The Campbell Report
Correspondence Chess
Computer CC Challenge Match

Steve Ham versus Nimzo 7.32

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

1. d4, Nf6 2.c4, e6 3.Nc3, Bb4 4.Qc2, ...

After 4. Qc2

I've long struggled with what to do about Black's Nimzo-Indian Defense. Originally I avoided it altogether with 3. g3, heading for a Catalan System. I scored many wins with the Catalan, but alas, my success was due to weak opponents rather than the value of my opening scheme. As I gradually met stronger opponents, I found them electing to play rock-solid closed positions against my Catalan. This led to "dull" chess. Instead, I can avoid the Nimzo-Indian via 3. Nf3. That's still too dull if Black plays either 3…b6 or 3…Bb4+. Besides Black can play the sharp Benoni Defense with 3…c5, but I've then eliminated the option of meeting that with the universally approved antidote, the Taimanov Variation, because I've already blocked my f-pawn. Also Black can transpose back to the Queen's Gambit with 3…d5, when having a Knight on f3 is too committal and restrictive for most tastes. I meet the Queen's Gambit with the exchange variation, leaving open the option to place that Knight on e2, depending upon Black's responses.

Later I questioned why I avoided the Nimzo-Indian. Well, Black has good winning chances; he can counter-attack quickly and positions can get wild rapidly, which can be plenty scary for somebody who values his technical skills over his tactical skills. But, White has one major mitigant that really appeals to me... White nearly always keeps his Bishop pair in relatively open positions while Black usually exchanges off his dark-squared Bishop early. This convinced me to meet the Nimzo-Indian on its own turf. But now, which of White's many variations should I play? I think the ubiquitous Rubinstein Variation gives Black the attacking opportunity he relishes. After reviewing my options, I chose to play the Classical Variation (sometimes called the Capablanca Variation) since it quickly causes Black to cede his Bishop pair while minimizing his attacking prospects. I began using this line about 6 years ago. I've had truly amazing results with it... I think I've won all my games but one in this line (the exception was a draw where I was a couple pawns ahead but botched a superior endgame). In truth, I don't think this line objectively gives White an edge. Instead my opponents failed to play the best lines. Still, given my results and the fact that the variation suits my style more than the other options, I'll keep playing it until something better is manifest.

I wonder if I should read anything into the fact that Nimzo 7.32 is playing a defense named after its namesake. Should I assume that it is particularly well programmed with a large/deep opening book for this defense? Let's see what happens.

4. ..., O-O 5.a3, Bxc3+ 6.Qxc3, b6 7.Bg5, Bb7 8.f3, h6 9.Bh4, d5

After 9... d5

Nimzo 7.32 has a well programmed opening book because it's my belief that all other replies (i.e. 9... c5, 9... d6, and 9... Nc6) give White an edge. I am now learning though that objectively correct opening books for chess engines are perhaps less meaningful than opening books subjectively 'tuned' to the strengths/weaknesses of the engine. So far this position has greater potential for a technical player than a tactical one. Let's see how the chess engine performs when it leaves its opening book.

10.e3, Nbd7 11.cxd5, Nxd5 12.Bxd8, Nxc3 13.Bh4, Nd5 14.Bf2, f5 15.Bc4, ...

White had two related mainlines to consider. Most recent Grandmaster games have favored 15 Bb5, c6 16 Bd3, e5 17 Ne2, Rae8 18 Rc1, exd4 19 exd4, N7f6 (19... Nf4) with equality. Instead, I won a nice game with rare text move a few years ago in Ham-Juan Martello (rated 2527), USA vs. South America correspondence match. I know that the ensuing strategic battle can be very complex. Let's see how Nimzo 7.32 handles it.

15. ..., c5 16.Ne2, Rac8 17.Bb5, ...

This is a relatively obscure line I've chosen yet we are still in Nimzo 7.32's opening book. This is most impressive. Several chess computer experts have independently stated that Nimzo 7.32 has the best opening book, so this is confirmation of its depth and quality. A review of the ChessLab two million game database shows only 7 games played with this line, 6 draws and a Black victory. Nothing has been played since 1995. Hmmm... that suggests my line is considered rather lame by theory. Still, I think Black's position is not easy, ... just ask Senior Martello.

After 17. Bb5

17. ..., Bc6

This is a relatively obscure line I've chosen yet we are still in Nimzo 7.32's opening book and it chose the best of the three known lines. This is most impressive. Several chess computer experts have independently stated that Nimzo 7.32 has the best opening book, so this is confirmation of its depth and quality. At this point the ChessLab database shows only 3 results recorded, 2 draws and a Black victory. Another path for Black is based upon 17... Rfd8 18 O-O, a6 19 Ba4, cxd4 20 Nxd4, Nc5 21 Bc2, a5 22 Rfe1, Ba6 23 Rad1, g6 24 Bg3, a4 25 Rd2, Bc4 26 Kc2, Nf6 27 Red1, Rd5 when Black has demonstrated equality. However I favor White after 25 e4. Opening theory used to bless 17... Rf7 18 O-O, a6 19 Bd3, b5 20 Rac1, c4 21 Bc2 but Ham-Martello, corr. 1996-7 later showed a White edge in all lines.

18.Ba6, ...

The text move forces Black's Rook to a new location, and hopefully an inferior one. The game that was the sole White loss in ChessLab's database followed 18 Bxc6?!, Rxc6 19 dxc5, bxc5 20 O-O-O, Ne5 21 Nf4, c4 22 Kc2, Nd3 23 Nxd3, cxd3+ when Black already has an edge. Finally, it is interesting to note that Nimzo 7.32 has no more pre-programmed opening moves. Now let's see how it performs on its own. [The computer's thinking time will be displayed after the move in hours and minutes, now that it is out of its book. -- JFC].

18. ..., Rcd8 (10:37)

Nimzo 7.32 calculated for 10 hours and 37 minutes when selecting this move. I think that three options were available to Black, the text move, 18... Rce8, and 18... Rb8. I think 18... Rb8 was best since it threatens to seal off White's light squared Bishop with 19... b5, thus leading to unclear play. The other moves also threaten that move, but it now has less force.

19.Bc4, ...

White has a number of goals to accomplish. They are as follows: 1) White will soon need to retreat the light squared Bishop along the a6-f1 diagonal before Black traps it with ... b5, 2) White needs to castle, and 3) White should develop the QR to c1. At some point White will try to alter Black's pawn formation with dxc5. If Black recaptures with ... bxc5, then he will have "hanging pawns" which can be targeted by White's Rooks. If Black recaptures instead with ... Nxc5, then White gains a nice Knight outpost with Nd4. Still, White is in no immediate hurry to effect this exchange since Black may hand White a tempo by playing ... cxd4 when White recaptures with Nxd4.

Given the above plans, White had three candidate moves. The text was the only move that avoided having the light squared Bishop being sealed off with ... b5. For example, 19 Rc1, b5 20 a4, bxa4 (Black correctly releases the Bishop because the option 20... Nb4? 21 axb5, Nd3+ 22 Kd2 give White a clear advantage) 21 dxc5, Nb4 22 Nd4, Nxa6 23 Nxc6, Ndxc5 24 Nxd8, Nd3+ 25 Ke2, Nxc1+ 26 Rxc1, Rxd8=.Also 19 O-O, b5 20 a4, bxa4 21 Rfc1, Rb8 is unclear. Notice again how in this last line the Rook moved to b8. This is another reason why I think the immediate 18... Rb8 was best. Notice too that Black has a positional weakness in the backward pawn on the e-file and the fact that his King is on that same a2-g8 diagonal that White's Bishop just moved to. I now think White has a slight edge.

19. ..., b5

Given what I wrote above, I was shocked to find that Nimzo 7.32 evaluates the position as 0.09 pawns in its favor. OK, I know that this low number means the machine thinks the position is equal, but I have to believe that most humans would favor White here for the reasons I already stated. More shocking yet, Nimzo's 2nd candidate move was 19... cxd4 which I think gives White a clear advantage due to the great Knight outpost after 20 Nxd4. Nimzo's third candidate move was the one that I might have played if Black, namely 19... Kh8. It removes the King from the diagonal occupied by White's light squared Bishop. Then 20 O-O, Bb7 21 Rac1 (21 Rfd1 and 21 dxe5 also maintain White's edge), e5 22 dxc5, Nxc5 23 b4 when White maintains his edge.

20.Ba2, c4

Nimzo 7.32 spent 11 hours and 5 minutes to search to a depth of 15/32 plys, assessing the position as being 0.12 pawns in White's favor, a small shift from its previous assessment. The text move is natural, seeking to limit the scope of White's light squared Bishop. Since there are no forced lines, other candidate moves, such as 20... Kh8 (removing the King from the long diagonal) may transpose back into this game line. For example, 21 O-O, c4 22 b3, cxb3 23 Bxb3, a6 24 Rfc1 retains White's edge. Nimzo also considered the purposeless 20... Rfe8?! 21 dxc5!? (simply 21 O-O looks good too), Nxc5 22 O-O, Nd3 23 Nd4, Ba8 24 Nxb5, a6 25 Bxd5 when White has a clear advanatge.

21.O-O, ...

The lack of immediate tactics dictates that White find a plan. My short-term strategy is to initiate a "minority attack" on the Queenside since Black's pawns are so far advanced. This minority attack will ultimately result in Black having an isolated pawn, which White can then target for attack. White's long-term strategy is to open the position to the benefit of White's Bishop pair. However, immediate implementation of these goals with 21 b3, N7b6 (21... cxb3 22Bxb3, N7f6 23 Nc1 [23 O-O, a5 24 Rfc1 gives White an edge], a5 24 Nd3, Rc8 [here's a shocker, Nimzo 7.32 also analyzed this line but favored 24... Rf7?! 25 Ne5, Nxe3?? when it said White was 0.22 pawns better. Humans know that Black is lost here. Franklin Campbell speculated that this is an example of the computer's horizon effect. I guess that must be it.] 22 bxc4, bxc4 23 O-O, Rc8 24 Rfc1, Bb5 25 Rab1, Ba6 26 e4, fxe4 27 fxe4, Nf4 28 Rxf4, Nxf4 is unclear. Also 22 O-O, Rc8 23 Rfc1, Bd7 24 Rc2, cxb3 25 Rxc8, Rxc8 26 Bxb3, Nc4 is only equal. Thus it's clear that White needs to patiently maneuver pieces into position prior to implementation of the minority attack.

21. ..., N5b6?!

Nimzo 7.32 calculated for 7 hours and 45 minutes to a depth of 15/32 ply. Things are getting strange here because after this unexpected retreat of the well-positioned Knight, the computer favors its position by 0.06 pawns. Also strange is the fact that it evaluated its best candidate move as 21... Rfe8, another purposeless move. Perhaps this is symptomatic of computer play in positions where tactics and material grabbing are minimized. The only point I can find to the text move is to meet 22 b3 with 22... Bd5. This is logical, but would a human retreat a well posted Knight for this stratagem? I think most humans would believe that more is being lost than gained. Instead, I believe the best moves for Black were 21... N7b6 or 21... Rc8, both of which allow White to retain his edge. For example, 21... N7b6 (21... e5? 22 b3 +) 22 Rfc1, Rc8 23 Rc2, Bb7 24 Rac1, Kh8 25 b3, Ba6 26 Nc3, Nxc3 27 Rxc3, Nd5 28 R3c2, Nb6 29 bxc4, bxc4 30 Bg3.

22.Bg3, ...

Since Black ceased his attack on e3, this frees my dark-squared Bishop to reactivate with planned advancement to d6. If the e3 pawn is later threatened, my King can now advance to f2 to defend it. Perhaps Nimzo 7.32 just doesn't understand the potential power of the Bishop pair. This power generally increases in direct relation to the opening up of the position.

After 22. Bg3

22. ..., g5?!

I never considered this computer-like anti-positional move either. But Nimzo 7.32 searched to a depth of 15/32 ply and assesses the position as 0.18 pawns in its favor. It is anathema for humans to advance the pawn wall in front of one's King, so I still don't understand why the chess engine chose this move. Perhaps it realized that after my Bishop lands on d6 that Nf4 would follow, and thus it now prevents the later move. Instead as Black I would have considered regrouping via ... Rf7, ... Nf6-d5.

23.Bc7, ...

The Bishop is still headed for d6 ultimately, but I couldn't resist the temptation to possibly gain a tempo by deflecting Black's Rook first. My analysis shows that I get better positions by so doing than from the immediate 23 Bd6. For example, 23... Rf7 24 Nc3 (It is still too soon to effectively open the long diagonal via 24 b3, Bd5 [24... Nd5 25 Kf2, cxb3 26 Bxb3, N7b6 27 Be5 maintains White's edge] when Black gains equality. Hence 24 Nc3 serves to deter Black from ... Bd5), Nf6 25 Be5 (25 Bb4!?) when White retains a small edge.

23. ..., Rc8

Nimzo 7.32 calculated 7 hours and 45 minutes to a depth of 15/32 ply and now assesses the position as improving to 0.22 pawns in its favor! Clearly I believe this is wrong and can not understand Nimzo's heuristics, which drive it to this conclusion. In a way, this is part of the fun in a human vs. computer match; both opponents disagree about the assessment. Let's see who's correct.

The only alternative worth considering is 23... Rde8 24 Bd6 (Nimzo 7.32 considered its main line as 24 Rfc1, a6 25 a4, Nxa4 26 b3, Nab6 27 bxc4, bxc4 28 Bd6, Rf7 when it favors Black by 0.16 pawns. I think this line is just plain silly; humans would have found more logical moves, especially for White. Regardless, 29 Bxc4, Nxc4 30 Rxc4, Bb5 is unclear.), Rf7 25 Nc3, Nf6 26 Rae1 when White manifests at least an edge.

24.Bd6, Rfd8

Nimzo 7.32 calculated for 21 hours to a depth of 17/35 ply and assessed the position as being 0.19 pawns in its favor. Although Nimzo's second choice was 24... Rfe8, I think the only logical choice for an alternative was 24... Rf7 25 Nc3 (25 b3, Bd5=. Nimzo 7.32 considered only 25 Rfc1 at this point. I think that move makes little sense in this position.), Rd8 26 b3, Nf6 27 bxc4, bxc4 and now both 28 Bb4 and 28 Be5 favor White with an edge.

25.Nc3, ...

This move is necessary in preparation for the opening of the long light squared diagonal with b3, as it keeps Black's Bishop off d5. Instead, the immediate 25 b3?, Bd5! 26 Nc3, Rc6 27 Bb4, cxb3 28 Nxd5, Nxd5 29 Bxb3, Nxe3 30 Rfe1, f4 31 a4 when White has no more than compensation for the sacrificed pawn.

25. ..., Nd5

This human reads into this move that the return of the Knight to its former post is an admission that its 21st move was erroneous. Nimzo 7.32 calculated for 8 hours and still assesses this position as 0.15 pawns in its favor, which is interesting since it only began assessing the position in its favor after first retreating this Knight with 21... N5b6. Regardless, I think the move was forced for positional reasons. As an example of how careful Nimzo 7.32 must be here, its 2nd candidate line was 25... a5? 26 e4, f4 27 g3! (27 b3!? maintains White's edge), Nf8 28 Bc5, Rb8 29 gxf4, gxf4 and here Nimzo thought the best continuation was 30 Rad1?, Rxd4?? and assessed the position as 0.03 pawns in its favor. Clearly the lost Rook is an example of Nimzo reaching its "horizon effect". I would have played 30 Kf2! instead with a large advantage. Similar examples of this behavior have been seen in both this game and the other three games.

26.Nxd5, ...

This exchange is part of my aforementioned plan of clearing the light squared diagonal for my Bishop and my ongoing preparation for a minority attack on the Queenside. Still, White had some amusing alternatives. One line involves a pawn gambit with 26 Rac1?, Nxe3 27 Rfe1, f4 28 Kf2, a6! 29 g3, Nb6 (The point of Black's last move. Now, Black's a-pawn isn't pinned the Knight after White plays Bc5) 30 Be5, Nef5 31 gxf4, gxf4 when Black has an edge. White has better prospects with another gambit via 26 e4?!, Nxc3 27 bxc3, fxe4 28 fxe4, Bxe4 29 Rae1, Bf5 30 Bb1, Bxb1 31 Rxb1, a6 32 Rbe1 when White seems to have compensation for the pawn. The text move however maintains White's edge.

26. ..., Bxd5

Nimzo's move is really the only logical continuation. Black has effectively stopped my plans to open the a2-g8 diagonal and to employ a minority attack. This was confirmed after 7 hours and 30 minutes of calculation to a depth of 17/35 ply. However, the darned contraption still favors its position by 0.16 pawns. The only candidate move worth considering besides the text move merely converts Black's Bishop into a big pawn. For example, 26... exd5 27 Bb1 (27 f4 keeps White's edge), Nb6 28 Bb4 (28 Be5, Bd7 29 e4, Na4 30 exf5, Nxb2 31 f4 retains White's edge), Bd7 29 e4, Na4 30 Bc3, Nxc3 31 bxc3, fxe4 32 fxe4, fxe4. Nimzo's analysis stopped here with the assessment that White only enjoys a 0.17 pawn advantage. However, 32 fxe4, dxe4 33 Bxe4 gives White a large advantage.

After 26. ... Bxd5

27.Bb1, ...

Observant readers have noticed that my style of play emphasizes flexibility in the long-range planning process. This position was expected back on my 23rd move, so I prepared having to abandon my aforementioned plans to one of preparing to advance my e-pawn, gaining a tempo on the Black Bishop. The net effect is a strong and mobile central pawn phalanx for White. A similar motif, one that involves freeing my Queen's Rook prior to the Bishop move to b1, is 27 Rae1. However, 27... Nb6 28 Be5 (28 Bb4, c3 29 Bxd5, Nxd5 30 bxc3, Nxc3 31 Bc5 is totally unclear), c3 29 Bxd5, exd5 30 bxc3, Rxc3 31 e4 is unclear. I predict that this technical game will soon show some aggression with White having the better of the action.

27. ..., Nb6

Nimzo 7.32 calculated for 17 hours and 30 minutes to a depth of 17/25 ply. It now believes it has a substantial edge with a 0.32 pawn advantage. I can't disagree more. I think I have the edge! Further examples of our differences of opinion can be found in an examination of Nimzo's rejected candidate moves. For example, Nimzo's 1st runner up candidate move was 27... Bb7 after which it presented a silly line, assessing it as 0.14 pawns in its favor. My plan however was 28 e4, Nb6 29 Be5, fxe4 30 Bxe4, Bxe4 31 fxe4, Rf8 32 Rxf8+, Rf8 33 Rf1, Rf1+ 34 Kf1, h4 (White welcomed the exchanges since Black has an inferior endgame. Black's aggressive 22nd move advanced the g-pawn within striking distance of my King and also placed it on a dark square... the same color as my remaining Bishop. White now plans to play g4! to lock Black's pawn on a dark square.) 35 Kf2, Kf7 36 h3, Ke7 37 Kf3, a4 38 g4, h4 39 d5!, exd5 40 Bd4, Nd7 41 exd5, Kd6 42 Ke4. Now, those of you who find endgames boring may w ant to move on and merely accept my verdict of a large White advantage. Instead, those of you who enjoy endgames may instead want to work out the best lines yourselves before reading my analysis below.

Black has three natural looking continuations. The first is 42... Nf8?? 43 Be5+, Kc5 44 Bf6, Nh7 45 Be7+, Kb6 46 Kf5 and White wins due to 47 Kg6. The second line is 42... Nc5+?? 43 Bxc5+, Kxc5 44 Ke5, b4 45 axb4+, axb4 46 d6 wins.

Instead Black must lose a tempo via 42... b4 43 axb4, axb4 44 Be3, Nf6+ and now how should White continue? If you guessed 45 Kf5, you need to know that Black can draw in King and minor piece versus King and two connected pawn endgames by keeping his King in front of the pawns while his Knight attacks from the rear. Thus 45 Kf5?, Nxd5 46 Bxg5, c3 47 bxc3, b3 48 Bc1, Nxc3 49 Ba3+, Kd7 50 g5, Nb5 (50... Ke8?? 51 g6, Nd1 52 Kf6 wins) 51 Bc1 (51 Bb2, Nd6+ 52 Kg6, Ke6 draws), Nd6+ 52 Kg4, Ke6 53 Kxh4, Kf7 54 Kg4, Nc4 draws. Therefore White must play 45 Kd4, c3 46 bxc3, b3 47 Kd3 when White has a large advantage.

28.Be5, ...

White had two attractive options, the text move and 28 Bb4. I think the text move allows Black more chances to err, although the two lines can transpose. The differences between the two are that the text keeps the Bishop centralized while the alternative shuts down Black's potential source of counterplay, namely ... c3. The main lines after 28 Bb4 follow with 28... Na4 29 Bc3 (29 e4, fxe4 30 fxe4, Bb7 31 Bc3, Nxc3 32 bxc3, Rf8 33 Bc2, Rxf1 34 Kxf1, Rf8 35 Ke2, a5 36 Ke3 when White has an edge), Nxc3 30 bxc3, Bb7 31 e4 (31 Bc2!?), fxe4 32 Bxe4 (32 fxe4 transposes into the 29 e4 line), Bxe4 33 fxe4, again with a White edge.

28. ..., Bb7

Nimzo 7.32 calculated for 18 hours and 15 minutes to a depth of 16/34 ply. It still favors its position by 0.31 pawns, the equivalent of an edge. I thought the best move was 28... Nd7, when I would have repeated the position with 29 Bd6. Since Nimzo believes it has an edge, it's unlikely that it would have repeated with 29... Nb6, but had it done so, I would then have transposed into the note to White's 28th move with 30 Bb4. Interestingly, Nimzo 7.32 erroneously believed the best line for White after 28... Nd7 was 29 Bg3, Bb7 30 h4?! (30 a4 is better), a5 31 Be1, Ra8 32 g4, f4 33 hxg5, hxg5, which it believed favored Black by 0.25 pawns. I also analyzed in advance 28... c3 (28... Na4?? 29 e4, fxe4 30 fxe4, Ba8 31 Rf6, Nxb2 32 Rxh6 wins for White) 29 bxc3, Nd7 30 Bd6, Rxc3 31 e4, fxe4 32 fxe4, Bc4 and now 33 Re1 and 33 Bb4 both maintain White's edge.

29.e4, ...

I now plan to transpose into the line I analyzed in the commentary to Black's 27th move.

29. ..., f4!

Drat! Nimzo 7.32 doesn't comply with my plan and finds the best move. Surprisingly though, it prefers its position even after 29... fxe4? because it doesn't understand the strength of 30 Bxe4, which transposes into my commentary to Black's 27th move. Instead, its analysis continued with 30 fxe4?!, Nd7 31 Bd6, a5 32 e5, Kg7 33 Be7 (I prefer 33 Bc2 when White is better), Rh8 34 Bc2, Bd5 which it evaluates as 0.21 pawns in its favor. This is strange because 35 Rf2 (planning Raf1) is better for White. Therefore I think it's uncanny how the chess engine found the correct technical response based upon flawed calculations. Still, Nimzo 7.32 made the best response, although it does require Black to somewhat overextend its Kingside pawns and lock them onto dark squares. This is a convenient target for my dark-squared Bishop to attack. Let's see if I'm able to exploit this before that Bishop is either driven away or exchanged. Nimzo 7.32 calculated 17 hours to a depth of 17/35 ply and now assesses the position as 0.31 pawns in its favor.

30.Bc2, ...

Although I saw Black's 29th move several moves earlier, I now believe I was too optimistic in my assessment (typical!). So in hindsight, 28 Bb4 was the best move rather than my choice of keeping White's dark-squared Bishop on a more central diagonal. 28 Bb4 keeps the advantage since it stops Black's threatened ... , c3. I've long been a believer in "long analysis is wrong analysis", but in correspondence chess one frequently has to dig deep into positions in order to determine what the best lines are. Well, I've just re-examined my previous long analysis and found a couple equalizing lines for Black now where previously I couldn't find any. In hindsight, they were not difficult to find. Since many of the up-coming lines are non-forced, several options are available to both players and the results of these lines simply don't resolve clearly until several moves later. I didn't explore all lines (who can?) and the ones that I immediately thought the most logical initially favored White.

White's text move frees the Queen Rook and prevents ... , Na4 which threatens White's b-pawn. As such, it is the most testing of the three candidates I explored. The alternatives are: 30 h4, Nd7 31 Bd6, Nb6 32 Bc5 (32 Be5 repeats the position, while 32 Be7?!, Rxd4 33 hxg5, hxg5 34 Bxg5, e5 gives Black an edge since White's dark squared Bishop is locked out of the field of action), Na4 (32... gxh4 33 Bc2 favors White since the following plans are available: Kf2 and Rh1, or Rd1 and Be7, or Kh2) 33 Bxa7, Nxb2 34 hxg5, hxg5 35 Bc2, Kf7 (25... c3 36 Bb3, Kf7 37 Rac1, Nd3 38 Rc2 is unclear) 36 Rab1 (36 Rfb1=, 36 Bb6 is unclear, 36 Kf2 [plans Rh1] is unclear, while 36 Bc5?, e5 gives Black a clear advantage) is equal. Instead, 32 e5?, gxh4 33 Be4 (33 Bc2, c3! 34 bxc3, Rxc3 35 Rac1, Rdc8 clearly favors Black), Rd7 34 Rac1, Kf7 35 Bxb7, Rxb7 36 Kf2, Kg6 37 Rh1, Kg5 gives Black a clear advantage.

Finally, 30 Rd1, a5 (Black removes White's chance to relocate the Bishop to b4. 30... Kf7 may transpose into any of several lines, while the immediate 30... Na4?! 31 Rd2, Kf7 32 g3, fxg3 33 hxg3, Nc5 34 Bc2, Nd7 35 Bd6, a5 36 Rh2 favors White) 31 h4 (31 Bc2, Kf7 32 h4, Nd7 33 Bd6, Nb6 [33... c3 34 bxc3, Rxc3 35 Bd3, Nb6 36 Be5, Nc4 37 Bxc4, Rxc4 38 hxg5, hxg5 39 Rab1 is unclear] 34 Bc5, Nd7 35 hxg5, Nxc5 36 dxc5, hxg5=), Kf7 32 hxg5, hxg5 33 Kf2 (33 Bc2=),Nd7 34 Bd6 (34 Rh1?, Nxe5 35 Rh7+, Kg6 36 Rxb7, Rxd4 37 Rxb5, Rd2+ 38 Ke1, Rcd8 39 Rd5 [39 Rxe5??, Rd1+ is mate in one], R2xd5 40 exd5+, Kf6 41 dxe6, Rb8 clearly favors Black), Nf6 35 Be5 and Black's Knight can repeat with 35... Nd7.

30. ..., a5

As expected, Nimzo 7.32 prevents White from playing Bd6-b4. It evidently realizes, as I mentioned in my previous comment, that White's dark-squared Bishop is more threatening to Black on that diagonal. Nimzo 7.32 calculated for 21 hours and 45 minutes to a depth of 18/37 ply and still favors its position by 0.42 pawns. To date, I'm impressed by Nimzo 7.32's play, but I question its ability to assess positions objectively. But then, how objective am I? I remember reading long ago about a comment made about GM Walter Browne. I don't recall the exact quote but it was something like, "He always thinks he's winning, or at least not so far behind that he can't draw at will." Perhaps Nimzo 7.32 and I fit into that quote somewhat.

31.h4, ...

White has three options here, but per usual, I chose the line that most unbalances the position in order to try to win. Playing for a win is often dangerous when prospects are essentially equal. However, I want to maximize dynamism since Black's dark squared pawns may be overextended and vulnerable to attack from White's dark-squared Bishop.

The other options are:

31 Rfd1, which transposes into the note to 30 Rd1, a5 31 Bc2, found in the commentary to White's 30th move.

31 Bf6, Rf8 32 Be7, Rf7 33 Bc5, Nd7 34 Bd6, Rc6 35 e5, Rc8 36 Bg6, Rg7 37 Bh5, Nb6 38 Bg4, Kf7 39 Rf2 (this allows Rooks to double on the c-file), Nd5 when the position is quite unclear.

31. ..., Nd7

Nimzo 7.32 really disliked my last move, because after calculating 9 hours and 29 minutes to a depth of 17/45 ply, it assessed the position as 0.74 pawns in its favor. The signal it gave was -/+. Clearly I have a much different opinion; the position is dynamically equal. Both sides have worked to unbalance this game so it's easy to slip up here. Unbalancing thus enables the better player the opportunity to punish any mistakes made by the opponent. Still, between Nimzo 7.32 and I, one of us has seriously misjudged this position. Let's see who.

32.Bd6, Kf7

Nimzo 7.32 calculated for 17 hours and 50 minutes to a depth of 16/34 ply and at an average speed of 336 kilonodes/second. It now downgraded its assessment to a 0.59 pawn advantage for itself. While closer to the truth than previously, this is still far too large a disparity from what I believe. I predict that very soon Nimzo will have to admit that it has no advantage.

The only other line that I considered for Black was 32... Nb6 when 33 Be5 repeats the position. Since Nimzo 7.32 assessed that it has an advantage, it thus rejected this line. Still, I would not have allowed a draw by repetition either. Instead, my options to avoid the draw can put Black under a lot of stress. For example, 33 e5 (33 Be7, Rxd4 34 hxg5, Rd7 35 Bf6, hxg5 36 Rfd1, b4 37 axb4, axb4 38 Bd4 [38 Bg5?!, e5 locks White's dark squared Bishop away from the action] with an unclear position.), gxh4 34 Rac1, Nd5 35 Kf2, c3 36 bxc3, Rxc3 37 Be4, Rb3 38 Rh1, b4 39 axb4 (39 Bxd5, Bxd5 40 axb4, axb4 41 Rxh4, Rc3 42 Rxc3, bxc3 43 Rxh6 [43 Ke2, Bc4+ is equal], c2 44 Rh1, Kf7=.),Rxb2 (the natural looking 39... Nxb4?! 40 Rb1, Rxb1 41 Rxb1, Bxe4 42 fxe4, Nc2 43 d5, a4 44 Rc1, Ne3 45 dxe6 gives White an edge) 40 Rc2, Rxc2+ 41 Bxc2, Nxb4 42 Ba4 (42 Bb1, Rc8 43 Bc5, Na6=.), Nd3+ 43 Ke2, Ba6 44 Kd2, Kf7 45 Rxh4, Rg8 is somewhat unclear, but probably equal. I think you can see from these notes that it's Black who is struggling for equality here, not White.

Also, I wouldn't be surprised if some improvements exist for White in the above lines. Therefore Nimzo 7.32 made the correct decision to avoid these lines. But even in the text line I believe that Black has no more than equality. Thus we still need to wait a few moves more in order to see who assessed the position accurately, and to see who has silicon all over their face.

33.hxg5, hxg5

Nimzo 7.32 calculated for 19 hours on its last move, searching to a depth of 18/37 ply at an average speed of 340 kilonodes/second. It now upgraded its assessment to 0.67 pawns in its favor. All I can say about that is Nimzo 7.32 seems to have a big ego.

After 33. ... hxg5

34.Kf2, ...

White has a lot of possibilities here. So to get organized, I wrote down my candidate moves and just began analyzing from there. Although I had an initial bias for 34 Kf2, I figured it was logical to consider putting a Rook on c1 to minimize the danger of ... c3 and to put a Rook on d1 since Black's Rook attacks it when White's Bishop moves away. Therefore my candidate moves were 34 Rac1, 34 Rad1, 34 Rfd1, and the text move. Here's a summary of my notes for the rejected candidates.

34 Rac1, Rh8? 35 Kf2, Nf6 36 Rh1, Rxh1 37 Rxh1, c3 38 b4!, a4 39 e5, Ne8 (39... Nd5?? 30 Rh7 wins) 40 Rh7+, Ng7 41 Be7 gives White a large advantage, possibly even a decisive one. Clearly Black must improve with 34... Nf6 35 Bc5 (White threatens 36 Bb6), Nd7 when the position repeats. If White avoids the draw, then 35 e5, Nd5 36 Be4 (White follows with Kf2) leads to equality.

34 Rad1, Nf6 35 Bc5, Nd7 36 Bd6 repeats the position. If White avoids the draw, then 36 Kf2, Nxc5 37 dxc5, Ke7 38 Rxd8, Rxd8 39 Ke2, Ba6 40 Rh1, c3 41 bxc3, b4+ 42 Ke1 is unclear. Black can also try 34... , Nb6 35 Bc5 (35 e5, Nd5 36 Be4 is unclear), Nd7 36 Bd6 repeats the position. White can avoid the draw with 36 Kf2, Nxc5, which transposes to the 34... Nf6 line.

34 Rfd1, Nf6 35 Bc5 (35 e5?!, Nd5 36 Be4, Rh8 37 Kf2, Ba8 38 Rh1, Ne3 39 Rxh8, Rxh8 40 Bxa8, Rxa8 41 Rh1, Nf5 42 Bc5, Kg6 gives Black an edge), Nd7 36 Bd6 repeats the position. If White avoids the draw with 36 Kf2, Nxc5 37 dxc5, Ke7 38 Rxd8, Rxd8 39 Ke2, Ba6 40 Rh1 then he transposes into the 34 Rad1, Nf6 line.

The conclusions I drew from the above analysis are that Black has at best a draw in those lines. Also, it was irrelevant where I posted a Rook since they never contributed to anything meaningful. Finally, Kf2 was a primary motif in all lines, so why not gain a tempo on the above lines by playing it immediately? Now White threatens to seize the h-file with 35 Rh1. Since we saw that Black fared poorly when contesting the h-file in the 34 Rac1, Rh8 line, White should enjoy that motif all the more since the text move gains a tempo.

34. ..., c3?

So was this what Nimzo 7.32 was counting on to justify it's overstated assessment of it's position? Now I feel pretty good because, as you readers know, I believe my previous maneuvers successfully anticipated this advance. Nimzo 7.32 calculated for 18 hours to a depth of 17/35 ply, at an average speed of 342 kilonodes/second. It now believes it has a 0.50 pawn advantage. Instead, Nimzo should have acknowledged it had no advantage at all and thus settle for 34... Nb6 35 Bc5 (35 e5 and 35 Be5 are both playable lines to avoid a draw. I would probably have chosen 35 e5.), Nd7 36 Bd6 when the position repeats. Instead, 35... Rxc5 36 dxc5, Rd2+ 37 Kg1, Rxc2 38 cxb6, Rxb2 39 Rab1, Rb3 40 Rxb3, cxb3 41 Rb1, a4 42 Kf2 gives White an edge. A similar alternative was 34... Nf6 35 Be5 (35 e5 is unclear but would have been my choice, while 35 Bc5?, Rxc5 36 dxc5, Rd2 gives Black a clear advantage), Nd7 36 Bd6 when the position repeats. Finally, 34... Rh8? 35 Rac1, Nf6 36 Rh1, Rxh1 37 Rxh1, c3 38 b4, a4 39 e5, Ne8 (39... Nd5? 40 Rh7+ wins for White) 40 Rh7+, Ng7 41 Be7 gives White a clear advantage. Therefore 35 Rh1(!) is maybe even stronger, in that it apparently gains a tempo.

As a subsequent event, Uri Blass, in a post entitled "Hard test for chess programs (Steve Ham-Nimzo 7.32)", asked others at the Computer Chess Resource Center whether their chess engines also played 34... c3 (I'll drop the "?" for the rest of this paragraph). Surprisingly they did! Uri wrote that Hiarcs 7.32 chose 34... Nf6 35 Bc5, c3 36 b4, which transposes back into the game line, evaluating the position as favoring Black by 0.15 pawns. Also, Crafty 17.10 played 34... c3, expecting 35 bxc3?, which is really silly. Uri did add though that after he manually played 35 b4, Crafty 17.10 evaluated the position negatively for Black, so perhaps with more time and a deeper search it would have selected a better move. Ed Schroeder, author of Rebel, responded by saying that Rebel 2.0 (beta) initially selected 34... c3, evaluating the line as 0.17 pawns better for White after 35 b4, then switched to 34... Nf6, and then back to 34... c3 35 b4, Rc4, favoring White by 0.13 pawns. Finally it settled upon 34... Nb6 35 e5, Nd5 36 Rh1, Rh8 37 Be4 which it thought favored White by 0.07 pawns. Ernst Walet stated that Schredder initially selected 34... Nb6 35 Bc5, Nd7 drawing by repetition, but then favored White after 35 e5. Then it switched to 34... c3 35 b4, Nb6 36 Bc5, Nd7 37 Rfd1?! (37 Rad1 transposes back to my game line) and favors White by only 0.15 pawns. Finally, it elected 34... Nf6 35 e5, Nd5 36 Rh1, Rh8 (this transposes to the same final position chosen by Rebel above) 37 Bh7 and favored White by 0.07 pawns. Finally programmer Ernst A. Heinz wrote, "DarkThought WCCC 99 never prefers 34... c3 as best. It knows about the weaknesses of advanced pawns that are 'detached' from their siblings. DarkThought WCCC 99 scored the position as 0.15 pawns in favor of White and wavered between Rook moves 34... Rd8, 34... Rg8, and 34... Rh8". I don't know what to conclude, other than these chess engines must share similar heuristics that compel them to blunder in even seriously considering 34... c3? and in failing to evaluate this position properly. The exception here is DarkThought WCCC 99. I think most humans would quickly reject 34... c3? after finding 35 b4. Agreed? Uri Blass summarized, "It seems that Nimzo 7.32 has a larger problem with this position relative to other programs because the other programs were less optimistic about 34... c4. I know of no other program that evaluated the position as favorably for itself either before 34... c3 or after 35 b4. It seems that Nimzo 7.32 gives itself a big bonus for the passed pawn on c3 and does not understand the weakness it created. Other programs that I tried see the position as equal. I guess they have some knowledge that tells them that 34... c3 is weak, but also have knowledge that 34... c3 35 b4 creates a passed pawn, so they give themselves a bonus and so select that move".

35.b4, ...

Although White has now allowed Black's c-pawn to change from half-passed pawn into passed pawn status, he has prevented any exchange allowing Black's Rooks to enter on the c-file. Instead, Black's c-pawn can now become a liability. White can, with adequate preparation, reposition the light-squared bishop along the a2-g8 diagonal, while Black's only Bishop is biting on a granite wall of White pawns. I therefore believe White has a clear advantage presently. The anticipated lines that I analyzed are pretty complex, but it was clear after double-checking my notes that White is much better in these lines. Let's see if I missed something or if the computer did. Clearly the machine and I have a major disagreement about this position, so it should be interesting to see how this sorts out.

35..., Nf6 36.Bc5, ...

Nimzo 7.32 calculated for 24 hours to a depth of 16/34 ply, at an average speed of 346 kilonodes/second. Nimzo now admits it has no advantage at all since it downgraded its assessment to 0.00 pawns; total equality. I believe that in another move or two it will start admitting that White is clearly better. It is confusing though why Nimzo 7.32 only reached 16/34 ply after a full 24 hours of calculations. I wonder whether it is possible that when chess engines downgrade their position assessments, they reduce their pruning of variations. In short, perhaps they look at more options (broader search than a deeper search) when they believe they are losing ground? Israeli computer chess expert, Uri Blass, has a possible explanation. "If a program evaluates 1.e4 as a 0.5 pawn advantage for White, it only has to prove that other moves are weaker. So it is enough for it to analyze 1.d4, d5 and see that White has only a 0.4 pawn advantage. Now suppose that in the last iteration the score of 1.e4 is declining and is now only equality. Seeing that 1.d4, d5 is 0.4 pawns better for White is insufficient to see if 1.d4 is better or worse than 1.e4. Thus it needs to analyze other options for Black. For example, it needs to see if 1.d4, Nf6 is better for White before it recommends 1.d4 over 1.e4. Analyzing 1.d4, d5 and 1.d4, Nf6 takes more time than analyzing only 1. d4, d5. This may be the reason that Nimzo 7.32 calculated fewer plies with the extra time"

White's move is the only one to keep the advantage. Instead, White should not play for cheap shots with 36 e5?, axb4 37 axb4, Nd5?? 38 Rh1, Ne3 (38... Kg8 39 Rh5 wins for White) 39 Ra7 (39 Rh7 probably wins too), Nxc2 40 Rxb7+, Kg6 (40... Kg8 41 R1h7 wins for White) 41 R7h7, Kf5 42 Rh6 when White wins. However, Black need not be so obliging. Instead, 37... Ra8 38 Bd3!? (38 Rxa8, Bxa8=), Nd5 (38... Ba6? 39 Ra5, Nd5 40 Rfa1, Bb7 41 Bxb5 + .) 39 Bxb5, Rxa1 40 Rxa1, c2 41 Rc1, Ne3 42 Bd3, Rh8 43 b5, Rh2 44 Bf1, Rh1 is unclear.

36. ..., Nd7 37. Rad1, ...

As predicted, Nimzo 7.32 is gradually admitting it had silicon all over its face. It now admits White is better by 0.19 pawns. It's interesting to see the chess engine "correct" its evaluation from -/+ to a White advantage all within the space of 5 moves! While this 0.19 pawn edge is minimal, I believe that it soon will agree with me that White has a clear advantage. I favor White because of the following static advantages:

After 37. Rad1

  1. Black's Bishop is biting down on a granite chain of White pawns.
  2. Black's Knight has minimal scope, so it will likely exchange itself for my dark-squared Bishop. Then I'll cede the advantage of my Bishop pair in exchange for a passed c-pawn and an open Queen file.
  3. Black's c-pawn is in danger of being lost.
  4. White's light squared Bishop has options of returning to the a2-g8 diagonal, targeting Black's backward e-pawn.
  5. White's Rooks have better prospects than Black's for penetrating the enemy camp.

Nimzo 7.32 calculated for 19 hours and 45 minutes to a depth of 15/32 ply, at an average speed of 346 kilonodes/second. As noted before, this is a relatively shallow search depth. Franklin Campbell reported that Nimzo 7.32 had already reached that depth after 8 hours. While it was unable to search deeper in the subsequent 12 hours, its evaluation changed from 0.00 to 0.19 pawns for White, so it was certainly calculating. This is confirmation for me that I'm not the only one who finds this position complex and tricky.

37. ... , e5!?

Nimzo 7.32 struggles in a difficult position. Although I'm not certain what Black's best defense is, I initially thought Black's best was 37…Nxc5 38 dxc5, axb4 39 axb4, Kf6 40 Ke2, Ke5 41 Rd3 when White has a large advantage. While Black is stuck with the terrible weakness in the pawn on e6, he at least gets his King favorably advanced to e5. Apparently Nimzo considered the potential target on e6 was untenable and so moved its pawn off that attack square. It came to this conclusion after calculating for 18 hours and 10 minutes to a depth of 16/34 ply, at an average speed of 347 kilonodes/second. It now favors White by 0.48 pawns. While it understated White's advantage, the important point is that it concurs that it is inferior and I suspect there will a be a trend over the next few moves where its assessment of White's advantage will continue to grow.

38. Rh1(!), ...

White's primary problem, as I see it, is how to sort through the numerous attractive continuations now. Only after analysis did it become clear that some lines are not as attractive as others. The candidate moves that I examined were, in the order that they popped into my mind: 38 Bd6, 38 Ke2, 38 Bb3+, and finally 38 Rh1. Here is a summary.

I don't know why I examined 38 Bd6?! first. At first look, the move seems illogical. Although White desires having Black exchange the Bishop with his Knight in order to obtain a strong passed pawn, Black's last move means it will first play ... exd4 to prevent this. So 38 Bd6?!, Ke6 39 dxe5, Nxe5 (39... a4?? [prevents 40 Bb3+] 40 Rh1, Rh8 41 Rxh8, Rxh8 42 Bd3, Nxe5 43 Bxe5, Kxe5 44 Bxb5, Rc8 45 Ke2 when White's advantage is very large and possibly decisive. Black's a and c pawns are endangered.) 40 Rh1, Nf7 41 Bb3+, Kf6 42 Bc5 gives White an edge. Given my belief that White should have a large advantage here, it's pointless to now examine 38... exd4.

Knowing that White should have more than just the edge found above, I selected 38 Ke2 (White moves closer to Black's targeted c-pawn), exd4 (38... Nxc5? 39 dxc5, axb4 40 axb4, Rd4 41 Rxd4, exd4 42 Rh1, Kg7 43 e5, Bd5 44 Kd3, Rd8 45 Kxd4, Bxf3+ 46 Kxc3, Bxg2 48 Rh5 wins.) 39 Bxd4, axb4 40 axb4, Nf8 41 Rh1, Ne6 42 Rh7+, Kg8 43 Rh8+, Kf7 44 Rxd8 gives White a large advantage. This line was my primary choice until I later discovered the related motif in 38 Rh1.

38. Bb3? is very tricky and I'm not at all certain I analyzed this line correctly. Black now has 38... Kg6 (38... Kf6?? 39 Ke2, Nxc5 40 dxc5, axb4 41 axb4, Rd4 42 Rh1, Kg6 43 Rxd4, exd4 44 Bd5!, Bxd5 45 exd5, Kf6 46 Kd3, Ke5 47 Re1+, Kxd5 48 Re4, Ra8 49 Rxd4+, Ke5 50 Kxc3 when White has a very large and possibly decisive advantage.) 39 Be6, c2 40 Rc1, exd4 41 Rxc2, axb4 42 axb4, d3 43 Rd2, Nxc5 44 Bxc8, Nxe4+ 45 fxe4, Rxc8 46 Rxd3, Bxe4. Now matters are pretty unclear to me. Therefore I didn't spend much time trying to make sense of this line since 38 Ke2 and subsequently 38 Rh1 looked so promising.

After the game, IM Tim Harding asked why I didn't consider 38 Rd3, Nc5 39 dxc5, Rxd3 40 Bxd3, Rd8 41 Bb5, axb4 42 axb4, Rd4 as a possibility. He then offered 43 Rb1?, Rd2+ 44 ke1, Rxg2 45 Bc4+, Ke7 45 b5, g4. He wrote, "The White Queenside pawns look like they ought to be strong but the Kingside gets undermined." Indeed, Black has a clear advantage here. Instead, White's best seems to be 43 Rc1, Rxb4 44 Bd3 (44 Be2 [44 Bd7??, Rc4 gives Black a clear advantage.], Rb2 45 Kf1, Ke7 46 Rxc3, Rb1+ 47 Kf2, Rb2 48 Ke1, Kd7 49 Rc1, Kc6 is equal.), Rb2+ 45 Kg1, Bc6 46 Rxc3 (White plans 47 Bc4+, Ke7 48 Bd50, Rd2 47 Kh2 (White plans Kh3-g4 or the pawn push g4), Ke7 48 Kh3, g4+ 49 Kxg4 (49 fxg4, Kf6 is equal.), Bd7+ 50 Kh4, Rxg2 51 c6, Rh2+ 52 Kg5, Rg2+ is equal.

My selected move 38 Rh1 is very similar to 38 Ke2. Both moves were prevalent in my analyses and sometimes transpose into each other, but 38 Rh1 seems to offer more attractive transpositions. Still, there is some risk here: Ke2 is always a strong move, while Rh1 can amount to a waste of time in some lines if Black can deny entry into its position. Thus, I'm left with trusting my analysis (not always a reliable decision!) to "ensure" that Rh1 works best here. In short, 38 Ke2 is the best "practical" choice in that this line is safe and very strong regardless of what Black does, while 38 Rh1 is apparently even stronger but poses risk in that one simply can't analyze everything. Since my philosophy is to always play "the best" line, regardless of the risks, then the choice of 38 Rh1 is clear. Let's see if I chose correctly.

38. ... , Nxc5 39. dxc5, ...

Nimzo 7.32 calculated for 18 hours and 45 minutes to a depth of 15/32 ply, with an average speed of 355 kilonodes/second. Surprisingly, it only favors White by 0.04 pawns, retreating from the 0.48 pawn advantage of its previous assessment. Thus it doesn't think much of my last move. I find this amazing. Sure, from a simplistic standpoint, White's lost the Bishop pair, but in exchange, the supported passed c-pawn is a greater advantage. Further, White's Rook will invade along the h-file and Black's c-pawn is will fall eventually. I wonder how much longer we have to wait before the chess engine's assessment shows White have a large advantage. Just compare this position to the related one after 38 Ke2, and you can see why I'm confident of White's potential here.

Black really had no acceptable alternatives at this point. After 38... exd4?, 39 Bxd4, I had analyzed many lines. Here is a brief summation of the main line. 39... axb4 40 axb4, Nf8 41 Bc5!, Kg7 42 Bxf8!, Kxf8 43 Rh8+, Ke7 44 Rh7+, Kf6 45 Rxb7, Rxd1 46 Bxd1, c2 47 Bxc2, Rxc2 48 Kg1 wins for White.

39. ... , axb4 40. axb4, ...

Aha! Justice at last! Nimzo 7.32 calculated for 17 hours and 30 minutes to a depth of 17/35 ply, at an average speed of 353 kilonodes/second. It finally admits it is in serious trouble, favoring White with a 1.03 pawn advantage. I wouldn't be surprised if it favors White even more in the future. For awhile, I questioned my sanity because Nimzo 7.32 refused to admit what seemed so obvious to me for many moves, that it's position is quite bad. Still, matters are quite tricky. For example, White had a seemingly attractive alternative in 40 Rh7?!, Kg6 41 Rxb7, Rxd1 42 Rb6+, Kf7 43 Bxd1, Rxc5 (43... bxa3?? 44 Rxb5, Ra8 45 Rb1, Ke7 46 Ra1, Kd7 47 Ke2, Kc6 48 Kd3 wins for White) 44 axb4, c2 45 Bxc2, Rxc2+ 46 Ka1, but White has only an edge here. It is not the large advantage I believe White gets after the text move.

40. ... , Rxd1

Perhaps this is Black's "only move. Nimzo 7.32 calculated for 18 hours to a depth of 18/37 ply, matching the record search depth of its 30th move. This was accomplished at an average speed of 355 kilonodes/second. It now favors White by 1.15 pawns, so as predicted, favor for the White position is increasing. The alternatives to the text move all lose rather quickly. For example, 40... Kg6? (40... Rd4? 41 Rxd4, exd4 42 Rh7+, Ke6 43 Rxb7 wins) 41 Ke2, g4 42 Rd3, Ra8 43 Bb3, Rd4 44 Rxd4, exd4 45 Kd3, gxf3 46 gxf3, Rd8 47 Rh4 wins.

41. Rxd1, ...

White has only one good move too, since 41 Bxd1?, Rd8 42 Bb3+, Kg6 abandons the d-file to Black.

41. ... , Ra8 42. Bb3+, ...

Nimzo 7.32 calculated for 17 hours and 15 minutes to a depth of 18/37 ply, again matching its previous record search depth. It calculated at an average speed of 353 kilonodes/second and now favors White by 1.12 pawns. Its move was the only one worth considering, but does set some tricky traps for an unsuspecting White to fall into.

For example, the biggest trap is 42 Rd7??, Ke6 43 Rxb7, Ra2 44 Ke2, Rxc2 45 Kd3, Rxg2 46 Kxc3, g4! (Did you see this coming?) 47 fxg4, f3 48 Kd3, Rxg4 49 Rxb5 and now 49…Rf4 wins for Black.

Other traps can be found after 42 Ke2, Ke7 (42…Ra3?? 43 Rd7+ wins immediately for White). Now 43 Bb3+ is best and will likely transpose into the game line. However, 43 Rd6?, Bc8 44 Kd3 (44 Rg6?, Be6 45 Kd3, Ra3 gives Black a clear advantage since he threatens 46…Bc4+ checkmate.), Ra3 45 Rd5, Be6 46 Rxe5, Kf6 47 Kd4, Ra2 48 Rxe6+, Ke6 49 Kxc3, Ke5 50 Kb3, Ra1 51 Bd3, Kd4 52 Bxb5, Rb1+ 53 Ka4, Ra1+ draws by perpetual check. Also, 43 Kd3?, Ra3 44 Rb1, Kd7 45 Rb3, Ra1 46 Kxc3, Rg1 47 Bd3, Rxg2 48 Bxb5+, Kc7 49 Be8, g4 is unclear.

Tricky chap, this Nimzo 7.32. I can see Nimzo 7.32 plans to defend tenaciously in order to avoid defeat. Also, this is the part of the chess game where I have a noticeable weakness; converting a large advantage into a win. Please forgive the lack of modesty, but after hard work I obtain large advantages in most of my games. However, I then feel the pressure to score the point and so react by relaxing, thinking the real battle is over anyway. This often leads to sloppy chess and lets the potential victim get away (see Ham-Fritz 6a for an example as early as the 14th move, when I thought I had that monster all but slain). Given that computer chess programs are famous for finding great defenses, it is imperative that I make every effort to calculate accurately and avoid getting lazy. 

42.  ... , Ke7 43. Ke2, ...

Nimzo 7.32 played exactly as forecast above. The chess engine now correctly favors White even more than last time, but only by 1/100th of a pawn! It says White is better by 1.13 pawns after deliberating for 16 hours at an average speed of 357 kilonodes/second and searching to a depth of 20/40 ply, clearly a new search depth record in this game.

White's move prepares to capture Black's over-extended c-pawn. The other route with 43 Rd3?, c2 44 Bxc2, Ra2 45 Rc3 only gives White a small edge. Now I think that with perfect play by both sides, White will win. Still, I have to prove the win exists to my all-seeing opponent while battling in a complex position. You may recall that previously I played in a style that created unbalanced positions to try to exploit any errors by my opponent. That style met with great success in this game, but now I'd like to reverse that ploy to simplify matters to minimize counterplay for Black's Rook along the a-file. Let's see what happens.

43. …, g4

After 6 hours of calculating to a depth of 18/37 ply, Nimzo 7.32 had already selected this move and assessed White's advantage as being 1.46 pawns. However, the computer was allowed to calculate for 19 hours and 15 minutes to a depth of 19/38 ply, at an average speed of 363 kilonodes/second. The final evaluation retreated slightly to a White advantage of 1.33 pawns.

Instead, I expected 43…Ra3?? 33 Rb1, c2 45 Bxc2, g4 46 Rh1, gxf3+ 47 gxf3, Kd7 48 Rh7+, Kc6 49 Bd3, Ra2+ 50 Kd1, Bc8 51 Rh6+, Kc7 52 Bxb5, Rb2 53 Rh7+, Kb8 (53…Kd8 54 c6 wins because White threatens 55 c7+ checkmate) 54 Re7, Rxb4 55 Bd3, which wins for White. I thought that the only reason that a human might have for selecting this text move is that it involves a "cheapo" that White could fall for. I therefore wonder whether chess engines are programmed to play for cheapo's as the best practical defense in lost positions? As a subsequent note, Mark Ryan asked my question at the Computer Chess Resource Center bulletin board. In response, Mike Scheidl wrote that Hossa does this in "lucky punch mode" while Djordje Vidanovic wrote, "Roland Pfister's 'Patzer' does something similar. When losing, it's ready to sac material to get a perpetual, etc. perhaps Wchess too". Wow, these machines are truly tricky. If they can't outplay you, they trick you!

44. Kd3, …

White will execute the helpless c-pawn. Then, if Rooks can be exchanged, the King will begin a flanking maneuver with an end-run around the Kingside pawn wall in order to win the game. However, if White captures the c-pawn with his Rook, he then cedes control of the d-file and thus falls for Black's aforementioned cheapo, giving Black a defense. For example, 44 Rc1, Rd8 45 Rxc3, Rd4 46 Bd5, Bxd5 47 exd5, gxf3+ 48 gxf3, Rxb4 (48…Rxd5?? 49 Rd3, Rxd3 50 Kxd3, Kd7 51 Ke4, Ke6 52 c6, Kd6 53 Kc7, Kxc7 54 Kxe5 wins) 49 d6+, Kd8 50 c6, Rc4 (50…Rd4?? 51 c7+, Kc8 52 Rc6, b4 53 Rb6, Kd7 54 Rb8 wins) 51 c7+ (here's Black's trick…51 Rxc4??, bxc4 wins for Black!), Kd7 52 Rxc4, bxc4 53 Kd2, e4 54 fxe4, f3 55 e5, c3+ 56 kxc3, f2 57 e6+, Kxd6 58 c8/Q, f1/Q 59 Qd7+, Kc5 (59…Ke5?? 60 e7 wins) 60 e7, Qc1+ draws.

44. …, Rd8+ 45. Kxc3, …

For those of you who enjoy some humor with your chess, please re-read my commentary to the last two moves. Now, for the punch line: I fell for Black's trap while thinking I had avoided it all along! Oops! To err is human and I just made a very human (read "extremely stupid") mistake and am horribly embarrassed. The highly public viewing of this match only magnifies my embarrassment.

For whatever reason (probably induced by late night fatigue), I expected 44…gxf3?? 45 gxf3, Rd8+ 46 Kxc3, Rxd1 47 Bxd1, Bc6 48 Bb3, Be8 49 Kd3, Bh5 50 Ke2, Kd7 51 Kf2, Ke7 52 Kg2, Be8 (52…Kd7 53 Bd1, Bf7 54 Kh3 wins) 53 Kh3, Bd7+ 54 Kh4, Kf6 55 Bd5, Bc8 56 Ba8, Be6 57 Bb7, Ke7 58 Kg5, Bh3 59 Bc6, Bg2 60 Kg4, Bf1 61 Kf5, Be2 62 Kxe5, Bxf3 63 Bxb5 wins. Instead, I overlooked a Black defense that couldn't be more obvious, namely 45…,Rxd1 46 Bxd1, g3, forcing a draw. White's King now can't penetrate along the h-file. White had no satisfactory alternatives because 45 Kc2, Rxd1 46 Kxd1, g3 47 Bc2 (47 Bd5??, Bxd5! 48 exd5, e4 wins for Black. Also 47 Kc2??, Bxe4 wins) draws. Strangely, on Nimzo 7.32's last move, which it calculated to a depth of 20/40 ply at an average speed of 367 kilonodes/second, it too didn't see the draw. That's because it favored White by 1.26 pawns. My conclusion is that Nimzo 7.32 is a very strong chess program, but its evaluation function should be ignored. I've found it to be inaccurate throughout this game. Anyway, I'll play on until Nimzo demonstrates that it sees 46…g3 and then offer a draw. After all, 46…gxf3?? transposes back into my notes above and Black's loss may be beyond the chess program's search horizon.

I'm now a convert to the point of view that at the higher levels of correspondence chess, humans may consult computers to "blunder check" their plans. Originally this notion was anathema to the way I thought correspondence chess should be played, as a match strictly between two humans. However, the use of computer chess programs is reportedly rampant in master-level correspondence chess and prohibition is unenforceable. Given the tremendous calculating skills demonstrated by Nimzo 7.32 and Fritz 6a, it is now too difficult for this 2500+ rated human to consistently calculate accurately versus such powerful hardware/software to the extent that I can defeat them. Yes, I earned winning advantages in both this game and Ham-Fritz 6a, but failed to capitalize upon my advantages.

I now think that in future serious ICCF games I will need to acquire these hardware/software tools as a synergy with my limited human skills in order to compete effectively. Drat! This will change the way correspondence chess is played. There will be fewer human tactical oversights to exploit when both parties use computers. One may then have to hope that one's opponent relies too heavily upon computer tactics/evaluations, thus failing to utilize their superior human long-range planning ability. That's not the type of correspondence chess that I'm familiar with and love to play.

45. …, Rxd1 46. Bxd1, g3 47. Be2 Draw

I hereby congratulate Nimzo 7.32 for playing well enough to be able to draw with me. Please re-read my commentary to White's 42nd move, where with remarkable precognition, I wrote about my laziness in going for the kill by giving away too many ½ points. Clearly, my tendency to psychologically relax when I've earned a large advantage is an area where I need to work on improvement.

After 47. Be2 ½-½

This draw is agonizing though. The computer program is too ignorant to recognize the draw because it still favors White by 1.21 pawns after searching to a depth of 24/46 ply! How would you feel about converting a big advantage into a draw with an opponent too stupid to know that it's a draw? Probably not half as bad as I feel in allowing the draw, given the public nature of this match game. Still, I don't mean to pick on the extremely powerful Nimzo 7.32. Misjudgment of this position is apparently common among chess programs. Postings at The Computer Chess resource Center reported the following: after 15 ply Fritz 5.32 favored White by 0.63 pawns, Goliath 2.7 favored White by 1.63 pawns, and Crafty 17.10 liked White by 1.87 pawns. After 16 ply, Chess System Tal 2 favored White by 1.67 pawns. This supports my belief that while top-level chess programs are extremely strong when allowed to calculate for long periods of time on very fast computers, their ability to accurately evaluate positions is quite suspect and often totally wrong. One viewer e-mailed me to say that his chess program sometimes favored itself when given both sides of the same position!

In summation, Nimzo 7.32 played extremely well…much better than I expected. My perception is that it is a powerful tactician, especially in open positions (see Nimzo 7.32-Ham). However, in closed positions where no clear plan is present, it tries to resolve maters tactically, which is intrinsically incorrect in correspondence chess. However, it was able to escape with a draw here because I too have certain weaknesses, chief of which is laziness when turning a big advantage into a win. My perception so far is that Nimzo 7.32 is a superior tactician than Fritz 6a, but the latter performs better in closed positions.

Last Updated: 2000.08.07

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