Nimzo's text move is highly logical. I knew that such a move was likely to be played but just didn't know when. It too looks decisive. Nimzo 7.32 calculated for 20 hours to a depth of 16/34 ply, at an average speed of 396 kilonodes/second. It favors White now by 2.08 pawns. Clearly Nimzo 7.32 selected a better move than Fritz 6a did. This example and the play in the other 3 games suggests, but doesn't confirm, that Nimzo 7.32 may be a superior tactician than Fritz 6a over an extended calculation period. Note too that Fritz 6a calculated deeper than Nimzo 7.32, but found an inferior move.

Does Fritz's greater speed and search depth mean that it has less "knowledge" than the slower Nimzo 7.32, and thus calculated faster due to having to make fewer judgements? I'm not qualified to judge but I'd guess that the opposite may be true. Fritz 6a seems to perform better in closed positions, while we saw some strange moves from Nimzo 7.32 when there were no tactics to calculate. My assessment based upon only 4 games is: Fritz 6a performs better in closed positions and has a more accurate evaluation assessment while Nimzo 7.32 is the tactical wizard. This may not be the case when played at normal tournament time controls though.

44. …, Rxb6

I believe this move is forced since the alternative loses quickly. For example, 44…Rxa4? 45 Rxc6, Rxd4 46 nxf6, Nxf6 47 Rfxf6, Bh7+ 48 Kc1, Kg7 49 Rfd6, Rxd6 50 cxd6, Bf5 51 d7, Bxd7 52 Rc7, h5 53 Rxd7+, Kf6 54 Kd2, h4 55 Ke3 wins.

45. cxb6, Bh7

Nimzo 7.32 calculated to a depth of 18/37 ply during a period of 17 hours and 45 minutes. With an average speed of 400 kilonodes/second, it now assesses the position as favoring White by 2.28 pawns.

Black's reply seems forced. For example, 45…Bd5? loses immediately to 46 Nxf6+, Nxf6 47 Qxf6+. Again, I believe White wins in this game line too, but I want to see the computer actually play the moves I found in my notes. "You can't win games by resigning!"

46. Re2, Nf4

The infernal machine now values its position by 2.59 pawns after a search depth of 17/35 ply. It reached this depth after 17 hours and 30 minutes at an average speed of 398 kilonodes/second.

47. Qd8+, Kg7

Well, the handwriting has been on the wall for some time. Nimzo 7.32 now assesses its position as being 3.19 pawns ahead, after searching for 16 hours and 45 minutes. During that time, it searched to a depth of 17/35 ply at an average speed of 406 kilonodes/second. As much as I hate losing to a machine, especially in public, and worse yet, in my beloved Dragon Variation, I have to admire Nimzo 7.32's tactical calculating skills that brought it to this point. I think that very few correspondence masters could have held their own against it in trying to match attack versus counter-attack in such complex open-board positions. Very impressive indeed!

48. Qe7+, Kg6

The chess engine calculated 8 hours and 30 minutes to a depth of 16/34 ply, at an average speed of 406 kilonodes/second. In now favors its position by 3.20 pawns, so the evaluation continues to grow.

Meanwhile Black's only alternative was 48…Kg8? 49 b7 when White threatens to Queen the pawn with checkmate. Black is thus forced to give up the Queen for a Rook and the queening pawn via 49…Qxb7+ 50 Qxb7, Nxe2 51 Qd5+, Kg7 52 Qd7+, Kh8 53 Qd8+, Kg7 54 Qxf6+, but White has a simple win now. Instead 51 a5?, Nc3+ 52 Kc2, Bxe4 (52…Nxe4?? 53 a6 wins) 53 Qc8+, Kf7 54 Qc3, h5 55 a6, Kg6 56 a7 still requires some technique from White in order to win. I trusted that Nimzo 7.32 can be counted upon to find the correct winning method on its 51st move.

49. b7, Qb6+

Black's move is more than just a spite check, although the result is clear. Nimzo 7.32 calculated for 18 hours and 30 minutes to a depth of 16/34 ply, at an average speed of 405 kilonodes/second. It still favors its position by 3.20 pawns.

50. Rb2, Qg1+

Like the previous move, Black's move is more than just a spite check. This will be evident on White's next turn since Black thus creates an opportunity for White to err. Well, I can dream, can't I? Nimzo 7.32 calculated for 16 hours to a depth of 16/34 ply, at an average speed of 404 kilonodes/second. It now favors its position by 3.88 pawns.

51. Kc2, Qg2+

The reason that my previous moves were not merely spite checks is because an unwary White player (read: human) may fall for 51 Ka2??, Bg8+ 52 Rb3, Bxb3+ 53 Kxb3, Qe3+ when White has to accept a draw by perpetual check.

Instead, Nimzo 7.32 found the only correct move. It calculated for 19 hours to a depth of 18/37 ply, at an average speed of 400 kilonodes/second. It now favors its position by 3.96 pawns.

52. Nd2, Qc6+

Now that the computer chess engine has survived the aforementioned pitfall, there is little reason for me to play on. However, I hope that you are like me in wanting to see how Nimzo 7.32 executes the win. There are at least 2 paths leading to a White victory, but one of them is a bit messy in allowing some chances to Black if White gets careless.

Nimzo 7.32 calculated for 15 hours and 45 minutes to a depth of 18/37 ply, at an average speed of 402 kilonodes/second. It now favors its position by 4.91 pawns.

53. Kb1 (!), …

Per usual, Nimzo 7.32 finds the most efficient kill. It computed for 17 hours to a depth of 17/35 ply, at an average speed of 401 kilonodes/second. It now favors its advantage by 4.58 pawns.

Nimzo 7.32 had an alternative in 53 Kb3, Bg8+ 54 Ka3, Qc3+ 55 Nb3, Nd3 56 Qe4+, f5 57 Qd4, Qc7 58 Rc2 (58 Qxd3, Qe7+ 59 Ka2, Qxb7 60 a5, Qb4 61 Kb1, Bc4 62 Qxf5+, Kxf5 63 Nd4+ when White also wins), Qxb7 59 Qd6+, Kh5 60 Rh2+, Kg4 61 Qxd3, Qxb3+ 62 Qxb3, Bxb3 63 Kxb3, h5 64 a5 when White wins.

After 53. Kb1 1-0

However, things get a little messy after 53…Nd5 54 Qxh7+, Kxh7 55 b8/Q, Qe6 (55…g4 56 Nc4, h5 57 Rh2 wins) 56 a5, h5 57 Qb7+, Kh6 (57…Kg6 58 Nc4, h4 59 a6 wins) 58 a6, Nc7+ 59 Ka3, Qd6+ 60 Ka2, Nxa7 61 Qb6, Qxb6 62 Rxb6, Nc7 63 Rxf6+ when White still wins.

53 …, Resigns 1-0

There is no point in prolonging what's quite obvious. After 53…Qh1+ 54 Ka2, Bg8+ 55 Nb3, Nd5 56 Qe8+, Bf7 57 Qa8, Nc3+ 58 Ka3, Bd5 59 Qe8+, Bf7 60 b8/Q, Bxe8 61 Qxe8+. This is clearly won for White. I bow to Nimzo 7.32's superior computer skills. A most impressive performance!

In summary, my final score against Nimzo 7.32 in this 2-game match was 0-1-1 for the disappointing score of only ½ of 2. I conceded a draw in a much superior position through sloppy analysis caused by sleep deprivation in Ham-Nimzo 7.32. But Nimzo's victory here was totally justified. Nimzo 7.32 displayed highly precise tactical skills that were clearly superior to mine. Yes, I could have chosen a less computer-friendly line to play than the Black side of the Dragon Sicilian, but that's not the point. Nothing is gained by having me defeat the computer (as if I could!) and nothing is gained in this test by having me play for a draw. The only way to really test the computer chess engine's ability in Correspondence Chess is to continually play for the win by unbalancing the positions, even when the wide-open position was full of heavy pieces.

Again, the primary purpose of this test was to determine whether a person could rely 100% upon moves generated by a computer chess engine in a Correspondence Chess match versus a highly rated master. In actual play, the master never knows if his opponent is human or silicon or a human assisted by silicon. Therefore this test shows that top rated chess engines such as Nimzo 7.32 and Fritz 6a can indeed hold their own with me when they are teamed with a super fast Pentium III computer and allowed to calculate for nearly a day/move.

Was the match realistic? I think so. For all I knew, my opponent was a sensationally gifted tactician who exhibited poor technique in closed positions. Was the match fair? Well the computer computed for about 17-20 hours per move while I calculated for 1-3 hours late at night after my family went to bed and I was already tired. I usually responded with my move the very next day. In this respect, the computer used far more time than I did, and the time when I analyzed was not optimal for me (this is the big difference between computer-human OTB chess, where the computer and the human use approximately the same amount of time/move). Still, this match is similar to any real-life scenario where a person uses a chess engine to generate moves against an unaided human.

Was anything learned by this match? I don't know about you, but I learned a great deal, including learning more about my own strengths/weaknesses. I also saw first hand how ferocious Nimzo 7.32 is in highly tactical positions… then it's truly scary to play against. But I also saw some positional/technical sloppiness (see Ham-Nimzo 7.32) that I failed to exploit properly. Computer chess experts/fans also got to see their chess engines perform against a 2500+ rated Correspondence Chess Master for the first time. I'm sure they learned more than the casual viewer. Their responses are welcomed.

What does this mean to other strong correspondence chess players? They will likely not suffer that same fate as I did here. The reason being that most ICCF tournaments are constructed so that one must play against 6 to 10 to14 opponents. Thus a person using a computer chess engine to generate all moves against all players would lose by time default when allowing the computer to use 17-20 hours per move. Any time reduction for the chess engines would adversely affect their performance. Consequently, strong masters should not fear this scenario.

However, a strong human who is assisted by a chess engine can indeed have a huge advantage over the human who works alone, as I do now. This is all the more true in positions where tactics and calculation prevail over technique and long-range planning. Given the proliferation of super-fast computers at affordable prices, I think it's only a question of time before more weak humans join the ranks of master-rated correspondence players due to the use of computer generated/assisted moves.

Last Updated: 2000.10.05

Copyright © 1999, 2000 by Steve Ham, all rights reserved
9/4/00 -
Free counters provided by Honesty.com.

Home    CC Challenge Page    All Four Diagrams    Java Replay
Contact Webmaster