|
The Campbell
Report |
|
| |
Steve Ham versus Fritz 6White: Steve Ham
Black: Fritz 6 Computer CC Challenge Match 1. d4, Nf6 2. c4, g6 3. Nc3, Bg7 4. e4, d6 5. f3, O-O 6. Be3, e5 7. Nge2, … My ideas about opening theory often differ from what is generally accepted by "the books" or current fashion. This is due in measure to the fact that I always try to play what I objectively believe to be best. Then, when multiple candidate lines still remain, I select the line that appeals most to me subjectively. NCO and most books on opening theory prefer 7 d5 for White, leading to lines where they believe White has an edge. This notion is thematic too because one generally wants to close the center in order to subsequently castle and then attack on the opposite wing. However, my assessment of the 7 d5 lines is that Black equalizes. Instead, I prefer to save the tempo and keep the center fluid while preparing to storm Black's King side. Interestingly opening theory says Black equalizes here, but I think White has an edge. Let's see who is correct. 7… , c6 8. Qd2, Nbd7 9. O-O-O, a6 10. h4, … Once again I diverge from theoretical fashion. The overwhelming choice of most masters here is 10 Kb1, b5 11 Nc1 preparing to play either on the Queenside with Nb3 or even Nd3. Sometimes White opens the center with dxe5. Instead my rare text move is an example of "red meat" chess. When young, I was influenced by the autobiography of Yefim Geller, who wrote that he'd played literally thousands of games where both sides castled on opposite wings. He noticed that generally the first player to mount an attack on his opponents King thus created an advantage. This is a maxim that I've mentally absorbed into my chess style. Perhaps it's not a commonly accepted maxim because 10 h4 is rarely played, although GM Beliavsky played it several times a few years ago. Once again, let's see if there is any value to my approach, which is similar to the Yugoslav Attack versus the Dragon Sicilian. I have a "feel" for these types of positions since I play the Dragon Sicilian as Black. Here however, I prefer White's chances. An unintended effect of 10 h4 is that Fritz has now left its opening book. 10… b5 The race to attack each other's King commences. Fritz spent 8 hours and 43 minutes in reaching what I believe to be the correct conclusion: it is better to try to attack White's King than to try to slow White's attack with 10… h5. I must admit to some human worry on my part here. Although I believe White is superior here, I worry about engaging in a mutual King hunt with a mechanical opponent that will not miss a tactical shot within its search depth (which must be very deep indeed). Fritz 6 will never tire, never has emotional worries, and always coldly plays the best move it can find. In this sense, I'm fighting on its "turf". Additionally, I'll have to gambit away some material to make my attack work. Let's see what happens. 11. h5, Nxh5?? Black had five reasonable choices here. After 11 hours and 35 minutes Fritz 6 chose a move that perhaps most appeals to chess engines … gobbling material. Like the previously mentioned relationship to the Yugoslav Attack versus the Dragon Sicilian, White gladly parts with a pawn in order to pry open the h-file in order to attack Black's King. I think White now has a very large advantage. Don Maddox said that Fritz 6's second favorite line was 11… bxc4 after an hour of calculation, but later 11… Qa5 and 11… b4 were a close second and third choice. I believe 11… bxc4 12 Bh6 gives White a big advantage Also 11… b4 12 Nb1, Qa5 13 Bh6, Qxa2 14 Ng3, exd4 15 Bxg7 leads to a clear White advantage. The moves recommended by opening theory are 11… Nb6
12. g4, Nhf6 13. Bh6, b4 Fritz 6 spent 10 hours and 1 minute on this selection. This move is a "theoretical novelty" since I've never seen commentary on this position and there are no games in the 2 million game ChessLab database after this move. Known theory stems from a game that continued with 13… Bxh6?? 14 Qxh6, bxc4 15 g5, Nh5 16 Ng3, Re8 17 Rxh5 when White had a winning position. Thus it's a simple conclusion that Fritz 6's theoretical novelty is an improvement upon human opening theory. 14. Bxg7??, … I confess to you readers that I played this move with great trepidation. Fritz 6 plays tactics perfectly within its search depth while I do not. Were my opponent human and not a computer chess engine, I could sleep at night with the conviction that my attack will land a harder punch to the chin of my opponent's King than he could land on mine. Now in hindsight, I see that my tempo gaining text move actually throws away a near winning position! I never considered Black's 15th move. How did I miss this maneuver when I played it myself in my Dragon game versus Nimzo 7.32? I guess I was just lazy; surely a human trait and not one that my opponent could ever be guilty of. Instead, White needs to actually "lose" a tempo with 14 Nb1!, thus compelling Black to move. Then Black has three reasonable looking responses:
The surprising thing about the tempo losing 14 Nb1! is that it eventually gains a tempo because Black isn't allowed to play …Kh8, …Rg8, and …Rg7. Surely you, dear readers, can understand the all too human desire to rush our pieces into the attack as I have done here, since that's usually the way one conducts an attack; get there first with the most. This shows that the mechanical brain has superiority in the respect that it considers all logical moves. However, that same mechanical brain also got itself into that nearly lost position by grabbing a pawn on its 12th move, something few strong humans would ever do. As a subsequent event, I asked Israeli computer chess expert Uri Blass whether chess engines could find 14 Nb1! The answer is that neither Hiarcs 7.32 nor Chess Master 6000 (selectivity=10) could, although both are considered top-notch programs. Next Uri gave them the Black position after 14 Nb1!, but they saw no problems for Black. After more than 10 minutes of calculation they both selected 14…Bxh6? as best. Chess Master 6000 saw no problems for Black until after 16 Ng3 (following my analysis above). Then after 28 minutes it chose 16…b3 17 axb3, Nf8 18 g5, N6d7 19 Nf5, gxf5 20 exf5, Nc5 and held to that choice after several hours of calculation. Then when it calculated that terminal position, it saw a checkmate in 16 moves in less than a minute! Meanwhile, Hiarcs 7.32 couldn't see problems for Black until after 18 Nf5, when it then saw "it is in big trouble". Well, I guess these mechanical marvels aren't such brilliant calculators after all. Maybe their calculations are OK, but their assessments of positions are truly bad, even tactical positions such as this. Now I don't feel so stupid about missing 14 Nb1!, when highly rated chess engines also fail to see that it is best too. Paulo Soares of Brazil provided a postscript to the above commentary. Paulo likes to analyze in partnership with his Fritz 6a chess program. Together, they found an improvement for Black with 14 Nb1, Qe7! 15 Bxg7, Kxg7 16 Qh6+, Kh8 17 Ng3, Rg8 18 g5, Nh5 19 Nxh5, gxh5 20 dxe5, Nxe5 when he assessed the position as unclear. I continued with 21 Qxh5, Rg7 22 Nd2 when I think White has an edge. Paulo/Fritz 6a then found the best line, namely 14 Na4! I had quickly rejected this in the game because I thought the Knight would remain out of play forever. However, now 14…Qe7 15 Bxg7, Kxg7 16 Qh6+, Kh8 17 Ng3, Rg8 18 g5, Nh5 19 Nxh5, gxh5 20 dxe5, Nxe5 when Paolo said White wins. Initially I didn't see any clear wins until I saw that 21 Rxh5! (21 Qxh5, Rg7 22 Nb6, Rb8 22 Nxc8, Rxc8 24 f4, Ng6 25 f5, Nf4 26 Qh4, Qxg5 27 Qxg5, Rxg5 28 Rxd6 gives White a clear advantage.), Rg7 22 Nb6, Rb8 23 Nxc8, Rxc8 24 f4, Ng4 (24…Nf3? 25 f5 wins) 25 Qxd6, Qxe4 26 Bd3, Qe3+ 27 Kb1, Re8 28 Bf5 wins due to unstoppable mate threats. 14… , Kxg7 Fritz 6 had a 31 hour and 23 minute calculation period, where I believe it found the only playable move. The alternative is 14… bxc3?? 15 Qh6, cxb2+ 16 Kb1, Re8 17 g5, Rb8 18 gxf6, Nxf6 19 dxe5, Rxe5 20 Bxf6, Qxf6 21 Qxh7+, Kf8 22 Nd4, Rc5 23 Qh8+ with an obvious win for White. 15. Qh6+, Kh8! Fritz 6 now assessed the position as 0.97 pawns in my favor after 10 hours and 40 minutes of calculation time. What does that mean in human terms? After all, I'm materially a pawn down. I think the computer acknowledges that its position is under attack from a superior force, but has not seen all the way to a resolution. Had it been able to calculate a couple of ply deeper, I think it would have seen that its disadvantage is minimal. Regardless, Fritz 6 found the correct move, preparing … Rg8-g7. The line I expected was 15… Kg8?? 16 Nb1, Re8 17 Ng3, Nf8 18 g5, N6d7 19 Nf5!, gxf5 20 exf5, f6 21 Rg1 when White wins. 16. Nb1, Rg8 Fritz 6 spent 6 hours and 43 minutes to search 16/16 ply and confirm its plan from the previous move. Its plan is the same as I am using in my Sicilian Dragon game versus Nimzo 7.32, namely … Rg7. Black now threatens 17… g5 to prevent White from advancing with 18 g5. 17. g5, … White's move was forced since the only logical alternative
17… , Nh5 18. Ng3, Qf8 Fritz 6 calculated for 21 hours to reach a depth of 18/18 ply and assessed the position as 0.22 pawns in my favor. The only logical alternative was 18… exd4? 19 Rxd4, Nf8 20 Nxh5, gxh5 21 Rxd6, Qxg5+ when Fritz continued with 22 Nd2?, Bd7 23 f4 and assessed the position as having a large White advantage of 0.81 pawns. This seems erroneous in light of 23… Qxh6 24 Rxh6, Rd8 25 R6xh5, Rg4 26 f5 when it looks unclear to me. Instead I would have played 22 Qxg5, Rxg5 23 Rxc6, Kg7 24 Kd2!, Ne6 25 Ke3 planning Nd2 when White has a clear advantage. 19. Qxf8, … Due to the error of my 14th move, White has little prospects for much advantage, although some Black pawn weaknesses may be exploited after the text line is resolved. Instead, the alternative was 19 Nxh5, Qxh6 20 gxh6, gxh5 21 dxe5 (21 Rxh5=),dxe5 (21… Nxe5?! 22 Rxd6, Nxf3 23 Rxc6 gives White an edge) 22 Rxh5 leads to equality.19… , Nxf8 Fritz 6 calculated for 6 hours and 30 minutes to a depth of 18/40 ply to confirm that the text move is the only acceptable defense. Instead, 19… Rxf8? 20 Nxh5, gxh5 21 Rxh5, Rg8 22 Rh6 (22 dxe5, dxe5 23 Rd6 and 22 Be2 planning 23 Rdh1 only confer a small advantage to White), Rxg5 (22… Rg6 23 dxe5, dxe5 24 Rxg6, hxg6 25 Rd6 gives White a clear advantage) 23 Rxd6 (Fritz 6 considered only 23 dxe5?, Nxe5 24 f4 claiming a clear White advantage of 0.91 pawns. But 24… Rg6 25 Rxg6, Nxg6 26 Rxd6, Nxf4 26 Rxc6, Kg7 28 Nd2, h5 favors Black. I've read that chess engines often fail to consider the potential of advancing passed pawns. This may be an example of that.), when White has a large advantage. 20. Nxh5, gxh5 21.dxe5, … White's alternative is 21 Rxh5, Ne6 22 dxe5, dxe5 23 g6, Rxg6 24 Rxe5 when While has a very small edge. The text line should transpose but offers Black a chance to err. Interestingly, Fritz 6 favored its position by 0.56 pawns after its 20th move. Since this is a radical change from its previous assessment, I'm curious to discover how Fritz 6 justifies this pronouncement. 21. … dxe5 Fritz 6 finds the correct move order, confirmed after 10 hours of calculating to a depth of 16/39 ply. Instead 21… Rxg5? 22 exd6, f5 23 Nd2, Kg7 24 e5 (24 f4 was preferred by Fritz 6, but I think this is less precise as now White's central pawn mass gets broken up after 24… Rg4 25 Rxh5, fxe4 26 Nxe4, although White has an edge) gives White a clear advantage.22. Rxh5, Ne6 Fritz 6 seems almost human in that its choice of moves actually sets a "trap" that I nearly stepped into. In so doing, I think it is demonstrating technical/positional skills that chess engines are usually thought to be deficient in. Fritz 6 calculated for 17 hours and 30 minutes to a depth of 17/39 play and assesses the position as 0.19 pawns in its favor. This massive search on the super fast Pentium III 500 MHz computer resulted in Fritz 6 finding the best of the three logical candidate moves available to it. The other two candidates were: 1) 22… Be6 23 Be2 (23 Rd6, Rc8 24 Nd2, Ng6 25 c5, Nf4 26 Rh1, Rxg5 27 Bxa6, Rc7 is both interesting and unclear), Ng6 24 Rdh1, Rg7 25 Rh6 and here Fritz continued with a silly line beginning with 25… Rag8? Instead, 25… Nf4 26 Bf1, Rd8 27 Nd2, Nd3+ is equal. Also, 2) 22… Ng6 23 Rh6 (23 c5, Be6 24 Rd6, Bxa2 25 Nd2, Be6 26 Nc4, Nf4 27 Rh1, Rxg5 28 Rxc6, Rag8 29 Kc2 is unclear), Be6 24 Rd2, Nf8 and now, instead of Fritz's 25 Rg2?!, White has 25 Rh5, Ng6 26 Rdh2, Rg7 27 Nd2, Rd8 again with equality. 23. g6, … This is the point of Fritz's wily last move. My planned line was 23 Rd6, which seems natural (read, "human"), since it activates a presently inactive Rook along a new front, the open d-file which is presently under my control. However, late last night I looked a little deeper and saw 23… Rxg5 24 Rxg5, Nxg5 25 Rxc6, Nxf3 26 Nd2, Nxd2 27 Kxd2 (my preliminary analysis ended here with the assessment of equality), Be6! 28 c5, Bxa2 29 Rxa6, Rd8+ 30 Rd6, Rc8 31 c6, Kg7 and Black has an edge since his h-pawn is a potential runner. Thus it's clear that the removal of my g-pawn creates a remote passed pawn of Black's h-pawn. My text move, oddly enough, forces my g-pawn's removal, but then I will capture Black's central pawn on e5. This allows my Rooks activity on both ranks and files and provides what I believe to be adequate compensation
As of today, Fritz 6 has been upgraded to Fritz 6a, a "new and improved" version, which has been shown to be even stronger than Fritz 6! This battle is tough enough as it is, without strengthening my adversary. Let's see what happens now. 23… , Rxg6 24. Rxe5, Rg3!? The diacritical marks are there only because Fritz 6a chose the most complex choice. As you readers know, I've been trying to unbalance this equal position in order to increase dynamism in an effort to exploit any computer mistake and thus win. However, Fritz 6a is clearly doing the same to me! The position is now extremely complex with potential pitfalls for both parties. Fritz calculated for 11 hours and 35 minutes to a depth of 18/42 ply (a new record?) and assessed the position as 0.16 pawns in its favor. In so doing, it rejected the clearly equal position reached after 24… Rg1 25 Bh3, Rxd1+ 26 Kxd1, a5 27 Kd2, a4 28 Ke3, Bd7 29 Nd2, a3 30 b3=. Surprisingly thisoption was not even one of Fritz's top 3 candidates, so perhaps it saw an "improvement" for White. Instead, Fritz's top rejected candidate was 24… Kg7 25 Rd6, Rg1 (now it's similar to my line above) 26 Nd2, c5 27 Kd1, a5 28 Ke2 (I considered 28 Rf5, Ra6 29 Rdd5, f6 [Black plans … Kg6 and then … Nc7] 30 Ke1, Kg6 31 Kf2, Rh1=.), a4 29 Kf2 when both the chess engine and I assess the position as equal. 25. Nd2, … The text move voluntarily traps White's Queen Rook in a passive position. While the effect is only temporary, White does abandon domination of the d-file. So why did I play this move? Well, the answer involves a very long story. Let me begin my paraphrasing Napoleon in "Animal Farm", "some positions are more equal than others!", and this is an example of less equal equality (read as, "highly unbalanced equality"). I spent the weekend analyzing the lines leading from my three candidate moves, 25 Rd6?!, 25 Rd3, and the text move. The first line, the one I planned to play, later proved to be inadequate since White doesn't get enough compensation for the gambit pawn. The latter two lines lead to equality, I think. However, I confess that this particular position is the most complex line I can ever recall examining! While the mere act of calculating these lines accurately was a great chore, assessing the resulting positions was an even greater problem. For example, in many lines I came to the conclusion that one side held an advantage. Then I'd find a new counter only to think that the other side had the advantage. This continued for two days! Most lines then evolved into a clouded assessment of unclear. Finally, last night, thinking I had a "feel" for this position, I was able to resolve some lines into an equal status. So why is this position so complex? Both sides have played sharply to unbalance the position. Black has a couple of trumps: a passed h pawn and a potential a-file opening for his Rook to penetrate into my back rank with … a5-a4, etc. White also has a couple trumps: White's pieces are presently more active than Black's and White frequently wins Black's c-pawn and then advances his c-pawn with great effect. Given that, here's a general summary of my much more detailed analysis. 25 Rd6?! (White sacs a pawn in order to activate his Rooks. This was my original intention, since it seemed my active pieces would recapture the material in a favorable position. However, we will see that Black is the one who's initiative prevails.), Rxf3 26 Nd2, Re3 27 Kd1, Nf4!? (Black plans … Bg4+. Instead, 27… c5 28 Nb3 [28 Rf5!?] is unclear.) 28 Rg5 (28 Re8+, Kg7 seems to give Black an edge while 28 Rxc6??, Bg4+ 29 Kc2, Re1 30 Kb3, Rxf1 31 Nxf3, Nd3 wins), Be6 29 c5, Bxa2! (29… a5 30 Rxc6, Rg8 31 Rxg8+, Kxg8 32 Ra6, Bg4+ 33 Kc2, Re1 34 b3, f5 35 Bc4+, Kg7 36 c6, fxe4=) 30 Rxc6,a5 (30… Ne6!?) 31 Rd6 (31 Bc4?!, Bxc4 32 Nxc4, Rxe4 33 Ne5, Rd4! [33… Rxe5 34 Rxe5, a4 35 Rb6, a3 36 bxa3, bxa3 37 Rb1, Kg7 38 Ra1, Kf6 38 Re3, a2 39 Rc3, h5 40 Rc2, Nd5 41 Ke2, Nb4 42 Rb2, Ra4 43 c6, Nxc6 [[43… Ke6?? 44 c7, Kd7 45 Rc1 wins for White]] is equal] 34 Kc2, Ng6 35 Nxf7, Kg7 36 Nd6, Rf4 37 Rc7, Kf6 38 Rh5, a4 and Black has a clear advantage) a4 32 c6, a3 and Black has an edge. My human emotions were at a low when I concluded that the machine was better here. 25 Rd3 (this line is similar to the text move, except White's Rook is feed to operate on the d-file before White traps it with a subsequent Nd2), Kg7 (25… a5!?) 26 Nd2, a5 27 Rd6, c5 28 Rh5 (28 Rf5, a4 29 Rdd5 is unclear as is 28 Kd1.), a4 29 Kd1 is unclear. However, 28 f4?, Nxf4 29 Rxc5, Be6! (29… h5?? 30 Rd8, Bb7 31 Rxa8, Bxa8 32 Rxa5 gives White a clear advantage) 30 Kd1, Re3 31 Rg5+, Kf6 32 Rb5, Ke7 33 c5, Ra7 and Black plans … h5 with a clear advantage. Black can also meet 25 Rd3 with 25… f6 when 26 Rh5 (26 Ra5!?), Nf4 27 Rd8+, Kg7 28 Rh1, Rxf3 29 Rg1+ (29 Nd2?!, Rg3 30 Rd6, c5 31 Rc6, h5 32 Rxc5, Bg4 33 Rc7, Kg6 34 c5, a5 35 Bc4, Rd8 and Black has the edge), Ng6 (29… Kh6?! 30 Nd2, Re3 31 c5, Re1+ 32 Kc2 gives White an edge) 30 Be2, Re3 (30… Rf2?! 31 Bg4, Bxg4 32 Rxa8, h5 33 Nd2, Nf4 34 Kc2, Ne6 35 Ra6 gives White the edge) 31 Bg4, Bxg4 32 Rxa8, h5 33 Rxa6, Ne5 34 Nd2, Re2=. Now that you have patiently read the above lines, you are ready to better understand why I chose the text move. Well, it was the only continuation where the first two Black lines I examined led to White's advantage. In fact, I had to actively look for Black equality. Psychologically, this fact made me favor this choice… a very human reaction I'd say. 25. … , Kg7 Fritz 6a calculated for 11 hours and 35 minutes to a depth of 18/42 ply, assessing the position as 0.28 pawns in its favor. This is large enough to say that it believes it has the edge. Of course, I disagree and call the position equal. To better understand our differences of opinion, a review of Fritz's candidate moves can be illustrative. For example, Fritz's 1st runner up candidate line was 25… a5 26 Rh5, Kg7 and now Fritz only considered 27 a3?? (Why would White voluntarily open lines that exposes his King?), bxa3 28 bxa3, Nd4 29 Bd3 and ends with the assessment that Black is better by 0.16 pawns. This erroneous assessment is strange. I see 29… Nxf3 30 Nxf3, Rxf3 31 e5, h6 32 Rg1+, Kf8 33 Kd2, Rb8 34 Rxh6, Rb3 35 Be4, Rf2+ 36 Ke1, Ra2 when Black wins the a-pawn and promotes his own to win the game. In this line, White instead maintains equality with 27 Rh2. Fritz 6a's other main candidate line was 25… Ra7 26 Rh5, Re7?! (This time the chess engine selects an inferior Black move. What future does the Rook have here? Better options were 26… Rg1 or 26… f6 with the idea of … Rag7 or 26… Nf4 planning … Be6 and then … Rd7.) 27 e5? (Yet another bad move. White gains an edge with 27 c5, a5 28 Nc4.), Nc5 28 Kc2, Re8?! (Best is 28… Be6 when Black has an edge after … Rd7) 29 Bd3, Nxd3 when the chess engine assesses Black as being better by 0.09 pawns. However I again disagree since 30 Kxd3, Bg4 31 Rh6, Bxf3 32 Nxf3, Rxf3+ 33 Ke4, Rf2 34 Rh1, f5+ 35 Ke3, Rxb2 36 Rxh7+, Kg8 37 Kf4, Rxd2 38 e6! (White eliminates a Black escape routs via … Kf7), Ra1 39 Rh8+, Kg7 40 R1h7+, Kg6 41 Rh6+, Kg7 42 Rxe8, Kxh6 43 e7 wins for White. I'll leave it to you readers to enjoy calculating the mechanics of the win. It can be a little tricky. In summary, these erroneous assessments and sloppy calculations prove the computer is only "human". Those errors occurred in lines involving precise calculation (the chess computer's forte) but also have a technical element (the chess computer's Achilles heel). Unfortunately for me, my silicon opponent's errors have only occurred in the side notes and not in the game line. 26. Rh5, … This is the only move I considered. Black's passed h-pawn is Fritz's biggest trump, so I aim to neutralize it. White now has the option of doubling Rooks on the h-file and is part of the plan behind my previous move. 26… , h6?! I always try to consider all logical candidate moves when analyzing positions. But, I never considered this move. I suppose that's because I expected Black to continue … Nf4 at some point, forcing my Rook to retreat and thus enabling Black's h-pawn to advance to h5 in one move. As such, Fritz's text move represents a potential loss of tempo. Thus the selection of this move is surprising considering that Fritz 6a calculated for 20 hours and 45 minutes to a depth of 19/19 ply (this must be a new record). Regardless, it assesses the position as 0.28 pawns in its favor, meaning it believes it has an edge. Instead, I thought the best lines for Black were the two "runner-up" lines that it rejected, namely 26… a5 27 Rh2, Rg1 or 27… Nd4. I won't go into any detailed analysis here other than to state that these lines, while still leading to equality, put a great deal of pressure on White. As such, they try to wrest the initiative. Black still needs to be careful here though since innocent looking lines such as 26… Kg6? 27 Rh1, h5 28 Nb3, Nf4 29 Rd6+, Kg7 30 Rxc6, Rxf3 31 Nd4, Rf2 32 c5 give White a clear advantage. In summary, Black's last move wasn't bad. The diacritical marks mean that better options were available; options that are more consistent with Black's previous moves. Both sides strove to unbalance the position to date with sharp lines, only to have this tempo loser remove some of the tension. 27. Rh2, … This is the only line that makes sense to me. I want to exchange material via an eventual Rg2 or Bh3. By exchanging material, my King will be safe to rush to the Kingside to aid in stopping Black's h-pawn. Also in some lines, Bg2 is best, so I need to support the Bishop. Instead, 27 Bh3??, Nf4 28 Bxc8, Nxh5 29 Bb7, Rd8 30 Bxc6, Nf4 wins for Black. 27. … , c5 Fritz 6a calculated for 18 hours to a depth of 18/43 ply and evaluates the position as 0.28 pawns in its favor, meaning that it claims a small edge. I still think the position is equal. Black had three other viable lines to analyze; I won't go into any detail other than to point them out. 27… Nd4 28 Rg2 (Fritz 6a thought 28 e5 best, but I'm not impressed with its analysis here), Rxg2 29 Bxg2, Be6 30 Rh1, Rh8 31 Rh4, Kg6 32 f4, h5=. Also 27… a5 28 Bh3, Nc5 (28… Nf4)=.Finally, 27… Nf4 (Black traps White's
28. Be2, … I looked at seven candidate moves here and chose this one since it creates the greatest chances for Black to err. White plans to double Rooks on the h-file. I will only list five of the six remaining candidates; the sixth move I looked at turned out to be so stupid that I'm embarrassed to admit to even considering it! I think this late night (1 am) chess analysis has a negative effect upon me. My first consideration was trading Rooks with 28 Rg2?, Rxg2 29 Bxg2, Nf4 30 Bf1, Bd7! 31 Nb3, Ba4 32 Rd6, Bxb3 33 axb3, h5? 34 Kd2, h4 35 Ke3, Ne6 36 Rd5, Rh8 37 Bh3, Kf6 38 Rd6=.However, when I checked my notes, I saw 33… Ne6! (this is easy to overlook) 34 Kd2, Nd4 35 Ke3, Nxb3 36 Rd5, a5 and Black has a clear advantage. 28 Rh3, Rg5 29 Rh4, Nd4 30 Bd3, Be6 31 f4, Rg2 32 Rdh1=. 28 Bd3, Bb7 29 Rdh1, Rh8 30 Bc2, a5 31 Rh5, Ba6=. I'm not confident of this analysis though so possibly Black's play can be improved. 28 Bh3?!, Nf4 29 Bxc8, Rxc8 30 Rhd1, Rc6 31 Kc2, Rcg6 when I favor Black with an edge. 28 e5? initially seemed attractive since it looks like a freeing move. However, 28… Bb7 29 Bd3, Rd8 30 Ne4, bxe4 31 Bxe4, Rxd1+ 32 Kxd1, Rg5 gives Black a clear advantage. 28. … , Nd4 This reply was a mild surprise. The alternative of 28… Nf4 is more logical and thematic (read better), since it supports 29… h5. For example, 29 Rdh1, h5 30 Nb3, Be6 (Black still needs to be careful since 30… Rg5?! 31 Bf1, Rg3 32 Rf2, Rg5 33 Kd2, a5 34 Ke3, Ne6 35 Rfh2 gives White an edge) 31 Nxc5, Nxe2+ (31… Rd8? 32 b3 is advantageous for White) 32 Rxe2, Bxc4 33 Rc2, Bb5 34 Rxh5, Rg1+ allows Black to draw by perpetual check. To avoid the draw, I would have to find a different 33rd move. Evidently Fritz 6a, after 18 hours and 40 minutes of calculating to a depth of 20/20 ply, sees something it likes in this line. This search depth is a new record! Dr. Robert Hyatt wrote that Deeper Blue searched to a depth of 14-17 ply when it defeated Garry Kasparov in their match. Thus it's really intimidating to have Fritz 6a search 17-21 ply deep against me! After this deep search, it evaluates the position as 0.34 pawns in its favor, meaning a small edge. Well, I don't know what it is finding in its search. I still think the position is equal; in fact, Black needs to search carefully in order to avoid giving White back the advantage. 29. Rdh1, … This was my plan all along… doubling Rooks on the h-file to negate Black's passed pawn, Black's only chance for an advantage. Still, the alternative 29 Nb3 is interesting. For example, 29… Nxb3?! 30 axb3, Be6 31 Rdh1, Rh8 32 Rh5, Rg5 33 f4, Rxh5 34 Rxh5, Rc8 35 Bf3 gives White the edge. Thus Black needs to find salvation in 29… Nxe2 30 Rxe2, Be6 31 Nd2 (31 Nxc5, Bxc4 32 Rc2, Bb5 3 f4, h5 is unclear), h5 32 Rh2, Rh8 33 Rdh1, Kg6 34 f4, Bg4 when the position is unclear. However, White lost a couple tempi in this line from shuffling the Knight around. Therefore my text line aims for similar strategic goals but without the loss of time. This means Black must be careful to avoid allowing White to develop an advantage. 29… , Be6!? Fritz 6a calculated for 21 hours and 30 minutes to a depth of 21/21 ply (a new record!) at an average speed of 412 kilonodes/second. It downgraded its assessment to 0.22 pawns in its favor. The general point of the text move is to move the Bishop to enable the Queen's Rook to slide over to h8 for defense of the beleaguered h-pawn. This is a controversial choice, however, as Black's Bishop can become a target after White plays f4-5.. Had I been Black here, I would have accomplished the same task with 30… Rh3 31 Rh3 (31 Rg1+, Kf6 is unclear), Bxh3 31 Rxh3 (31 Bd1, Be6 32 f4 planning f4-5, is unclear), Nxe2+ 32 Kd1, Nf4 (32… Nd4 33 Rh5, Rc8=) 33 Rh4, Nd3 34 Kc2, Rd8 35 Rh5, Ne1+ 36 Kc1, Nd3+, which draws by perpetual check. I'm impressed that Fritz 6a is "smart" enough to detect that its strong Knight is better presently than my weak Bishop. Thus it correctly
30. Bd1, … I feel all the more intimidated by Fritz 6a's exceptionally deep 21/21 ply search when I confess that I made no search here at all! My instinct was to relocate my Bishop to a better location where it now threatens to become active with Ba4. I also plan to drive away Black's Bishop with f4-5. This combined plan offers White chances to assume the initiative again. This was the reason why I thought Black's last move was controversial. This is also a good time to discuss technique. Most of us know that in middlegame positions, we can benefit by posting our pawns on the same colored squares as the opponent's Bishop to minimize its scope. However, in the endgame, we should almost always post those same pawns on the opposite color to avoid having the Bishop prey on the now weak pawns. When a lot of pawns remain on the board during the transition from middlegame to endgame, this change in strategy can cost a lot of time in moving them. One thus tries to develop a long-range strategy where pawn advances to opposite colors figure into the tactics. Presently 4 of my 5 pawns remain on White squares (the same squares attacked by Black's Bishop) so I need to develop a long-range strategy with this concern in mind. Given that I'm planning f4-5 soon, this can be a double-edged plan with some short-term gain but potential long-term problems because my pawns will remain on White squares. Therefore I'll analyze more deeply on my next turn before committing to this scheme. Again, it needs to be kept in mind that White has several other options here that maintain easy equality. However, my style of play is to always push to unbalance positions to the point where the chances for a decisive result increase. There is clearly a fine line that one should not cross. On one side of the line, the position can be controlled and a draw can be obtained. But on the other side of the line, loss in unavoidable. My style of play, while risky, has given me positive results to date against opponents I presumed to be human. However, when playing against chess engines searching in excess of 20 ply, this can be courting disaster. Still, I promised to play as if I my opponent were human and to remain true to my style. I think, regardless of the result, this will prove most educational regarding human versus computer battles in correspondence chess since the chess engine will be under constant pressure, as will I, to find the proper lines. 30. … , Rh8 31. f4, … Fritz 6a calculated for 11 hours and 30 minutes on its last turn, analyzing to a depth of 20/20 ply at an average speed of 418 kilonodes/second. It evaluated the position as 0.25 pawns in its favor.A natural human response now would be to throw one's arms up and say, "I can't compete with 20-22 ply searches!" However, I think that this is the stage of the game where humans can outperform even deep searching chess engines. In transitions from middlegames to endgames, human technique and intuition has historically proven superior to chess computers in OTB play. Admittedly in OTB chess, the chess engines are not allowed to calculate as deeply as in this correspondence match. So, let's see if this highest rated of all chess engines is correct in its assessment of this position when allowed to search over 20 ply deep. I think it's wrong. White's text move was chosen after minimal calculation. I admit to apprehension in selecting this move without analysis while my opponent is calculating every logical move within its search horizon. But the options for both sides are so great that I didn't know where to begin to analyze. Instead, I intuitively chose to remove my f-pawn from the double attack on it, thus gaining space on the Kingside while creating additional scope for my Bishop. This plan, threatening f4-5 with all the risks mentioned in my previous commentary, gives White chances to regain the initiative. 30 f4 also prevents … Rg5 as a response to Rh5. For example, the immediate 31 Rh5 prevents the advance of Black's h-pawn and attacks Black's pawn on c5, but now 31… Rg5 32 f4, Rxh5 33 Rxh5, Rc8 34 f5, Bd7 35 Nf3 allows dull equality. Now I think Black has a difficult choice to make in finding a path to equality. 31… Bd7?! Wow! What perfect timing for Fritz 6a to err; right after my previous comment. Fritz 6a calculated for 17 hours to a depth of 20/20 ply, at an average speed of 416 kilonodes/second. It now assesses the position as only 0.09 pawns in its favor, meaning total equality. However, I think White is now better! Black's move prevents Ba4 and prepares redeployment with … Bc6. My previous commentary suggested that Black has to struggle for equality. I think Black's best chance was 31… f5 32 e5, Bf7 33 Re1 when White may have a minute edge. Black's f-pawn is now fixed on a light square, blocking his Bishop, while White's Kingside pawns have successfully shifted to dark squares. This is the culmination of White's strategy mentioned in the notes to my 30th move. Black
32. Nf1, … It is interesting to note that Fritz's downgrade of its position came during a period of super-deep calculations by the machine, while the human instead made intuitive moves based upon almost no analysis. However, I sensed immediately that Fritz's last move was inferior and so spent a great deal of time in analysis. The text move was not my first choice. Initially, I was attracted to the nice and technically superior White positions reached after 32 Rh5 followed by 33 Rf5. But late last night (it's dangerous to analyze deep into the night, in this case 1:30am), I saw that the text move allows me to eventually sac my c-pawn and then force Black to sac his Knight for my e and f pawns. In exchange, White gains a ferocious attack using the extra piece. The final result was too complex for me to calculate, but every defense I examined for Black looked pretty bad. In summary, were I allowed to play in anti-computer strategy, I'd opt for the superior endgame White achieves after 32 Rh5. However, the potentially sharper text move would have been my choice versus a human and I have promised to play versus the chess engines as if they were human. Given their calculating skills, this can be suicidal! A summary of the alternative is, 32 Rh5, Ne6 33 Rf5 (now 33 Nf1, Rd3 34 Rf5, Rd4 35 Nd2, Rd8 36 Bh5, Be8 37 Rh2 is unclear), Bc6 34 Bh5, f6 35 Re1, Rc8 36 Nb3, Ba4 37 Na5 (37 e5, fxe5 38 fxe5, Bxb3 39 axb3, Rc7 is unclear), Rg2 38 Nb7! (38 e5, fxe5 39 fxe5, Rd8 allows Black to draw by perpetual check after … Rdd2), Nd4 39 Rxc5, b3 40 a3, Rc2+ 41 Kd1, Rxc4 42 Rg1+, Kh8 43 Rxc4 (43 Rxc8+, Rxc8 44 Kd2 gives White an edge), Rxc4 44 e5, fxe5 45 Nd6, Rc5 46 fxe5, Bc6 47 Kd2. White has the edge here since he is the only one with a meaningful passed pawn. If 47… Rc2+??, then 48 Ke3, Ne6 49 Nf7+ wins. Please forgive this long analysis. I'm a strong believer that long analysis equals wrong analysis. Worse yet, the analysis supporting my text is move is just as long and much more tactically complex. Please bookmark this position since I believe that the ultimate result for this game can be traced back to this decision. I now think that the text line grants White at least an edge. However, if I'm ultimately proven wrong, then we can see that the human passed up a chance to transpose into a superior endgame in favor of an attacking middlegame. If wrong, this will be the second time that I got the superior position and then misplayed it (see my 14th move). 32. … , Rg6! Fritz 6a had a choice between an active defense (32… Rd3?) which I expected, or the more passive text move. Fritz 6a calculated for 18 hours and 45 minutes, to a depth of 20/20 play at an average speed of 420 kilonodes/second. It downgraded its assessment of the position further to only 0.06 pawns in its favor. The reason that I chose my 32nd move was because the natural looking 32… Rd3? leads to 33 Rg1+ (33 Rh5, Ne6 transposes back to 32 Rh5 lines analyzed previously. Instead 33… Rxd1+?! 34 Kxd1, Bg4+ 35 Kd2, Bxh5 36 Rxh5, Rc8 37 Ng3, Kg6 38 Rd5, a5 39 e5 gives White an edge), Kf8 34 Rh5, Ne6 35 f5, Ng7 36 Rh2, Rd4. Now, as previously mentioned, White has a diabolical line here that exchanges the c-pawn and the e and f-pawns for a Black piece. White's extra piece then joins in a fierce attack when it was difficult for me to find a satisfactory Black defense. Here is a summary: 37 Bc2! (37 Nd2, bc6 is unclear), Rxc4 38 Nd2, Rd4 39 Nf3, Rd6 40 e5, Rb6 (40… Rc6 41 e6!, Nxe6 42 fxe6, Bxe6 43 Ne5 gives White a large advantage) 41 e6!, Nxe6 (41… , fxe6?? 42 f6, Ne8 43 Ne5, Bc8 44 Ng6+, Kf7 45 Rf2, Rg8 46 Ne5+, Kf8 47 Rxg8+, Kg8 48 f7+ leads to checkmate) 42 fxe6, Bxe6 43 Rh5, c4 (43… Bxa2!? 44 Rxc5, Be6 45 Rc7 gives White a large advantage) 44 Nd4, Ke7 45 Re1 (45 Nxe6?, fxe6 46 Rg7+ is unclear, given the number of Black pawns. Black can probably draw this, since even if Black's Kingside pawns were eliminated, the remaining pawns are all on the Queenside.), when White has a much better position. It was late at night when I found this line, so I may have missed better lines for Black, but it just seemed to me that adequate defenses were scarce for Black. 33. Bh5, … This was my initial plan and the only one that I believe maintains a White edge. Instead, 33 f5 (This move takes a great deal of courage to play, since White puts both Kingside pawns on light-squares. Thus White potentially burns his bridges with such a plan. Meanwhile the slow and methodical 33 Ne3, Re8 34 e4, Bc6 35 Nd5 [35 Rf1, Be4 36 Rh5, Rg3 37 Re1 is unclear], Rd8 is only equal.), Rg5 34 Ne3, f6 (34… Re8 35 Rxh6, Rxe4 36 f6+, Kf8 37 Rh8+, Rg8 38 Nd5 gives White an edge) 35 Nd5, Be8 allows Black to create a fortress-like position where White finds it difficult to effectively penetrate. For example, 36 Nf4 (36 Nc7, Bc6 37 Rh4, a5 38 Nd5=. Instead, 38 Na6?, Nxf5 39 exf5, Bxh1 40 Rxh1, Rxf5 gives Black a clear advantage.), Bc6 37 Re1, Re8 38 Nd3, Rxe4 39 Rxe4, Bxe4 40 Nxc5, Bxf5 41 Nxa6, Nc6 42 Bf3, Ne5 is only equal. Yet more tries are, 36 b3, (36 Bc2?! Nxc2 37 Kxc2, Bf7 38 Re1, h5 39 Re3, Rg4 [39… h4? 40 Reh3, Rg4 41 Kd3. White threatens to wins Black. s h-pawn.] 40 Kd3, Re8 gives Black an edge.), Bc6 37 Bc2, Nxc2 38 Kxc2, h5 39 Rh4, Rg2+ 40 Kb1, Rg4!=. 33…, Re6 34.Nd2 … On Fritz's last move, it calculated for 17 hours and 30 minutes to a depth of 18/44 ply over an average speed of 411 kilonodes/second. It believes it is better by only 0.16 pawns. I expect it to recognize a White advantage very soon now. 34. …, Bc6 35.Re1, … Fritz 6a calculated on its last move for 17 hours to a depth of 20/47 ply, at an average speed of 425 kilonodes/second. To my surprise, it believes it's regaining a miniscule edge with an assessment of 0.34 pawns in its favor. 35. …, Kf8 36.Bd1, … Fritz 6a calculated for 19 hours and 30 minutes to a depth of 20/45 ply, at an average speed of 422 kilonodes/second. It now downgraded its claimed advantage to 0.28 pawns. White's move prepares Rh5, threatening penetration into Black's position, with advantage to White. Instead, 36 Bg4?, f5 37 Bd1, fxe4 gives Black an edge. 36…, Ke7 Fritz 6a was given a tough choice; try to find the lesser of two evils. Fritz calculated 18 hours and 30 minutes to a depth of 21/21 play, at an average speed of 419 kilonodes/second. Surprisingly, it still favors its position by 0.22 pawns. Objectively, Fritz may have selected the best move, but it will have another tough decision on its 38th move, in trying to avoid giving White a large advantage. The alternative here was 36…Rf6 when White can choose between 37 Rh5, Rxf4 38 Rxc5, Rg8 39 Rh5, Rf6 40 e5 or 37 f5, Ke7 38 Bg4, Rd6 39 e5, Rdd8 40 Ne4 (40 f6 also favors White) when White has a small advantage. 37. Rh5, … White continues with his plan commenced by the 32nd move. There were no
37. …, f5 Although Black gives White a passed pawn, Black now takes total control of the a8-h1 diagonal. As such, this offers Black the best practical chances for equality, since this potentially enables a Black Rook to claim the g-file. Instead, 37…Kd6 38 Bc2 presents Black with a tough problem regarding how to proceed. Now 38…Nxc2? (38…Rg8?? 39 f5, Re5 40 Rxh6+ wins) 39 Kxc2, Rhe8 40 e5+, Kd7 41 Nb3, Ba4 42 Reh1, Rh8 43 Kd3, Bxb3 (43…Kc6? 44 Ke4, Kb6 45 f5, Bc6+ 46 Kf4, Re7 47 Rxh6+ wins) 44 axb3, Rg6 45 Ke4 gives White a large advantage. This may be one of those positions where one should not try to actually do anything. Here it's now best to simply hold things together, forcing White to prove any advantage. The best Black line I could find was 38…a5 39 Nb3, Nxc2 40 e5+, Ke7 41 Kxc2, Ba4 42 f5, Rc6 43 Reh1. In this static position White has the edge. However with advance preparation, White will play f5 and then Rxh6 with probably decisive consequences. Since Black's only "trump card" is his passed h-pawn, its loss leads to a miserable existence and an inevitable defeat for Black. I don't know how much of the above Fritz 6a saw. Regardless, it calculated for 17 hours and 30 minutes to a depth of 21/21 ply, at an average speed of 424 kilonodes/second. Surprisingly, it evaluates the position in its favor by 0.13 pawns, which is still equality. I can't imagine how Black can even remotely consider this position equal, given the pressure it is under 38. e5, … Now White's pieces are no longer needed to support the e-pawn. The plan here is to exchange Knights with 39 Nb3, Nxb3 40 axb3. This opens the a-file for potential penetration of Black's defense. Now please envisage this position with all pieces removed. You will see that the resulting King and pawn endgame is easily won for White. Since Black's King must blockade White's e-pawn, White's King is free to capture Black's h-pawn and then triangulate to victory. That means that piece exchanges favor White and thus are integral to White's long-range strategy and short-term tactics. 38. …, Rg8 Exactly as forecast in my commentary to Black's 37th move, Black stakes claims to the open g-file. A similar motif is 38…Rg6 39 Nb3, Nxb3+ 40 axb3, Be4 41 Bc2, Bxc2 42 Kxc2 which clearly favors White. Fritz 6a calculated for 20 hours and 10 minutes to a depth of 21/47 play, at an average speed of 421 kilonodes/second. To my surprise, it still favors its position by 0.16 pawns, meaning dead equality. 39. Nb3, … As mentioned in my notes to White's 38th move, the King and pawn endgame is won for White. Presently, Black's weak f-pawn is supported only by a Knight. I expect to exchange that Knight, so Black's Bishop will then have to support that pawn. I plan to exchange Bishops too, similar to the line in Black's 38th move commentary. In short, White thus exploits the weakness in Black's pawn structure to facilitate a liquidation of pieces. I am therefore curious how much longer Fritz 6a will continue to evaluate the position as equal. A human player in the Black position would feel quite uncomfortable. 39. …, Nxb3+ 40. axb3, Be4 41. Bc2, … Fritz 6a calculated for 19 hours and 30 minutes to a depth of 24/24 ply (a new search depth record for this game!) for its last move. This was accomplished with an average calculation speed of 423 kilonodes/second. Fritz 6a continues to call the position equal, favoring its side by 0.09 pawns. Black will likely see that 41…Bxc2? leads to a bad endgame, so I expect Fritz 6a to calculate that 41…Bf3 is forced. Ideally, it would be nice to proactively prevent 41…Bf3 with 41 Rh3, but Black's Bishop can now simply chase the Rook away. Instead, if White tries 41 Rh2??, Reg6 42 Bc2, Bxc2 43 Kxc2, Rg2+ 44 Re2, Rxe2+ 45 Rxe2, h5 46 Rh2, Rg4 47 Rxh5, Rxf4, then Black has a clear advantage. 41. …, Bf3 42.Rh2, … As expected, Fritz 6a saw that it's survival chances after 41…Bxc2? were very slim, so it elected to keep Bishops on the board for the time being. Fritz 6a calculated for 18 hours to a depth of 24/24 ply, tying its search depth record. Its average speed was 419 kilonodes/second, calling the position equal by favoring its side by 0.03 pawns. These computer assessments amaze me. Like humans, they optimistically favor their side. However, I don't think any human player with the black pieces would feel comfortable in this position. If Black can really equalize, he can do so only by the skin of his teeth. I think White has the edge, since Black has potentially exploitable pawn weaknesses. Also, Black must live in fear of trading pieces, since the present King and pawn endgame is an easy win for White. Meanwhile, White's text move is the only attempt for advantage since 42 Rxf5 (42 Baxf5??, Bxh5 43 Bxe6, Kxe6 44 Rh1, Bg6 45 Rxh6, Kf5 46 Rh4, Rd8 wins), Bg4 43 Rf6, Rxf6 44 exf6+, Kxf6 45 Re5, Rc8 46 Kd2 is unclear, but probably equal. 42. …, Bg4 43. Kd2, … Fritz 6a's last move was forced. Still, after calculating for 16 hours and 45 minutes to a depth of 24/24 ply, it kept the same assessment as for its last move. White's move, preparing to centralize the King and allow Rook access to the a-file, is also the only continuation to trouble Black since 43 Bd1, Bxd1 44 Kxd1, Rg4 45 Rf1 is unclear. 43. …, h5 The computer chess engine calculated for 21 hours and 45 minutes to a depth of 23/23 ply. It now rates the position as 100% equal at 0.00 pawns in nobody's favor. At least it's nice to know that it no longer believes that it has any shred of an advantage. Fritz 6a's last move is the most logical since Black is going to play this move anyway. Therefore it's interesting that Fritz 6a favored 43…Rh8 after 18 hours and later liked 43…Reg6. All of these choices can transpose, but why play moves that aren't essential to the plan? Since the text move always figures into Black's plan to advance its passed h-pawn, the other moves are thus seen as imprecise. Nonetheless, a fourth alternative was available in 43…Rd8+. My original analysis considered 44 Ke3, h5 (see, here's this essential move again) 45 Ra1?!, Rc6 46 Ra5, Kd7 47 Rd2+, Ke7 48 Rd5, Rdc8 49 Ra1, h4 50 Bd3 (50 Rh1, Rh6 51 Rh2 may or may not amount to anything for White), Rh8 51 Rh1 (51 Be2?, h3 52 Bxg4?, fxg4 53 Kf2, h2 54 Rh1, Rh3 55 Kg2, Rch6 gives Black a large advantage), when White may have a microscopic edge. It's therefore probable that White should gain time by not probing the a-file. Instead, White's Bishop's position should be improved.44. Ke3, … This is the only way to maintain what I believe to be a very small White edge. White's edge consists of the fact that Black has no hopes for advantage once the promotion of his h-pawn is negated. Meanwhile Black must respond to White's initiative, rushing his pieces to different squares to protect his pawn weaknesses, such as after White plays Ra1. Black's Bishop is now a big pawn, but White's Bishop has a future elsewhere. White now centralizes the King with options of moving to the Kingside to Blockade Black's h-pawn. My plan is to improve the position of White's Bishop with either Bd3-e2 or Bd3-f1-g2-d5, depending upon circumstances. White's only other option at this point is 44 Ra1?!, Rd8, transposing into the line given to Black's last move.44. …, Rh8 Fritz 6a selected this move fairly early in its search. After 6 hours and 40 minutes, it reached a depth of 22/22 ply, assessing the position as dead equal. However its search was not completed until 18 hours and 15 minutes had elapsed. By then it had searched to 23/47 ply and favored White microscopically by 0.03 pawns. My first thought was, how can any human compete with search depths with extensions to 47 ply? This is truly inhuman! Can you imagine competing against a never tiring, tactically precise, 100% logical opponent who searches some lines 23 ½ moves deep? But my second thought, after advance analysis, was that it missed the most precise move of 44…Rd8!, preventing the move played in the game. So maybe humans can handle 23 ½ move deep searches by their opponents. Let's see what happens. 45. Bd1, … My latest analysis convinces me that the deep-searching computer's assessment is indeed correct; the position is objectively equal. Yes, White does have an initiative, but it evaporates against correct defense. The text move is the most effective way to mobilize the Bishop since the fastest way to the a8-h1 diagonal is Bd1-f3. This creates a sharp position where both sides have chances of losing, although objectively Black should be able to hold this position. Instead, the alternatives fizzle out after, for example: 45 Rh4, Rd8 46 Reh1 (46 Bd3, Rd4=), Rd4 47 Ra1 (47 Rxh5?, Bxh5 [47…,Rxe5?? 48 fxe5, Bxh5 49 Rxh5 wins] 48 Bxh5, Rg6 49 Kf3, Rg1 gives Black a clear advantage)=. Also 45 Bd3, h4 46 Bf1, Reg6 (46…, h3?! 47 Bg2, Rh4 48 Rg1 [48 Kf2, Bh5=], Reg6 49 Bb7, Kd7 50 Kf2, Ke8 [what else?] 51 Rg3, Rh7 52 Bd5, Kd7 53 Rd3 may give White a small edge) 47 Bg2, Bh5 48 Bd5 (48 Bb7?, Rg3+ 49 Kf2, Rxb3 50 Kg1, Bf3 favors Black), Rg3+ 49 Kf2, Rxb3 50 Rxh4, Rxb2+ 51 Ke3, Rh6 52 Reh1, Rg6 53 Rxh5, Rg3+ 54 Bf3, Rb3+ 55 Ke2, Rbxf3 56 Rh7+ draws by repetition. Finally, 45 Reh1 (45 Ra1, h4=), Rd8 (45…,Kf7!?) 46 Rxh5, Bxh5 47 Rxh5, Rg6 48 Kf3, Rd2 49 Bxf5, Rgg2 50 Rh7+, Kf8 51 Bd7, Rgf2+ 52 Ke4, Rd4+ 53 Kf5, Rdxf4=. 45. …, Rg6 Fritz 6a calculated 9 hours and 35 minutes to a depth of 22/22 ply, and now assesses the position as favoring Black by 0.13 pawns. Black threatens to exchange Bishops and then win a pawn with …Rg3+. These computer programs always look for forcing tactical solutions, rather than "playing the position" as humans usually do. By playing the position, I refer to a patient maneuvering of the pieces in accordance with the dictates of the position, with the intent to avoid being too committal. One thus tries to accumulate small advantages by exploiting the opponent's positional weaknesses while minimizing the opponent's ability to do the same. I expected the more human looking 45…h4. Then a brief summary of my notes continues: 46 Bf3 (46 Rg1, Bxd1 [46…, Rg6 transposes back to the game line} 47 Rxd1, Rg6 48 Rhd2, Rg3+ 49 Kf2, Kf7=), Bxf3 (46…h3?! 47 Bd5, Rb6 [47…Rg6?! 48 Ra1, Rc8 49 e6 + ) 48 Ra1, Rc8 49 Kf2 and White gets an edge after Ra5, Kg3, and Rh1-a1) 47 Kxf3, Reh6=. 46. Rg1, … White hereby counters Black's threat and sets a trap that an unsuspecting human might fall for if he continues with his planned threat. However, I don't expect any computer program to miss 46…,Bxd1?? 47 Rxg6, Bxb3 48 Rxa6, Bxc4 49 Ra7+, Ke6 50 Rd2, Bd5 51 Ra6+, when White wins Black's Bishop. White only logical alternative was 46 Bf3, Bxf3 47 Kxf3, h4 48 Rh3=. This line is too insipid, relinquishing White's small initiative. 46. …, h4 47. Bf3, … The chess program calculated for 17 hours and 45 minutes to a depth of 23/23
47. …, Kd8?! Black's reply is really strange and I can't figure out what the purpose of this move is. Possibly it's to lure White's Rooks to the d-file in pursuit of some ill-fated attack, when Black subsequently exchanges Bishops and then advances the h-pawn. I doubt that this is a move many humans would find; most humans wouldn't play it without first deeply analyzing the consequences. Franklin, the computer operator wrote, "This position was difficult for Fritz. When I checked this morning it was considering three moves. When I shut down tonight, it was still considering two moves, very closely ranked. The other move was 47…Rhh6 when the evaluation was about 0.06 pawns behind the recorded move. The third move that it considered was 47…Kd7." Regardless, Fritz 6a calculated to a depth of 20/46 ply after 16 hours and 50 minutes. The average calculation speed was 411 kilonodes/second. It now favors its position by 0.09 pawns. Instead, I expected 47…h3 48 Rg3 (48 Bd5, Rh4 49 Rg3=), Rh4 49 Bg2 (49 Bxg4!?), Rgh6 50 Bd5, when White has the subjectively better side of objective equality. Why wasn't this one of the top three candidates for Fritz? 48. Bd5, … Objectively, I think that Black's 47th is inferior to the 47…h3 line that I anticipated since I now think White may have a small edge. White plans Ra1 when Black's King must rush to the defense of his menaced Queenside pawns. White can then switch back to Re1 to advance the e-pawn and stretch Black's defenses very thin. This move order avoids 48 Ra1?, Bxf3 49 Kxf3, Rg3+ 50 Ke2, Rxb3 when Black has a clear advantage. However, the text line is very tricky and it's quite easy to miscalculate, so we'll just have to wait to see who was correct in calculating through these complications. White's alternative line was 48 Rd2+, Ke7 49 Rgd1 (49 Rd5?, Bxf3 50 Rxg6, Bxd5 51 cxd5, h3 52 Rg7+, Kf8 53 Rg1, h2 54 Rh1, Rh3 55 Kd2, Rxb3 56 Kc2, Rh3 gives Black a clear advantage), Re8 50 Rd7+, Kf8 51 R1d6, Rxd6 52 Rxd6, Bxf3 53 Kxf3, Ke7 54 Rh6, Rd8 55 Ke2, Rg8 56 Rh7+, Ke6 57 Rxh4, Rg2 58 Ke1, Rxb2=. Black had an inferior alternative in 48…Kc7?! 49 Bd5 (49 Rd5, h3 leads to unclear lines), Rgh6 50 Re1! (50 Ra1, R8h7 51 Kf2, Rg7 51 Kg1, Bd1+ 53 Rg2, Bg4 54 e6, Kb6 55 Rh2, Re7=), h3 51 Rh2 (white plans Bf3 or Kf2) when White has an edge. 48. …, Rgh6!? Fritz 6a calculated for 20 hours and 40 minutes to a depth of 22/22 ply, at an average speed of 419 kilonodes/second. The chess engine favors its position by 0.06 pawns. While most moves are of equivalent value presently, Fritz 6a selected a move that I had not considered. Franklin Campbell wrote that the chess engine also examined 48…Kc7 and 48…h3, believing them to be 0.03 pawns less attractive. The later two moves were also considered by me and can transpose into what I thought to be the mainline, which begins with a fourth alternative: 48…Rh7 49 Ra1, Kc7 50 Kf2, Kb6 51 e6, h3 52 Rhh1, Re7 53 Rhe1, Rh6 54 Re3, h2 55 Kg2, Rg7 56 Kh1, Re7 57 Rc1 (White plans Rc2) when I think White has a minimal edge. 49. Ra1, … I believe the text move is White's only try for an advantage. The alternative was 49 Bf3, which can present some difficulties if Black were to misjudge the position. Were I to employ "anti-computer" tactics rather than play what I deemed best, then I might hope to expect 49…Bxf3? 50 Kxf3, Ke8 (50…R8h7 51 Rg8+, Kc7 52 Rg5, Rf7 53 Re2, Kd7 54 Kg2, Ke7 55 Kh3 + ) 51 Rg7, R6h7 52 Rg5, Rh5 53 Rxh5, Rxh5 54 Kg2, Rh7 55 Kh3 +. Instead Black's proper reply is 49…Rg8 50 Rd2+, Kc7 51 Rdg2 (51 Rd5, h3=), Rhg6 52 Bd5=. White has accomplished nothing with his maneuvering. The point of White's move is to spread Black's defenses thin by pulling Black's King away from the center and over to b6. White's Rook can then race back to e1 in one move and support the pawn advance e6 which divides Black's forces and makes it difficult for his Rooks to quickly support the queenside pawns that can be threatened with a doubling of White Rooks on the a-file. Black will have to dedicate on Rook to a blockade on e7, which may help White overload other areas of Black's defenses. At least I've been able to work out an edge with this philosophy on my board. Let's see what this great inhuman defender can do. 49. …, Kc7 50. Kf2, … The chess engine calculated for 18 hours and 15 minutes to a depth of 23/23 ply, at an average speed of 423 kilonodes/second. It favors its position by 0.19 pawns. Meanwhile, White's last move, as planned, clears the e-file for an eventual Re1, supporting the advance of the e-pawn. There is a secondary point to the move as well. If White's king can replace the Rook as the blockader on the h-file, then White's Rook pair is freed to seek out targets in Black's camp. 50. …, R8h7 Fritz 6a calculated for 17 hours to a depth of 23/23 ply, at an average speed of 420 kilonodes/second. It still favors its position by 0.19 ply. The text move was apparently Fritz's choice from the beginning since Franklin reported that this move was favored when he checked it at the 8-hour stage. It is probably the best choice. After all, it must be played eventually, so to maintain flexibility it should be played now. Black plans to double Rooks on the 7th rank, creating a barrier to stop White's passed e-pawn. A related line runs 50…Kb6 51 Rhh1, h3 52 Rh2, R8h7 53 Re1, Rg6 54 Re3, Rhg7 55 e6, Re7 (55…Bh5?? 56 Rhh3 wins for White) 56 Rh1, Kc7 57 Kg3, Rgg7 (57…Kd6?? 58 Ra1, Ra7 59 e7 wins for White) 58 Kh2 when White has an edge. In this position White's Rooks are free to probe and exploit Black weak points. For example, White can double Rooks on the a-file or instead play Re5. Meanwhile Black's Rooks are tied to passive defense and his Bishop is a "big pawn". Still, Black's defenses may hold unless White can either overload his forces on a defensive weakness or penetrate Black's wall. 51. e6, … White divides Black forces, forcing Black's King to rush to the b6-square to cover his weak a-pawn. Therefore I offered a conditional move here: "if 51…Kb6, then 52 Rhh1". This line may eventually transpose into the line given on Black's last move. White tries to draw some of Black's forces away from the center in an effort to use White's greater mobility to double Rooks on the e-file. Black will then have to shuttle his forces efficiently to cover weaknesses on a6 and h4 and to stop White's passed pawn. Correspondence Master Joe Shipman subsequently wrote, " the poor machine has no clue that you plan to make like Andre Agassi with it (keep it running from one side to the other until it falls down). 51 e6! is a beautiful move, but the machine's limited aesthetic sense only allows it to appreciate coarse concepts like 'pawn on the same color as the Bishop' but not refined ones like 'cuts the opponent's position in half' ". Thanks Joe. You did a better job communicating my plan than I could have. 51. …, Kb6 52. Rhh1, … On its last move Fritz 6a calculated for 17 hours to a depth of 25/25 ply, at an average speed of 433 kilonodes/second. Now the darned machine favors its position by 0.34 pawns! I'd really like to know what it "thinks" is so favorable about its position. Black is much less comfortable than White. White has greater space, more options for piece mobility (Black's Bishop is a "big pawn"), and an initiative due to some Black weaknesses that may be exploitable. Objectively, Black can probably hold this position together, but Black is compelled to react to White's actions and thus is certainly not superior here. Others have contacted me to say that their chess programs often will favor one side of an "equal" position. Then when given the opposite side, they rate their new color as having the edge. This tendency by the chess engines to always favor their position is something that I recommend that chess
52. …, Rg7?! We've seen repeatedly in these 4-game matches that the chess engines often don't know what to do in positions with few clear threats to analyze. This problem is much more evident in OTB chess where their calculating times are severely restricted, but even after calculating 18 hours to a depth of 23/23 plies, Fritz 6a played a technical mistake. For the record, its average computational speed was 421 kilonodes/second and it still favors its position, but now only by 0.25 pawns. Most humans know that opponent's passed pawns should be blockaded. This Rook will need to blockade with …Re7 on the next move so why waste the tempo? Normally in closed positions such as this with no immediate threats, a wasted tempo is of minor importance. However, two of my analytical main lines now lead to White wins! Yes, I believe Black's play can be improved in those lines, but the point is that the tempo still should not be squandered. In short, the chess engine exhibited poor technique even after 18 hours of rapid non-stop analysis. I believe Black's best defense was to immediately begin the blockade with 52 …Re7. Then, White has two lines that seem to offer a small edge: 1) 53 Ra2 (53 Kg2, h3+ 54 Kh2, Rexe6 55 Bxe6, Rxe6 56 Rhe1=), R6h7! (53…h3? 54 Rha1, Ra7 55 e7, Bh5 56 Rae1, h2 58 Re5 + .) 54 Kg2, h3+ 55 Kh2, Rhg7 56 Rha1, Ra7 57 Re1 when White plans Re3 and Raa1-e1 with an edge. 2) 53 Rhe1, h3 54 Re3 (54 Kg3?!, h2 when Black plans …Rh3+ with an unclear position), h2 55 Kg2, Rg7 56 Kh1, Re7 57 Rc1 planning Rc2 when White has an edge. 53. Rhe1, … Now White plans to go into Andre Agassi mode (see the commentary to White's 51st move) and force Black's defenders to switch back and forth and from side to side to cover weak spots in their lines. As such, the text move is the only serious attempt at a White advantage. 53. …, Rhh7 !? Amazing! I confess I never considered this move. My mind knows that passed pawns must be blockaded so I naturally assumed that 53…Re7 was the only move. This is the second consecutive time that Fritz 6a has deferred blockading my e-pawn. Still, there may be some reason for the blockade deferment. My guess is that Fritz saw tactical value in posting a Rook on the g-file. Now if White's King moves to the h-file to blockage Black's pawn, Black then delivers a discovered check that enables it to activate its passive Bishop. Instead, my late-night analysis produced 53…Re7 54 Re3, h3 55 Rh1, Kc7 56 Kg3, Kd6?? 57 Ra1, a5 (57…Ra7? 58 e7 wins) 58 Rxa5, h2 59 Ra6+, Kc7 60 Re1, Rh3+ 61 Kg2, Kd8 (61…Rxb3 62 Rea1, Rxb2+ 63 Kh1 wins) 62 Kh1, Rd3 63 Ra8+, Kc7 64 Rea1, Rd1+ 65 Kxh2, Rxa1 66 Rxa1, Rh7 67 Kh1, Rh3 68 Ra7+, Kb6 69 Rb7+, Ka6 70 Rb8 wins. This illustrates that Black's position is inferior and that his defensive resources are being stretched thin. Fritz 6a calculated for 19 hours and 15 minutes to a depth of 23/49 ply. This was done at an average speed of 423kilonodes/second. Much to my surprise, the chess engine has downgraded its assessment of its advantage to only 0.16 pawns. Fritz has yet to acknowledge its difficulties, but this may be movement in that direction. 54. Re3, … This move is a logical continuation of my plan. At the appropriate moment I plan to double Rooks on the e-file. Another option is to move this Rook to d3 and reposition my Bishop, giving me potential penetration into Black's weak center. The other candidate move was 54. Ra2, threatening to double Rooks on the a-file and then subsequently move at least one Rook back to the e-file. Both moves have their merits but are somewhat mutually exclusive. I chose the text because it's not clear that I need to double Rooks on the a-file, but all my plans involve having a Rook on e3. 54. …, Re7 55. 55.Kg2, … Black evaluated the position after its 54th move as favoring itself by 0.13 pawns, so it's continuing to gradually downgrade its assessment. It came to this conclusion after computing for 18 hours and 35 minutes, to a depth of 24/24 ply. The computations were done at the speed of 431 kilonodes/second. White is consistent with his aforementioned plan. If Black's h-pawn is to ever advance, it would be ideal to blockade it with the King on a dark square (h2) so that Black's light-squared Bishop would continue to lack any counterplay. Now White maintains prospects of either doubling Rooks on the e-file or even playing Bf3 at the appropriate moment. This later idea weakens Black's support of the f-pawn, which can then be attacked via Re5. It also opens the d-file for penetration into Black's defenses. Timing is the key and White is in no hurry. Sometimes the best execution of a plan is to delay it by repositioning ones pieces to anticipate counterplay. This stratagem gives the opponent, especially an opponent lacking a plan, chances to err if it doesn't "see" what's eventually coming. This works well versus humans who can miss short-term tactical shots, so let's see how it works against a machine who's short-term tactical ability is nearly perfect. 55.…, Rhg7 56. Kh2, … As expected, Fritz 6a chose a tactical solution based upon discovered check. It calculated for 16 hours and 20 minutes to a depth of 23/48 ply, at an average speed of 431 kilonodes/second. It now favors its position by 0.16 pawns. The other option
56. …, Rg8 Fritz 6a finally sees the light by acknowledging that it has no advantage. In fact, after 15 hours and 5 minutes of calculating, it now favors my position by 0.12 pawns. It reached this decision at a depth of 25/25 ply and an average speed of 428 kilonodes/second. Of course, 0.12 is so insignificant as to be meaningless, but since I've detected an innate preference to always favors its own position, the 0.28 pawn shift to White has relevance. Fritz's move is best in that it frees this Rook to transfer to other areas in need of bolstering. It also anticipates White's opening of the d-file by defending with a timely …Rd8. 57. Rae1, Rgg7 Fritz 6a calculated for 18 hours at an average speed of 432 kilonodes/second, to a depth of 24/24 ply. It continues to favor White by 0.12 pawns. I expected Black to leave the Rook on the 8th rank in order to have the flexibility to rush that piece to areas that need shoring up. However, 57…Rh8 (Franklin Campbell noted that the chess engine's second choice was 57…Rd8, which leads to a similar conclusion) 58 Bf3, Bxf3 59 Rxf3, Rg6 60 Rfe3, Rg4 61 Rfe3, Kc6 61 Re5 gives White an edge. Fritz 6a may have calculated the above line as well, since its move now prevents White's plan. For example now, 58 Bf3, Bxf3 59 Rxf3, Rg6 60 Rfe3, Rg4 61 Rf3, Rg6, when the position repeats. 58. Bg2, … There's more than one way for White to follow his plan to exchange Bishops. White now plans an eventual Re5 and Bh3. Yes, White's Bishop is superior to Black's, but the resulting trade exposes many weak points in Black's pawn structure. As a consequence, Black's Rooks will be tied down to the defense of his pawns while White's Rooks go on the offensive, as does White's King. Black's King can only play only a defensive role, guarding his backward a-pawn. 58. …, Rg6 Fritz 6a calculated for 19 hours and 15 minutes at its last turn, searching to a depth of 20/22 ply. It calculated at an average speed of 443 kilonodes/second. It now thinks the game is totally even, showing a 0.0 pawn assessment. Darn! I hoped it was getting tired of defending and thus growing despondent. I think a human with the Black pieces would be worried about now. 59. Re5, h3 This move is very committal, since Black's h-pawn is now fixed on a light square (the same as my Bishop) and within potential capturing distance from my King. Because of this, I originally expected a more flexible move, biding time until White commits to a plan of action. Thus I think most humans would have selected 59…a5 60 R1e3 (60 Bh3??, Bxh3 61 Kxh3, Rg3+ 62 Kh2 [62 Kxh4??, Rg4+ is mate in one], Rxb3 when Black has an edge.), Rh6 61 Bh3, Bxh3 62 Kxh3, Kc6 63 Re1, Rg6 64 Rxf5, Rexe6 65 Rg5, Rxe1 66 Rxg6+, Kd7 67 Kxh4, Re2 68 Rg7+, (68 Ra6?, Rxb2 69 Rxa5, Kc6 70 Ra6+, Kb7 71 Re6, Rxb3 72 f5, Rf3=.), Kc6 69 Rg3, Rxb2 70 f5, Kd7 71 Kg5, Ke7 72 Kg6, Kf8 73 f6 (73 Kf6!?), Re2 74 Rg5, when White has a clear advantage. It's possible that Fritz 6a calculated much or all of this line and thus preferred the text move. The chess engine calculated for 18 hours to a depth of 25/49 ply, at an average speed of 447 kilonodes/second. It still evaluates the position as being totally even, with a 0.0 pawn assessment. 60. Bd5!, … White had several options here, though 60 Bxh3??, Bxh3 61 Kxh3?, Rh7 checkmate was certainly not one of them. While examining various options, I couldn't help but recall Lasker's maxim that, "The threat is greater than its execution". I think that applies to my approach here since by returning the Bishop to its former square, I show that I'm in no hurry to exploit Black weaknesses yet. Instead, I plan to maneuver to create more. If correct, Black's defenders will become contortionists in trying to defend weak points while I defer threat executions until a time of my choosing. If timed properly, my break-though will catch Black in an inferior defense. Psychologically this should have no effect upon silicon, but humans with the Black pieces would really suffer now. An example of a premature threat execution is the natural 60 Bf1, Rf6 when White has two mainlines: 1) 61 Bd3 and 2) 61 R1e3.
Sorry for the reams of analysis, but the point is that I can presently find no way to load up on Black's h-pawn and still reach a favorable conclusion. The threat remains but I'll now pursue options of doubling Rooks on the a-file. This will draw Black's Rooks to the defense of its a-pawn. Meanwhile, Black must maintain a blockade of my e-pawn. When I first examined this motif, I got the Black pieces tangled up on a couple occasions. So Black's defense is difficult. Let's see how Fritz fares. I still think White is better here but the position is too complex for me to give an accurate assessment. Some people have written to me saying they think White is winning. I think it's safe to say that White has at least an edge. 60. …, Rh6 Fritz 6a calculated for 21 hours and 30 minutes to a depth of 26/52 ply, at an average speed of 436 kilonodes/second. It continues to assess the position as totally equal at 0.00 pawns. Fritz really need not find any critical moves at this juncture, since it need only wait to see what White plans to do before committing itself to any specific defense. 61. Ra1, … The commentary to my last move indicated that I believed I had some advantage. However, I now believe that the position is indeed even with best play from Black. What caused my change in assessment? Well, I think that my initial optimism was fueled by a difficulty in finding adequate defenses for Black. However, eventually I found a defense. Once I understood the position better, I began to find several adequate Black defenses. So it seems I suffered from an all too human trait of optimism, which caused me to fool myself. When my error was discovered, another human trait manifested itself…depression and “burn-out”. This might be a good time to explain to you readers what it is like to be involved in a match of this type. In a word…scary. The inhuman chess engine never tires. It has no family obligations and needs no relaxation time. Having no human feelings, it never feels emotional exhaustion caused by long hours at a high-stress job followed by late night suppers and dealing with family dynamics. So virtually every day, there's a new move waiting for me from Fritz 6a, which I must deal with. Since I usually respond the next day, the only time I have for chess analysis is late at night after my family goes to bed. This nocturnal behavior is a problem for my wife, which further compounds our “family dynamics” on subsequent days. Also late night analysis after long days at work leads to fatigue and potentially sloppy analysis. Since my moves are on public display, any error by me shows the world what an idiot I really am. This adds additional pressure. This routine has been repeated day after day and night after night and month after month for over a year now. So what kept me going in this game? It was the belief that I was gradually outplaying the machine to redeem myself for not winning the game at the beginning when I should have. Gosh, how I wish I could go back in a time machine to replay my 14th move, and thus win a miniature! Anyway, when the truth dawned upon me that the position is really equal after all, I simply couldn't look at this game for about two weeks! Last night I somehow dragged myself back to the chessboard and to a review my notes. A brief survey of the alternatives is as follows:
61. …, Rg6 Fritz 6a continues to "mark time" until White's intentions become clear. The chess engine calculated for 19 hours and 30 minutes to a depth of 25/25 ply, at an average speed of 427 kilonodes/second. It still rates the position at totally even (0.00 pawns). 62. Ree1, … White elects the most flexible path. There is still the option to double rooks on the a-file or to wait for an opportunity to challenge along the g-file. Both sides now decide to maneuver to wait for the other to commit to a plan. 62. …, Rgg7 Fritz 6a calculated for 16 hours and 40 minutes to a depth of 27/27 ply, at an average speed of 438 kilonodes/second. It still rates the position at totally even (0.00 pawns). Black momentarily doubles Rooks on the 7th rank to ensure that White's e-pawn won't advance. The text move is much like the others I examined in that Black is maintaining flexibility for any eventuality. Assuming that White plans to double Rooks on the a-file, then other lines also work. For example: 62 …Rg8 (62…Rh6 63 Ra2, Rh8 64 Rea1, Ra7=) 63 Ra2, Rh8 64 Rea1, Ra7=. If however, Black wants to deviate from the above plan and try something active rather than continue with passive defense, then his position deteriorates. For example, in the above line, 63…Bh5? 64 Rea1, Ra7 65 Rg1, Rxg1 66 Kxg1, Bd1 67 Ra1, Bxb3 68 Re1, Re7 69 Kh2, a5 70 Kxh3 gives White a large advantage. 63. Rg1, … This is the only way that I can find to maintain the tension. The plan involves possibilities of trying to activate White's King with hopes of exploiting weaknesses in Black's Kingside pawns or of trying to create connected passed pawns after exchanging Bishops. For example, 63…Rg6 64 Rae1 (64 Rg3??, Rgxe6 65 Bxe6, Rxe6 and Black is better due to 66…Re2+), Rg8?? 65 Bf3, Kc7 66 Bxg4, fxg4 67 f5, Kd6 68 Re2, Rf8 69 Rd1+, Kc6 70 Re5 wins for White. Yes, Black can improve and hold his defense together with 64…Rgg7, but it's an illustration of a plan that shows that Black must remain vigilant to any White initiative. 63. …, Rg6 64. Rae1, … Please read the comment to the previous move for an explanation of events. Fritz 6a calculated for 19 hours and 5 minutes to a depth of 26/26 ply, at an average speed of 431 kilonodes/second. It continues to evaluate the position as totally equal (0.00 pawns). 64. …, Rh6!? Black moved off the g-file in order to avoid getting hit with 65 Bf3. Still, I think that 64…Rgg7 65 Rg3, Rg6 66 bf3, Rexe6 67 Rxe6+, Rxe6 68 Bxg4, fxg4 69 Rxg4, Re2+ 70 Kxh2, Rxb2 71 Rg3, a5 72 Rf3, a4 73 bxa4, Rc2 74 f5, Rc3 gives Black the safest path to equality. Instead, by moving off the g-file, White has the constant threat of an exchange sac on g4. Regardless, Fritz 6a calculated for 18 hours to a depth of 28/28 ply, at an average speed of 438 kilonodes/second. It continues to evaluate the position as (0.00) pawns. 65. Re5, … White prepares 66 Rxg4 by trying to gain a vital tempo. This can be effective if Black is unaware of the threat. In so doing, White has just transposed to a line analyzed after White's 61st move. I originally planned the immediate exchange sac with 65 Rxg4, fxg4 66 f5, Rf6 67 Re5, but Black can draw without too much effort. The three main lines are:
So, while I make no claims for a White advantage after 65 Re5, I like the fact that the exchange sac is still a possibility, with the potential to gain a favorable position if Black's defense is inexact. 65. …, Rf6 Fritz 6a is apparently determined to allow me to sacrifice the exchange. Instead, it
Black could also have safely vacated the g-file with 65…Rh8 when I would have elected 66 Rg3=. Instead, 66 Rxg4?!, fxg4 67 f5, Rh6 seems easier for Black to play. Regardless, the chess engine calculated for 19 hours and 30 minutes to a depth of 27/27 ply, at an average speed of 432 kilonodes/second. It still believes the position is totally equal, claiming a 0.00 pawn advantage. 66. Rxg4, … White has no winning prospects in 66 Rg3, Rh6, transposing into the 65…Rg6 66 Rg3, Rh6 discussed on the previous move. Therefore I've elected to sacrifice the exchange to unbalance the position. Now 66…fxg4 67 f5 transposes back to the lines analyzed after White's 65th move. While this unbalancing motif is a standard operating procedure for me against humans, it is extremely risky against calculating machines that search over 25 ply deep. Any error in calculation by me will probably be swiftly refuted. But, since chess engines are materialistic, it is possible it will incorrectly assess the resulting positions and thus play an inferior line. But this debate is irrelevant; I've promised to play against the chess engines as I would against a human opponent. So my risky text move is thus “forced”. Let's see what happens now. 66. …, fxg4 I expected the materialistic chess engine to now assess the position in its favor, given its material advantage and appetite for material. To my surprise and delight, it now favors my position by 0.06 pawns. Whew! That's a relief because I was afraid that it would immediately find a bust to my analysis. But, maybe it will anyway over the course of the next few moves. Fritz 6a calculated for 17 hours and 10 minutes to a depth of 25/25 ply, at an average speed of 419 kilonodes/second. 67. f5, Rh6 (!) Fritz 6a unbalanced matters as well by selecting the sharpest response. This move was not one of those I considered in my notes to White's 65th move, but may transpose at some subsequent point. Nonetheless, this line is so sharp and double-edged that both sides may walk a tightrope while trying to make the other fall off. Objectively, this line seems to offer Black the greatest chances of victory and therefore seems best. Fritz 6a calculated for 17 hours and 25 minutes at an average speed of 415 kilonodes/second. During that time, it calculated to a depth of 23/56 ply! This 56 ply extension is remarkable because it's the deepest search I can recall to date. Imagine the feeling I had when I read that, knowing also that Fritz 6a is the highest rated commercial chess engine in the world. So when it searches to a depth of 56 ply, my natural responses are of frustration and worry. What chance do I or any humans have in such sharp positions after a 23/56 ply analysis against the strongest chess program on a Pentium III computer? Well, let's find out. My one consolation is that the chess engine still does not believe it has any advantage. Instead, it returned to its former assessment of 0.00 pawns. Subsequent to writing the above paragraph, Israeli computer chess expert, Uri Blass, wrote the following to me.
Thank you for your input, Uri. 68. Kg3, h2 This is the point of the chess engine's previous move. It threatens to promote the h-pawn, so White's reply seems forced. Fritz 6a calculated for 16 hours and 45 minutes at an average speed of 441 kilonodes/second. During that time it searched to a depth of 23/23 ply. This depth is an interesting surprise in that it didn't perform any huge search extensions, like the 56 ply search it conducted last time. The chess engine now favors my position by 0.03 pawns. I think the position is objectively equal, but matters will get tricky over the next few moves. Both Fritz and I soon have to make some big decisions about moves that will affect the future course of the game. 69. Bh1, Re8 (!) In this position, Fritz 6a has no option other than to find a waiting move. Black can't improve his position and so must remain non-committal until White shows his plan. For example, Black's rooks must remain on their files. One must remain on h6 to both protect the passed h-pawn and to prevent White from playing f6. The other rook must stay on the e-file to prevent the advancement of White's past e-pawn. Black's King needs to defend its c-pawn. Therefore the only candidate moves for Black are 60…a5 and 69Re8. In general, piece moves are superior non-committal to pawn moves, for the simple reason that a pawn can never retreat once advanced. So for this reason, in theory, 69Re8 is the better waiting move. As we will soon see, it is also better in reality. After 69…a5?!, White's primary winning chances are seen in lines after 70 Kxg4. Instead, 70 Re2, Kc7 71 Re5, Kb6 repeats the position. Also, Black can continue to wait with 70…Re8 (70…Rh3+?? 71 Kxg4, Rh6 72 Kg5 wins) 71 Rxh2, Rxh2 72 Kxh2, Kc7 (72…Rh8+?? 73 Kg3!, Rxh1 74 e7 wins) 73 Kg3, Kd6 74 Kxg4, Ke5 75 Bc6, Re7 76 Bd7, Rg7+ 77 Kf3, Kf6 78 Ke4, Rg5 79 Kd5, Rxf5 80 Kd6, Rf2 81 Kxc5, Rxb2 82 Ba4, Kxe6 83 Kb6 which draws. So, the dynamic lines are seen after 70 Kxg4, Rg7+ 71 Kf3, when Black has three options:
The point of showing these variations is not to maintain that they are concrete. Instead, it's to illustrate ideas and motifs. All three Black options are quite similar. White can try to use the motif seen in the 71Rh3+ line to show that 69a5 is probably the second best waiting move. I don't know how Fritz 6a came to the same conclusion I did, but it made its decision after 18 hours of calculating, at an average speed of 419 kilonodes/second. During that time it reached a search depth of 22/52 ply. The chess engine now assesses the position as being totally equal at 0.00 pawns. 70. Kxg4, … Just as in the previous example, White's primary chances lie in taking the pawn. Another advantage of 69…Re8 is that White can't bail out with a draw now after 70 Re2??, since 70…Kc7 (70…a5?? transposes to 69…a5) 71 Re5 (71 Rxh2, Rxh2 72 Kxh2, Rh8+ 73 Kg1, Kd6 wins), Kd6 (in the 69…a5 lines, e7 was unavailable for Black's King) and Black now wins. However, if White is afraid of entering the dynamic positions ahead, he can instead find a draw with 70 Rd5, Re7 71 Re5, repeating the position. Also 71 Rd6+, Ka7 72 Rd2, Rh3+ 73 Kxg4, Rh8 is equal. After the text move, I think the position is objectively equal. Still, it is dynamically imbalanced so any error in either calculation or assessment may prove fatal. Subjectively, I think the burden is upon Black to find a defense, since White's attack plays itself. A human playing Black would experience a great deal of pressure now. 70. …, Rg8+ The text move seems forced, preventing White from playing Kg5 to support the advance of the connected passed pawns. Fritz 6a calculated for 18 hours and 20 minutes to a depth of 21/21 ply, at an average speed of 424 kilonodes/second. The chess engine "thought" that my pawn capture motif unbalanced the position too much since it's formerly even/draw assessment just changed to -0.94 pawns in its favor. This is a major devaluation, signifying a large Black advantage. But my belief is that my calculations are correct and instead, the chess program is once again showing error in its ability to evaluate positions. Let's see who is correct here, the human or the machine. 71. Kf3, … White's only other choice is 71 Kf4, Rh4+ (71…Rg1 72 e7, Rh4+ 73 Kf3, Rh8 with equality) 72 Kf3, Rh3+ 74 Kf4, Rg1 with equality. These lines transpose into lines resulting from the text move. Still I preferred the text move on principle. Black will likely chose to transfer his Rook to g1 and then subsequently check my King from behind and attack my passed pawns from behind. After all, Rooks generally are most effective behind passed pawns, whether one's own or one's opponent's. The text move thus keeps White's King near the back rank to minimize Black's drawing chances by attacking from the rear. For example, after my text move, assume (for sake of argument) that Black's Rook on g8 can be transferred to f1 with 71…Rf1+. Now White can reply with 72 Kg2, forcing Black's Rook to lose a tempo to retreat. White then wins with 73 e7. So although it would be nice to have the King assist with the advancement of his passed pawns, the King is instead more effective guarding the back rank. 71. …, Rg1 Fritz 6a calculated for 18 hours and 5 minutes to a depth of 18/44 ply, at an average
Black's primary alternative was 71…Rh3+ 72 Kf4, Rg1 73 e7, Rh8 74 Re6+, Kc7 which can transpose into the expected game line. Then, White has two options:
72. e7, Rh8 Fritz 6a calculated for 19 hours and 30 minutes to a depth of 21/54 ply, at an average speed of 422 kilonodes/second. To my relief, it now downgraded its assessment to a 0.50 pawn advantage for itself. Although this still represents a large and possibly winning advantage to Fritz's “mind”, I think the trend is more meaningful. In short, Fritz is getting deeper into the position with each move and is now beginning to see what I've said all alongBlack really has nothing here other than a draw with best play. In fact, these positions are quite easy for Black to lose. Black had two other options:
73. e8/Q, … White's reply is forced since 73 Re6+?? enables Black's King to gain a tempo over the text line. This tempo allows Black's King to approach White's f-pawn so, no Rook need be diverted to prevent its advance. This is illustrated by 73Kc7 74 e8/Q, Rxe8 75 Rxe8, Rxh1 76 Rh8, Rf1 77 Kg2, Rb1 78 Kxh2, Rxb2+ 79 Kg3, Rxb3+ 80 Kf4, Rb1 81 Rh7+, Kd6 82 Rh6+, Ke6 83 Rxa6, b3, which wins for Black. I now offered some conditional moves: if 73…Rxe8 then 74 Rxe8, and if 74…Rxh1 then 75 Rh8. 75 Rh8 is forced because 75 Kg2??, Rb1 76 Kxh2, Rxb2+ 7 Kg1, Rxb3 78 Kf2, Rd3 79 Re6+ (79 f6, Rd6 wins), Ka5 80 f6, Rd8 81 f7, Rf8 82 Re7, b3 wins. 73. …, Rxe8 74. Rxe8, … The first conditional move was accepted after 12 hours of calculation. During that time Fritz 6a searched to a depth of 22/50 ply at an average speed of 422 kilonodes/second. The chess engine still favors its position by 0.50 pawns. I still predict that Fritz 6a will soon downgrade its assessment to equality once it has searched a few ply deeper into the position. 74. …, Rxh1 The darned machine just upgraded its assessment to 0.53 pawns in its favor. This conclusion was reached after 17 hours and five minutes of calculations to a depth of 25/25 ply at an average speed of 426 kilonodes/second. This 0.53 pawn advantage is apparently considered to be quite a large advantage by the chess engine. Therefore one of us is seriously wrong in the assessment of this position based upon our respective calculations. This makes for some excitement to find out just who is in error, since this error can be potentially disastrous for either side. 75. Rh8, Kc7 Fritz 6a “thinks” it's making progress because it again upgraded the evaluation to 0.56 pawns in its favor. It came to this conclusion after calculating for 21 hours and 15 minutes to a depth of 24/51 ply. Calculations were performed at an average speed of 436 kilonodes/second. These evaluation upgrades amaze me since I can find no advantage for Black. But then, I may be overlooking something obvious. To date, these four match games have shown both chess engines to have a great ability to find strong moves, but no reliable capacity to evaluate positions, especially technical positions such as endgames. Perhaps my exchange sacrifice has confused the chess engine's evaluation process to the point where Fritz 6a believes its extra material conveys an advantage. 76. f6, … This endgame has a “forced” nature to it, so I'll continue to offer conditional moves to speed matters along. Therefore, if 76…Rf1+ then 77 Kg2, and if 77…h1/Q+ then 78 Rxh1. 76. …, Rf1+ 77. Kg2, … Now this is getting positively scary. Fritz 6a again upgraded its assessment to 0.72 pawns in its favor. It calculated for 17 hours, to a depth of 25/55 ply at an average speed of 437 kilonodes/second. This trend of upgraded evaluations at a moment when I forecast downgraded evaluations worries me. Like most paranoid humans, I'm beginning to doubt myself and wonder what, if anything I missed. If I did miss something, is it because I'm stupid? If I'm stupid, then the entire world will see just how stupid I am. This is not a pleasant feeling. Fortunately for the chess engine, it never experiences psychological doubts. It feels no emotion when I sacrifice the exchange to advance my passed pawns or evaluate the position in opposition to its claims. One aspect of human-chess engine battles that gets overlooked is psychology. Even in correspondence chess, psychology exists, although seldom to the degree seen in OTB chess. But, versus machines, we get no subliminal input from seeing our opponents shaky or suddenly sloppy handwriting. There is no communication exchanged, such as a postcard or e-mail saying, "Help!" Even if no conscious communication exists, we humans can still convince ourselves that we can discern something from our adversary's communications, even if it is only reading something into how long it takes them to respond. Instead, the machine is cold, giving no clues and having no feelings. Another aspect that's probably only found in correspondence battles against chess engines, is the physiological angle. The machine never experiences fatigue. It never comes home late from work or gets exhausted from stress. It has no family obligations or gets nagged by an unhappy spouse. It never gets bored. It never has interruptions in its "thought" and never fights to find time to analyze. These games have continued for about 1.5 years and the pressure never lets up. I'm tired but Fritz 6a isn't. I feel like I'm running a marathon while Fritz 6a is fresh and ready to sprint to the finish. I'm not making excuses, dear readers. Merely making observations of fact. I feel something like Yuri Gagarin, the first man to fly into outer space and return safely. In this respect, I'm probably the first and only person in the world to play a 4-game match against the two highest rated chess programs on a Pentium III computer, giving the chess engines 17-22 hours to calculate for each move, and allow the match to last for 1.5 years without let-up. I lack motivation to drag myself to the chessboard. Worse yet, I don't have the desire to double/triple check my analysis anymore for blunders, or even to analyze at all. Instead I have a fatalistic mentalitywhatever happens, happens. I believe that I have nothing to worry about in this position, and yet I worry because of Fritz 6a's evaluation upgrades. Let's see who's evaluating incorrectly. I still think I'm right and Fritz is wrong. 77. …, Rxf6 Something is very strange here. Fritz 6a calculated for 16 hours and 50 minutes to a depth of 25/55 ply and further upgraded its evaluation to 0.84 pawns in its favor! The strange thing for me is that 77…Rxf6 looked so drawish when I prepared my exchange sac that I didn't even bother to set up this position. After all, we've entered a Rook endgame where all pawns are confined to the a-c files and are either blocked or partly blocked. On principal alone this should be a draw. But then, maybe I need to learn more about endgames. My impression here is that this is yet another example of chess engine inability to understand endgames. This is understandable at the short time controls of OTB chess. But given almost 17 hours of calculation at 422 kilonodes/second to a depth of nearly 26 ply, this is difficult to accept. My guess is that the materialistic machine calculated that it can now capture my pawn on b3 but didn't notice that my King will then capture its pawn on c5, creating a passed pawn. The alternative is 77…h1/Q+ 78 Rxh1, Rxf6 79 Re1 (as long as White can protect both b2 and b3, the position is drawn), Rd6 80 Kf3, Rd2 81 Re2, Rxe2 82 Kxe2, Kd6 83 Kd2, Ke5 84 Ke3, a5 85 Kd3, Kf5 96 Kd2, when white maintains the opposition and draws. As such, Fritz's text move is tougher to meet since White can't interpose a Rook between his King and b-pawns as in this example. A similar line is 78…Rxh1 79 Kxh1, Kd6 80 Kg2, Ke6 81 Kf3, Kxf6 82 Kf4, a5 83 Ke4 draws. 78. Rxh2, Kb6 ?! My previous comment discussed how Fritz played the best line, making my life a little tougher because I can't immediately interpose a Rook between my King and b-pawns. But its last move returns the tempo, allowing me to reposition my Rook. I thought computer chess engines are supposed to be logic machines. But where's the logic in this last move? Nonetheless, here's an embarrassing story. For whatever reason, I didn't get Franklin's e-mail of Fritz's last move until after I saw it posted here at Franklin's web site. A shiver ran down my spine when I saw that Fritz played a move that I never considered. Knowing that Fritz's evaluations claimed a win for itself, I suspected the worst. I suspected that this surprise move must be part of a winning plan (remember my earlier comments about paranoia?) that I'm just too stupid to see. At this point I must point out that my chess vision is terrible when looking at a 2-dimensional computer screen. I can't analyze at all that way, so I always set up my pieces on my board at home for analysis. So, while sitting here at work and looking at the screen, I assumed that Black had used the tempo to protect his c-pawn prior to winning one of my b-pawns, while simultaneously preventing my King from moving to the Queenside. So, I thought this might be evidence of a strong plan and called Franklin to tell him that! I was impressed by Fritz's idea since machines aren't supposed to be able to plan, although its super-deep analysis gives the impression of constructing a plan. But when I got home last night and set the position up on my board, my worries and chess blindness evaporated in seconds. The reason that I never considered Fritz's move is because… it's illogical to allow White to free his Rook from behind his King! No informed human would have played such a move here, so why should I have considered it? Now the only mystery is why the second highest rated chess program (Fritz 6a was the highest rated program until just a few days ago, when its younger brother, Deep Fritz, pushed it into the number two position) can't see that this is an elementary draw. Fritz 6a calculated for 18 hours, to a depth of 25/25 ply at an average speed of 444 kilonodes/second. It downgraded its evaluation to 0.66 pawns in its favor. While the downgrade is correct, it should have assessed the position as a dead draw, rather than a substantial advantage. The proper (read logical) method of play still leads to a draw, but forces White to find the correct moves. This line continues with 78…Rd6 79 Kf3, Rd3+ 80 Ke4, Rxb3 81 Kd5, Kb6 82 Rh6+, Ka5 83 Rh2, Rg3 84 Kxc5 (perhaps Fritz didn't want to allow me to obtain a passed pawn), b3 85 Rd2, Rg5+ when White has two routes to a draw:
79. Kg3, … Virtually everything draws here. For example 79 Rh5, Rd6 80 Rf5, Rd2+ 81 Rf2 is a draw. The point of the text move is similar. Now 79…Rd6 is met by 80 Rf2. 79. …, Rd6 80. Rf2, … Exactly as forecast on my last move. Fritz 6a calculated for 18 hours, to a depth of 25/48 ply at an average speed of 442 kilonodes/second. But once again I'm disappointed by the chess engine's inability to properly evaluate this position. It has now slightly upgraded its assessment to 0.69 pawns in its favor. Hopefully chess program authors will use these match games as evidence that they really need to improve the ability of their programs to assess certain positions: principally endgames and closed positions. All informed humans can see that this position is drawn within a few seconds. So why can't this highly rated program, mounted on a fast Pentium III computer and given 18 hours of searching at the rate of 442,000 positions/second, “understand” this as well? 80… Rd3+ The chess engine calculated for 19 hours and 50 minutes to a depth of 24/24 ply, at an average speed of 379 kilonodes/second. The darned machine still favors its position, but by 0.63 pawns this time. 81. Rf3, Rd6 Fritz 6a calculated for 19 hours and 15 minutes to a depth of 25/25 ply, at an average speed of 379 kilonodes/second. It again downgraded its assessment, this time to 0.53 pawns in its favor. 82. Rf2, … I've just repeated the position after White's 80th move to emphasize that this position is a draw. Let's see if the machine gets the hint. 82. …, Kc6 Fritz 6a got the hint and played something new after calculating for 19 hours. During that period, it searched to a depth of 24/47 ply at an average speed of 377
83. Rf3, … This “simple” and statically dull endgame can be converted into a dynamically rich position if White is willing to sacrifice a pawn to unbalance matters with 83 Kf3!?. The point of this endgame gambit is to offer material in exchange for time to capture Black's c-pawn, thus making White's c-pawn a passed pawn. Normally, I wouldn't hesitate to unbalance matters, as evidenced by my play in these match games to date. However, just as in business and life, a risk/reward consideration exists. My opponent has shown great calculating ability in the endgame, so I should not expect chances for victory unless Black's chances to err are significant. They are not. Instead, it is possible that I may have miscalculated the pawn sac lines. If so, my error would likely cost me the game. As such, I think the risk is unwarranted. Therefore my choice to play the dull text move would be the same even if I thought my opponent were human. Specifically there just are insufficient winning chances for White unless Fritz has a “bug” in its endgame programming. To date, given its extraordinary search depths and Nalimov 5-piece endgame tablebases, the likelihood of making a fatal mistake is remote. Nonetheless, please allow me to share with you an interesting and educational Rook endgame. After 83 Kf3, the best line for both sides is 83…Rd3+ 84 Ke4, Rxb3 85 Rf6+, Kb7 86 Kd5, Rxb2 87 Kxc5, a5 88 Rb6+, Ka7 89 Rb5, Ka6 90 Rb8, Ra2 91 Kc6, Ka7 92 Rb7+, Ka6 93 Rb6+, Ka7 94 Rb7+ draws. The most interesting line occurs when Black's King avoids being trapped on the “short side.” This is a winning attempt for Black, which almost allows White to win. Black instead plays 88…Kc8 89 Ra6, Ra2 90 Kc6 (White threatens 91 Ra8 checkmate), Kd8 (Black responds by threatening 91…b3) 91 Rb6 (White can probably bail out with a draw after 91 Kd6, Ke8, etc.), Ra3 (91…Rb2 92 Ra6, Ra2 repeats the position while 92 Kb5, Kc7 93 Kxa5, Rxa2+ 94 Kb5, b3 95 c5, Rc2 96 Kb4 also draws). White now has two options: 1) drive Black's King away before advancing the passed c-pawn, and 2) immediately advancing the passed c-pawn.
Black actually had a much simpler draw in the last line with 92…Kc8 93 Ra6, Kb8 94 Kb6, Rc3 95 Rxa5, b3 96 Rb5, Kc8 97 c6, b2 (97…Rg3?? 98 Ra5, Kd8 99 c7 wins) 98 Rxb2, Rc1. Black draws this because his Rook remains behind White's passed pawn while his King occupies the Queening square. Isn't it ironic that Black must give up his pawns in order to draw, after White abandoned his b-pawns in a winning attempt? 83. …, Rd2 While it's clear to informed humans that the game is drawn, the machine doesn't know that. To my disappointment, Fritz 6a raised its evaluation in its favor to 0.63 pawns after calculating for 20 hours and 20 minutes. During that time it reached a search depth of 23/46 ply, at an average speed of 373 kilonodes/second. Franklin Campbell wrote, “I think the computer will continue to calculate a similar advantage for itself till the bitter end. After all, it won't conclude that it isn't making progress. It will continue to use the same formula and come up with similar results till we declare the draw.” Franklin is correct, so we will declare a draw after a few more moves. Still, I can't shake the disappointment that such a strong chess engine can't correctly evaluate simple endgame positions. For the record, Black's other option also draws. After 83…a4 84 Rf8, Rd3+ 85 Rf3, Rd2 86 Rf2, Rd1 87 Rf6+, Kd7 88 Rf7, Ke6 89 Ra7, Rd3+ 90 Kf4, Rxb3 91 Ra6+ (a zwischenzug, pushing Black's King into a more passive position), Kd7 92 Rxa5, Rxb1 93 Rxc5, Rc2 94 Rb5, Rxc4+ 95 Ke3, Kd6 96 Kd3, Rh4 97 Kc2, Kc6 98 Rb8 draws. 84. Rf2, Rd6 Very disappointing. Fritz can not make any progress yet evaluates the position as 0.56 pawns in its favor. It reached this assessment after calculating for 24 hours and 15 minutes to a depth of 25/25 ply, at an average speed of 381 kilonodes/second. Black had a couple other options. I expected 84…Rd1 85 Rf3, Rg1+ 86 Kh3 (prevents 86…Rg2), Rg8 87 Rf6+, Kb7 88 Rf7+, Kb8 89 Rf6, Rd8 90 Rxa6, Rd3+ 91 Kg4, Rxb3 92 Rc6, Rxb2 93 Rxc5, b3 94 Rb5+, Kc7 95 Kf4, with a draw. Black's other option was analyzed by Helmut Froeyman of Belgium. He considered 84…Rxf2 85 Kxf2 Kd6 86 Kf3, Ke5 87 Ke3, a5 88 Kd3, Kf4 89 Kd2, Ke4 90 Ke2, Kd4 91 Kd2, a4 92 bxa4, Kc4 93 b3+??, Kxb3 94 a5, Kb2 95 a6 (95 Kd3, b3 96 a6, Kc1 97 a7, b2 98 a8/Q, b1/Q+ wins), c4 96 a7, c3+ 97 Kd3, c2 98 a8/Q, c1/Q when it seems that Black wins. To his credit, Helmut later discovered that White need not offer up his pawns with a loss of tempo. It was my plan to play instead 93 Ke3, Kb3 (93…b3 94 a5, Kb5 95 Kd3, Kxa5 96 Kc3 draws, as does 93…Kd5 94 a5, Kc6 95 Ke4, Kb5 96 Kd5.) 94 a5, c4 95 a6, c3 97 bxc3, bxc3 98 a7, c2 99 a8/Q, c1/Q+ 100 Kd4 which seems to draw. It is my guess that Fritz 6a's search, aided by its 5-piece Nalimov endgame tablebase, saw only draws in this line and therefore repeated the position to try something new on its next turn. 85. Rf3, … I may as well make things as clear to Fritz 6a as I can by repeating the position reached after the 83rd move. 85. …, Rd2 Fritz 6a is clearly confused. It just repeated the position after it's 83rd move.
86. Rf2, … I've just repeated my 84th move. If the chess engine replies with 86…Rd6, then we have a draw by three-fold repetition. This position will have occurred after Black's 82nd move, after Black's 84th move, and after 86…Rd6. 86. …, Rd6 Well, there we have it. Fritz 6a calculated for 18 hours to a depth of 25/25 ply, at an average speed of 377 kilonodes/second. In spite of the three-fold repetition, it still favors its position by 0.53 pawns. I suppose that if this were an OTB game, I would summon the arbiter to report the draw. Possibly I'd have to play 87. Rf3 to prove it. Regardless, The TD Ralph Marconi and Franklin Campbell and I have agreed that the game is a draw at this point. Franklin wrote, “The difference in precise evaluations of the same position is probably due to differences in times used. I just let the computer run as long as is convenient for me [and was] not based on some theoretic reasoning.” ½ - ½ At this point I'd like to thank Franklin Campbell for many reasons. I'm a difficult person to work with, so Franklin had to show a great deal of patience in order to deal with my stubbornness and strong opinions. Franklin also dedicated enormous amounts of time daily on his personal computer so that this 4-game match could even happen. This match also cost him considerable personal time due to hosting this event at his web site and for the almost daily updates caused by having to post moves and commentary. I think we readers all owe Franklin a large amount of gratitude for his daily sacrifices on our behalf over the past 1.5 years. Now he gets a much needed and well-deserved rest. I'd also like to thank Ralph Marconi for his contribution as the official arbiter. Finally, I am extremely grateful to Don Maddox of ChessBase USA. He generously donated Fritz 6a and Nimzo 7.32 to our cause, which made the whole match possible in the first place. In gratitude, I will support ChessBase USA with my future purchases. Well readers, I hope you enjoyed this match and benefited from the experiment. What conclusions can be drawn for the data? Well, it is clear now that top rated chess programs teamed with fast Pentium III computers can play correspondence chess at a very high level when allowed very long reflection times for all of their moves. While they do not yet show evidence of actually outplaying a strong master, they are capable of winning if that master unbalances the position too much in order to try to win. A clear example is Nimzo 7.32-Ham. Fritz-Ham also presented the chess engine with a winning endgame, but the win was beyond its capacity to execute. Still, I was not able to defeat the chess engines, although I nearly had an early tactical win in Ham-Fritz 6a and a long technical win in Ham-Nimzo 7.32. In both cases, human error caused me to err at the decisive moment, allowing the chess engines to draw. Had I chosen to play using anti-computer strategies, I think I would have scored much better than I did, since the chess engines have some clear weaknesses. However, that would have defeated the point of this experiment and little would be learned from that. Does this mean that masters, unaided by computers, now need to worry about meeting these mechanical monsters in correspondence events? Can an uninformed computer operator rely 100% upon using a chess engine to defeat strong human opposition? I don't think chess engines alone will have a decisive impact for the next few years. In my match, these chess engines were operating nearly 24 hours/day on each game. However, most ICCF events require one to compete against multiple opponents. So no single chess engine would be able to dedicate the amount of time per game that my opponents did against me. However, I do think that strong chess-playing humans, having access to such powerful assistance, have a great edge over unaided humans, especially in tactically sharp positions. I'm convinced of this and therefore plan to buy a computer soon, using Fritz 6a and Nimzo 7.32 for assistance. During this match, I lived in constant fear of missing tactical shots that a computer would find. My fear came true in Nimzo 7.32-Ham. Therefore, using these chess engines as blunder-checkers should give me significant peace of mind in future ICCF games. Now, a final thank you is owed to you readers, many of whom followed these games from beginning to end. Thank you for your interest and support. Last Updated: 2001.04.05 |
Free counters provided by Honesty.com.
| Home | CC Challenge Page | All Four Diagrams | Java Replay |