From this reduction we can conclude that if White can do this on move <11>, then Black has seriously goofed-up his move-order, or at the very least, left out or missed a move in his development (thus bringing on his ruin)

All in all, a nice combination by White

That is fine <AL wazir> except that 16 move should be 16. Qg7

If you have had good sense to take me off ignore, you would be able to read this>

After 15…♗f6, 16.♕g7 is illegal. After 15…♗f8, 16.♕g7 appears to be refuted by 16.♗xg7.

click for larger view

Analysis by Houdini 4: d 25 dpa done

1. = (0.21): 10...Bf8 11.Ne3 Nc6 12.0-0 g6 13.f4 Be7 14.b3 0-0 15.Qe2 Nd4 16.Qf2 Nc6 17.Qf3 Be6 18.Bb2 d4 19.Ng4 Bd5 20.Be4 Bxe4 21.Qxe4 Nb4 22.a3 Qd5 23.Qe2 Nc6 24.Nf2 Rad8 25.Ne4 d3

2. + / = (0.29): 10...Bxf5 11.Bxf5 Nc6 12.e6 g6 13.Bh3 Qd6 14.exf7+ Kxf7 15.0-0 Rhe8 16.g3 Bf6 17.Bf4 Be5 18.Bh6 Bg7 19.Be3 Bd4 20.Bd2 Bf6 21.c3 Rad8 22.Bh6 Ne5 23.Re1 c4 24.Bf1 Kg8 25.Be3 Kg7 26.Be2 Re7 27.Qc2 Rde8

3. + / = (0.57): 10...g6 11.Nxe7 Qxe7 12.0-0 0-0 13.Bh6 Re8 14.Qd2 Bd7 15.Rfe1 Bc6 16.Bg5 Qf8 17.c3 Nd7 18.Bh6 Qe7 19.f4 c4 20.Bc2 Qd8 21.b3 Qb6+ 22.Qe3 Qxe3+ 23.Rxe3 Nc5 24.Rd1 a5 25.Kf2 Ne6 26.f5

After

15. Bg6 Qf8
16. Bxf8 hxg6

Instead of 17. Qxg6+ Kxf8, White also has 17. Qxh8+ Kxh8 18. Bxe7, when he emerges ahead by a clear exchange.

click for larger view

Gelfand played 11...Kxg7 exposing himself to a murderous and irresistible attack after 12.Qh5 and the mate threat on h7. But Gelfand does <not> have to take the knight, which he might have figured out if he hadn't likely been surprised by 11.Nxg7. If instead 11...c4 ! (I think that I'm entitled to give this move at least one "!") since after going over the game I saw that if the Bd3 retreats, then the threat to h7 disappears and after 12...Kxg7 Black is simply up a piece with no attack for White. And if 12.Qh5 cxd3 Black is again up a piece with no good attacking prospects for White.

According to Stockfish 9 at d=34 White can keep the game even with 12.Nf5 but everything else apparently loses. These are Stockfish 9's top 5 moves after 11...c4:

1. [-0.23]: 12.Nf5 Kh8 13.Nxe7 Qxe7 14.Be2 Qxe5 15.f4 Qd6 16.Kf2 Nc6 17.Re1 Be6 18.c3 d4 19.Bf3 Rfd8 20.b3 Rac8 21.bxc4 Bxc4 22.Bb2 Qxf4 23.Kg1 Ne5 24.cxd4 Nxf3+ 25.Qxf3 Qxf3 26.gxf3 Bd5 27.Kf2 Kg7 28.Rec1 b6 29.Rxc8 Rxc8 30.Rc1 Rxc1 31.Bxc1

click for larger view

The BOC give White good drawing chances even though Black will be a pawn up after 31...Bxa2. Restarting the analysis from this position Stockfish 9 evaluates the resulting position at [-1.72], d=47 after 31...Bxa2 but it drops to [-1.40] after d=48 and goes no higher for Black through d=57.

The other 4 lines reflect the fact that White remains a piece down without adequate compensation for a likely loss.

2. [-3.02]: 12.Bxh7+ Kxg7 13.e6 Bg5 14.Qg4 Bxe6 15.Qxg5+ Qxg5 16.Bxg5 Kxh7 17.Rd1 Nc6 18.Rg1 Kg6 19.Be3 Rh8 20.h3 Rae8 21.c3 Ne5 22.Kf1 Nd3 23.Rh1 f6 24.Rd2 a5 25.Re2 b5 26.Bd4 Bf5 27.f3 Rxe2 28.Kxe2 Re8+ 29.Kd2 Nxb2 30.Re1 Rxe1 31.Kxe1 Nd3+ 32.Kd2 b4

3. [-3.22]: 12.Nh5 cxd3 13.Qf3 Bh4 14.cxd3 Nc6 15.Kf1 Nxe5 16.Qf4 Qe7 17.Bd2 Ng6 18.Qh6 Qe5 19.Bc3 d4 20.Re1 Bg5 21.Qxf8+ Kxf8 22.Rxe5 Nxe5 23.Bxd4 Nxd3 24.h4 Be7 25.Bg7+ Kg8 26.Bc3 Bf5 27.g4 Bxg4 28.Nf6+ Bxf6 29.Bxf6 Re8 30.Rg1 Re1+ 31.Kg2 Re2 32.Bd4

4. [-3.67]: 12.e6 cxd3 13.Nf5 Kh8 14.Qxd3 Bxe6 15.Nxe7 Qxe7 16.Qa3 Qd8 17.f3 Nc6 18.Bf4 Rc8 19.Rd1 Qf6 20.Bg3 Nd4 21.Rd2 Rfe8 22.Kf2 Nxc2 23.Qxa7 d4 24.Rhd1 Bf5 25.Kg1 d3 26.Qxb7 Re7 27.Qb3 Nd4 28.Qb4 Ne2+ 29.Kh1 Nxg3+ 30.hxg3 Re2 31.Qf4 Kg8

5. [-3.94]: 12.Bf5 Kxg7 13.Qg4+ Kh8 14.Bxc8 Qxc8 15.Qxc8 Rxc8 16.f4 Nc6 17.Bd2 Bc5 18.Rf1 Rg8 19.g3 Rge8 20.Kd1 Kg7 21.g4 d4 22.Rf3 Rad8 23.Be1 a5 24.Rb1 b6 25.a4 Be7 26.g5 Bc5 27.Ra1 d3 28.cxd3 cxd3 29.Rc1 Ne7 30.Bf2

You are right - there is a reason <AL wazir> put me a n ignore.

Reason is that I was being a jerk.

I can forgive Gelfand for not playing 11. ... c4 for a number of reasons. First, it is tough to decline the knight. Second, by declining the knight, Black not only drops a pawn, but he concedes his king side as lost, which given Black's position, would make for a very difficult game. Third, White must play precisely to win.

I found the first three moves of the game line pretty easily. I eventually found 14. e6! followed by the rest of the game line. I looked at 11. ... c4 as being a decent counter move, but Black faces an ugly, defensive game with White having a slight but permanent positional advantage.

Well, I think that I have to take back my conclusion in V Gashimov vs Gelfand, 2009 (kibitz #31). Those of you who (unfortunately) are familiar with my analyses posts know that I think that to get the best evaluation of a position it is essential to conduct the analysis using multiple engines since each engine uses different evaluation functions and search tree pruning heuristics so they could come up with different evaluations and different move rankings. If different engines of comparable playing strength (i.e. ratings) come up with different evaluations and different "best" moves, how do you know which one is correct? Or at least closest to the true evaluation of the position and ranking of possible "best" moves.

So, in my rush to validate my impression of 11...c4 I and get some results quickly, I did the analysis using only Stockfish 9 using 4 threads and a hash table size = 1024 MB. Then, possibly to my regret, I decided to go back and validate the analysis using Houdini 6 and Komodo 11.2 in addition to Stockfish 9, what I refer to as "Team HKS". Because I have an archaic 32-bit computer with 4 GB of memory, I can only allow each engine to use 1 thread and a hash table size = 256 MB. These smaller values are needed to prevent excessive disk swapping when running all 3 engines simultaneously. Yes, those analyses each take approximately twice as long but since I can then run them simultaneously I can let them run overnight and all I have to do is consolidate the results in the morning.

The results surprised me. Both Houdini and Komodo also evaluated White's best move to be 12.Nf5 with evaluations of [0.00] and [-0.17] respectively, not too different than Stockfish's initial evaluation of [-0.23], and indicating a position with approximately equal chances for both sides. But this second Stockfish analysis evaluated 12.Nf5 at [+1.05], indicating that White had a significant advantage.

But still not too surprising. The initial Stockfish analysis was run using 4 cores and threads and multi-core chess engines are notoriously non-deterministic, and so it was almost certain that analyses performed of the same position, on the same computer, using the same engine, and to the same depth will give different results on subsequent runs (if you don't believe me, try it yourself). And the second analysis was run to a greater depth, d=44 compared with the first analysis' d=34, and a lot can change in 10 plies. Besides, using only 1 thread eliminates the engines' non-determinism, so the result of the second analysis seemed more credible. But in the first Stockfish analysis all the evaluations of 12.Nf5 were negative indicating an advantage for Black and in the second Stockfish analysis all the evaluations of 12.Nf5 were positive indicating an advantage for White. Now, <that> didn't make any sense.

As I was going to run yet another analysis (see the definition of insanity apparently erroneously attributed to Einstein) I realized that I needed to explicitly specify that White's 0-0 and 0-0-0 options are still open. I use Arena 3.5 as my GUI and its default is to specify that neither side has the option to castle. So I suspect that I had failed to do so in Stockfish Analysis #1 (4 threads, 1024 MB hash table) and the Houdini and Komodo analyses (1 thread, 256 MB hash table). So I re-ran the Team HKS analyses overnight after making sure to indicate that White had the options of 0-0 and 0-0-0 and below is a summary of the results. The <Avg> column lists the unweighted average of the 3 engine evaluations and the <RWAvg> lists the ratings weighted average of the 3 engines evaluations based on the latest (5-13-18_ CEGT 40/20 64-bit, single engine ratings, the closest I could find to a 32-bit single engine ratings. After all, it makes sense to give more value to the engine with the highest rating and less value to the engine with the lowest rating. The <True Rank> column lists the engine rankings with no more than a [0.50] centipawn difference of each other, the type of move ranking that a human player might do when evaluating positions.

White's Houdini 6 Komodo 11.2 Stockfish 9

Move d=36 d=33 d=44 <Avg> <RWAvg> <TrueRank>

---------- ---------- ----------- ---------- ---------- ----------- -----------

12.Nf5 [+0.92] [+0.66] [+1.05] <[+0.88]> <[+0.89]> <1>

12.Bxh7+ [-1.96] [-1.68] [-1.96] <[-1.87]> <[-1.87]> <2>

12.Nh5 [-1.90] [-1.76] [-2.26] <[-1.97]> <[-1.99]> <2>

12.c3 [-2.21] [-2.13] [-2.24] <[-2.19]> <[-2.20]> <2>

12.Bh6 [-2.17] [-1.98] [-2.58] <[-2.24]> <[-2.26]> <2>

And these are the PV=1 lines for the 3 engines after 11.Nxg7 11...c4:

<Houdini 6>: [+0.92], d=36 12.Nf5 Kh8 13.Nxe7 Qxe7 14.Be2 Qxe5 15.0-0 Nc6 16.Re1 Re8 17.Bd2 Qg7 18.Bxc4 Bg4 19.f3 Qd4+ 20.Kh1 Rxe1+ 21.Qxe1 Qxc4 22.fxg4 Qxc2 23.Bc3+ d4 24.Bxd4+ Nxd4 25.Qe5+ Kg8 26.Qxd4 Re8 27.Rf1 Qe2 28.Qf6 Re7 29.h3 Qe5 30.Qxe5 Rxe5 31.Rf2 Kg7 32.Kh2 Kg6 33.Rd2 Ra5 34.b3 f5 35.Kg3 Ra6 36.Kf3 fxg4+ 37.Kxg4 Rf6 38.Rd7:

<Komodo 11.2>: [+0.66], d=33: 12.Nf5 Kh8 13.Nxe7 Qxe7 14.Be2 Qxe5 15.0-0 Nc6 16.Re1 Qf5 17.b3 Qf6 18.Rb1 Rd8 19.h4 Qxh4 20.bxc4 dxc4 21.Bb2+ f6 22.g3 Rxd1 23.gxh4 Rxe1+ 24.Rxe1 c3 25.Bxc3 Kg7 26.Bd3 Bd7 27.Kh2 Rc8 28.Kg3 Re8 29.Be4 b6 30.h5 Ne5 31.Bd4 Nc6 32.Bb2 Ne5 33.Rd1 Be6 34.f4 Ng4 35.Rd6 Bxa2

<Stockfish 9>: [+1.05], d=44: 12.Nf5 Kh8 13.Nxe7 Qxe7 14.Be2 Nc6 15.0-0 Qxe5 16.Re1 Re8 17.Bd2 Qg7 18.Bxc4 Bg4 19.f3 Qd4+ 20.Kh1 Rxe1+ 21.Qxe1 Qxc4 22.fxg4 Qxc2 23.Bc3+ d4 24.Bxd4+ Nxd4 25.Qe5+ Kg8 26.Qxd4 Re8 27.Rf1 Qe2 28.Kg1 Qe5 29.Qxe5 Rxe5 30.Kf2 Rb5 31.b3 Ra5 32.a4 Rc5 33.Re1 Kf8 34.Re3 Rc2+ 35.Kg3 f6 36.Rd3 Ke7 37.Kf3 Ke6 38.h4 Ra2 39.g3 Rc2

So all was now relatively well in the world. All the analyses of 11...c4 12.Nf5 were consistently positive and all the analyses of 11...c4 12.(any White move) were consistently negative.

This is the position after 10...0-0.

click for larger view

And here is what Team HKS determined were White's best move 11. Because all 3 engines did not agree on White's top 5 moves, neither the unweighted nor ratings-weighted evaluation averages could be directly calculated. So I needed to fill in values for the missing evaluations and I did this by considering the missing evaluations to be the lowest evaluation of each engine less the average of the magnitude of the differences of each of the evaluations of that engine. I thought that this would be the most realistic value for the missing evaluations although it's still just guesswork. However, it mostly affects the lowest ranked moves so it probably doesn't have any adverse practical effect. The other columns have the same meaning as for the summary of 11...c4.

White's Houdini 6 Komodo 11.2 Stockfish 9

Move d=37 d=34 d=47 <Avg> <RWAvg> <TrueRank>

---------- ---------- ----------- ---------- ---------- ----------- -----------

11.Nxg7 [+1.00] [+0.86] [+1.15] <[+1.00]> <[+1.01]> <1>

11.Nxe7+ [-0.13] [+-0.26] [0.00] <[-0.07]> <[-0.05]> <2>

11.0-0 [-0.19] [-0.22] [0-00] <[-0.21]> <[-0.14]> <2>

11.c3 [-0.18] [-0.21] [-0.12] <[-0.17]> <[-0.17]> <2>

11.f4 [0.00] [-0.61] [-0.08] <[-0.23]> <[-0.21]> <2>

11.Qf3 [-0.49] [-0.24] [-0.60] <[-0.44]> <[-0.45]> <2>