You know you have hit the big time when I create this game collection.

Game Collection: WEI YI'S BEST GAMES

click for larger view

<Moszkowski012273> above questioned whether there were better moves for White than 12.Bxb5, and <kingscrusher> wanted to know the basis for that implied opinion and whether it was based on engine evaluation. Well, below are the evaluations of 4 engines to try to determine whether White had a better move than 12.Bxb5.

First let me say that this position is relatively rare at the master level. The 365Chess database only has 2 games with this position, this one and the earlier Bologan vs Shirov, 2005, won by Shirov. In the earlier game Bologan also played 12.Bxb5 and after 12...axb5 13.Ndxb5 Qb7 14.Qxd6 Shirov played 14...Be7 instead of 14...Bxd6 and eventually went on to win. So Shirov was certainly familiar with the position through 14.Qxd6, and it's even possible that 14...Bxd6 was the losing move. And, needless to say, countless games have been played with positions similar to the position after 11...Rb8 and the 12.Bxb5 and 12.N4xb5 have been played many times. Of course, the kind of success achieved by White depends on the actual position, but White will likely get 3 pawns for the sacrificed piece.

Back to the original question. Here is a summary of 4 engines' evaluations, sorted in order of descending Ratings Adjusted Average as a result of being White's move, whenever there's more than one engine with an evaluation for that move. The Ratings Adjusted Average is the weighted average of the evaluations using the engine ratings from the Dec-06-15 CCRL 40/40 engine tournament.

a. Moves not reported by the engines as one of their top 3 moves are considered to be null when calculating the averages.

White's Gull 3 Houdini 4 Komodo 9.2 Stockfish 6

Move d=25 d= 30 d= 29 d= 36 <St.Avg> <RAdj.Avg>

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

12.Ncxb5 [------] [+1.14] [+0.91] [------] <[+1.03]> <[+1.00]>

12.Ndxb5 [------] [+1.14] [+0.91] [------] <[+1.03]> <[+1.00]>

12.h3 [+0.29] [+0.50] [+0.53] [+0.28] <[+0.40]> <[+0.42]>

12.Kb1 [+0.27] [+0.31] [+0.53] [+0.28] <[+0.35]> <[+0.37]>

12.Rhe1 [+0.26] [+0.30] [+0.62] [+0.19] <[+0.34]> <[+0.37]>

12.Rhg1 [+0.25] [------] [------] [+0.27] <[+0.26]> <[+0.26]>

12.Bg5 [+0.35] [------] [------] [+0.20] <[+0.28]> <[+0.25]>

b. Moves not reported by the engines as one of their top 3 moves are considered to be evaluated one centipawn less than the move with the lowest evaluation when calculating the averages. This is the best evaluation that the move could receive from the engine and therefore the average evaluations will likely be optimistic since the engine's actual evaluation of the move could be much lower.

White's Gull 3 Houdini 4 Komodo 9.2 Stockfish 6

Move d=25 d= 30 d= 29 d= 36 <St.Avg> <RAdj.Avg>

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

12.Ncxb5 [+0.24] [+1.14] [+0.91] [+0.18] <[+0.62]> <[+0.66]>

12.Ndxb5 [+0.24] [+1.14] [+0.91] [+0.18] <[+0.62]> <[+0.66]>

12.h3 [+0.29] [+0.50] [+0.53] [+0.28] <[+0.40]> <[+0.42]>

12.Kb1 [+0.27] [+0.31] [+0.53] [+0.28] <[+0.35]> <[+0.37]>

12.Rhe1 [+0.26] [+0.30] [+0.62] [+0.19] <[+0.34]> <[+0.37]>

12.Rhg1 [+0.25] [+0.29] [+0.52] [+0.27] <[+0.33]> <[+0.36]>

click for larger view

And here is a summary of how the 4 engines ranked the moves, without regard for the value of the evaluation.

White's Gull 3 Houdini 4 Komodo 9.2 Stockfish 6

Move d=25 d= 30 d= 29 d= 36 <Avg> <TrueRank>

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

12.h3 2 3 4 1 <2.5> <1>

12.Kb1 3 4 4 1 <3.0> <2>

12.Ncxb5 6 1 1 6 <3.5> <3>

12.Ndxb5 6 1 1 6 <3.5> <3>

12.Bg5 1 6 6 4 <4.3> <4>

12.Rhe1 4 5 5 5 <4.8> <5>

12.Rhg1 5 6 6 3 <5.0> <5>

True Rank: 1 = [ 12.h3 ]; 2 = [ 12.Kb1 ]; 3 = [ 12.Ncb5, 12.Ndxb5 ] ; 4 = [ 12.Bg5 ]; 5 = [ 12.Rhe1, 12.Rhg1 ]. But this ranking is possibly misleading since 2 of the engines did not consider 12.Nxb5 (either knight, since the position will transpose after 12...axb5 and the capture of the Pb5 by the other knight) to be one of White's top 5 moves.

Note that 12.Bxb5 is not evaluated to be one of White's top 5 moves by any of the engines. And 2 of the engines, Houdini 4 and Komodo 9.2, evaluated 12.Nxb5 to be White's best move by a substantial margin. The other 2 engines, Gull 3 and Stockfish 6, did not consider either the knight or bishop sacs on b5 to be one of White's top 5 moves and preferred quieter continuations, although they did not achieve the high evaluations that Houdini and Stockfish achieved with 12.Nxb5. It would be interesting to force 12.Nxb5 into Gull and Stockfish to see how they evaluate it.

So at this point, according to these engines, <Moszkowski012273> is correct in wondering whether White has "plenty" of better moves than 12.Bxb5 (I'll let others decide whether "plenty" is appropriate here). But to reach a definitive conclusion we have to look at the positions after 12.Bxb5 and slide forward somewhat. If, as <kingscrusher> indicated, at least one engine shows a slight edge for White after 17.Rd4, then unless 16...Rxb7 is a better move than 16...Bxb7, then 14...Bxd6 is certainly a questionable move since all the other moves are pretty much forced.

click for larger view

In my previous posts (Wei Yi vs Shirov, 2013 (kibitz #54) and Wei Yi vs Shirov, 2013 (kibitz #55)) I showed that none of the 4 engines I used to analyze the position after 11...Rb8 considered 12.Bxb5 to be among their top 5 moves, and their evals ranged from [+1.14] to [+0.19] (more on this later). So how do these 4 engines evaluate the move actually played, 12.Bxb5?

Here is a summary of the same 4 engines' evaluations, sorted in order of ascending Ratings Adjusted Average as a result of being Black's move, whenever there's more than one engine with an evaluation for that move :

a. Moves not reported by the engines as one of their top 5 moves are considered to be null when calculating the averages.

White's Gull 3 Houdini 4 Komodo 9.2 Stockfish 6

Move d=25 d= 30 d= 30 d= 40 <St.Avg> <RAdj.Avg>

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

12...axb5 [-0.16] [0.00] [0.00] [+0.00] <[-0.04]> <[-0.02]>

12...Qb7 [+0.86] [+0.66] [+0.92] [+0.91] <[+0.84]> <[+0.85]>

12...e5 [+1.03] [+0.98] [+1.27] [+1.13] <[+1.10]> <[+1.13]>

12...Be7 [+1.43] [+1.81] [+1.53] [+1.76] <[+1.63]> <[+1.65]>

12...Bb7 [+1.43] [------] [+1.61] [+1.85] <[+1.63]> <[+1.67]>

12...Rb6 [------] [+1.81] [------] [------] <[+1.81]> <[+1.81]>

b. Moves not reported by the engines as one of their top 5 moves are considered to be evaluated one centipawn less than the move with the lowest evaluation when calculating the averages. This is the best evaluation that the move could receive from the engine and therefore the average evaluations will likely be optimistic since the engine's actual evaluation of the move could be much lower.

White's Gull 3 Houdini 4 Komodo 9.2 Stockfish 6

Move d=25 d= 30 d= 30 d= 40 <St.Avg> <RAdj.Avg>

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

12...axb5 [-0.16] [0.00] [0.00] [0.00] <[-0.04]> <[-0.02]>

12...Qb7 [+0.86] [+0.66] [+0.92] [+0.91] <[+0.84]> <[+0.85]>

12...e5 [+1.03] [+0.98] [+1.27] [+1.13] <[+1.10]> <[+1.13]>

12...Be7 [+1.43] [+1.81] [+1.53] [+1.76] <[+1.63]> <[+1.65]>

12...Bb7 [+1.43] [+1.82] [+1.61] [+1.85] <[+1.68]> <[+1.70]>

click for larger view

And here is a summary of how the 4 engines ranked the moves, without regard for the value of the evaluation.

White's Gull 3 Houdini 4 Komodo 9.2 Stockfish 6

Move d=25 d= 30 d= 30 d= 40 < Avg> <TrueRank>

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

12...axb5 1 1 1 1 <1.0> <1>

12...Qb7 2 2 2 2 <2.0> <2>

12...e5 3 3 3 3 <3.0> <3>

12...Be7 4 4 4 4 <4.0> <4>

12...Bb7 5 6 5 5 <5.3> <5>

12...Rb6 6 5 6 6 <5.8> <6>

True Rank: In the order listed.

A remarkable uniformity. All 4 engines agree on their top 4 moves and their ranking, and 3 out of the 4 engines agreed on their top 5 moves and their ranking. I don't recall ever seeing a uniform agreement such as this before.

And clearly 12...axb5 is Black's best move, with a large gap in the evaluations between this move (approximately even) and the other moves (White having a significant advantage to a winning advantage). And the gap in the evaluation of 12.Bxb5 compared to the evaluation of other White moves, while not great, seems significant, from [+0.35] to [+0.66] based on their Ratings Adjusted Averages when the best possible evaluation for those moves not listed by the engines to be among their top 5 moves are included in the averages. So, as far as these 4 engines are concerned and without additional forward sliding, there are "plenty" of better moves for White than 12.Bxb5.

So why did Yi play 12.Bxb5 if the evaluations were not that favorable for White? As I indicated earlier, this position has not appeared many times at the master level, and both of the games I found has Shirov playing Black. While Bogolan in Bologan vs Shirov, 2005 in 2005 might not have had access to top engines (engines have improved a lot from 2005 to 2013), I would think that Yi did, even in China. So I suspect that Yi had some "home cooking" ready against Shirov if Shirov duplicated his moves against Bogolan (a game which, after all, he won). Shirov, probably suspecting this, deviated from his game against Bogolan after 12...axb5 13.Ndxb5 Qb7 14.Qxd6 by playing 14...Bxd6 instead of the 14...Be7 he played against Bogolan (probably found OTB). And unless we ask Yi (and assuming that he's willing to discuss it), we'll never know what he had in mind had Shirov played 14...Be7. But this is all pure conjecture on my part.