< Earlier Kibitzing · PAGE 3 OF 3 ·
|Jan-19-15|| ||AylerKupp: <<cro777> The move 13...Nf4 with the idea to sacrifice this knight for the attack (instead of the positional approach 13...Nf6) is, in effect, a computer move. ... GM Baburin: "My Houdini likes the idea: 13...Nf4 14.Nf3 Nxh3 (13...Nxg2 14.Kxg2 a5) 15.gxh3 c5 (15...Bf6)">|
Well, maybe GM Baburin's Houdini likes the idea of 13...Nf4 but my Houdini 4 doesn't, or at least not as much as either 13...Nf6 [+0.10] or 13...c5 [+0.18] at d=27 after about 22 minutes of analysis (it ranks 13...Nf4 as its 3rd best move, evaluating it at [+0.32]). Perhaps if I had let it run longer it would have evaluated and ranked 13...Nf4 higher. I don't know how long GM Baburin let his Houdini run or what version of Houdini he was using.
On the other hand, Stockfish 5 evaluated 13...Nf4 as its top move as early as d=25 after less than 2 minutes of analysis, but it alternated between 13...Nf4 and 13...Nf6 as its suggested top move until I stopped the analysis after about 30 minutes at d=33. At that time it was evaluating 13...Nf4 as its top move at [-0.34], with 13...c5 as second best at [+0.19] and 13...Nf6 as third best at [+0.25]. So it is possible that Stockfish would change its move rankings again if I let it run longer, but I really have no basis for thinking that.
Finally, Komodo 8 did not evaluate 13...Nf4 as Black's top move (at [0.00] until d=24 and more than 30 minutes of analysis, with 13...Nf6 as #2 [+0.07] and 13...c5 as #3 at [+0.25]. Who knows how it would have evaluated 13...Nf4 if I had let Komodo run longer?
Therefore, in my usual style, this is how I would summarize the engines' assessment of their top 3 moves, sorted by descending Ratings Adjusted Average:
Black's Houdini Komodo Stockfish
Move d=27 d=24 d=33 <St.Avg> <RAdj.Avg>
13...Nf4 [+0.32] [+0.07] [-0.34] <[+0.02]> <[+0.01]>
13...Nf6 [+0.10] [0.00] [+0.25] <[+0.12]> <[+0.11]>
13...c5 [+0.18] [+0.25] [+0.19] <[+0.21]> <[+0.21]>
And here is a summary of how the 3 engines ranked the moves, without regard for the value of the evaluation.
Black's Houdini Komodo Stockfish
Move d=27 d=24 d=33 <Avg>
13...Nf6 1 1 3 <1.7>
13...Nf4 3 2 1 <2.0>
13...c5 2 3 2 <2.3>
The position after 13.Rxe5 is a complicated one and, if basing Black's best response on computer analysis, highly dependent on which engine they were using and how long they let it analyze. Maybe Aronian was a little too harsh in firing his seconds after they failed to find 13...Nf4. And maybe he is searching for replacements. Would you give me a recommendation? :-)
|Jan-19-15|| ||Eyal: From looking a bit at this 13...Nf4 line with my own engine shortly after the Anand-Aronian game, when I first saw it mentioned, I got the impression that engines tend to have serious horizon-effect problems in evaluating its true strength Ė probably the reason why Aronianís seconds missed it (not the move Nf4 itself, that is, but the knight sac + rook lift idea which follows). In several lines, when you start sliding forward in the analysis (which isnít difficult, since the lines tend to be very forcing), you start seeing the evaluations improving for Black. Thereís a small sample of that in some of the analysis that was posted here Ė Anand vs Aronian, 2014 Ė following my initial post. In any case, from a human-practical point of view, the most important thing is not how many centipawns exactly separate the evaluation of 13...Nf4 from 13...Nf6 or 13...c5, but rather realizing the many problems that the sharp and forcing sequence leading to 15...a5 can pose for White, especially if he isnít prepared in advance. |
And btw, Iím not sure if this whole thing about Aronian "firing his seconds" is even serious, or whether it was said jokingly by Aronian to So and/or by So to Seirawan. But at any rate, Aronian certainly had good reasons to feel hugely frustrated about the opportunity that he missed in such an important game, considering he was the one who introduced 11...Qd7 as a novelty.
|Jan-19-15|| ||cow: Wow <Thorsson> back at the site after a 6 year absence! Welcome back.|
|Jan-19-15|| ||tamar: Komodo 5 64 bit (on Quad Core) is similarly recalcitrant about accepting 14...Nxg2 as a magic bullet. |
Even when I enter the moves 14...Nxg2 15 Kxg2 a5 manually, 16 minutes later it is still showing a slight advantage for White with 16 a4
Granted this is only at 26 depth, so there may be horizon issues, but it is easy to see how Aronian's seconds could miss that there was buried treasure.
The brilliant nature of the move is that the subsequent moves are not immediately decisive, but have created highways for all Black's pieces to become involved, so there are several waves to be calculated.
|Jan-19-15|| ||tamar: Letting the engine run longer, at 28 ply depth, Komodo begins to see a substantial edge for Black, and after 16 a4 Ra6 basically gives the continuation N Guliyev vs J Gustafsson, 2014 which ended in a draw after Gustafsson had a lapse and missed a zwischenzug keeping a large edge.|
|Jan-19-15|| ||AylerKupp: <tamar> I think that knowing how long to run an engine analysis is one of the, if not THE, critical questions in computer assisted chess. Run it longer than needed and you waste time. Not run it as long as needed and you will get a misevaluation of the position. In the recently concluded The World vs Naiditsch, 2014 we had a similar situation in that all engines except, interestingly, Komodo, evaluated one particular line near [0.00] but if the analyses were run longer than we usually ran them they would eventually start to show an advantage for White. We referred to this as the "Hocus Pocus" variation because Black's position seemed to deteriorate as if my magic. So it seems to me that the position after 13.Rxe5 might be a similar type of position.|
I have thought that looking at the evaluation trends might provide a clue. If multiple engines agree as to what the top move is (and provide similar continuations) or if the difference between the evaluation of the top line and the others is increasing, then the position is stable and the evaluations probably correct and the analysis can be stopped. But if multiple engines are showing different evaluations and each engine's top 3 moves are changing from ply to ply, then the position is unstable and the analyses need to be run longer until a stable position is reached.
Of course, a stable position might never be reached or the time to advance one ply might become impractically long. That would also be an indication that the analysis must be stopped and that forward sliding is necessary.
|Jan-19-15|| ||tamar: <AylerKupp: <tamar> I think that knowing how long to run an engine analysis is one of the, if not THE, critical questions in computer assisted chess.>|
I agree. For the most part, the eval you get in the first five minutes is going to stay the same, but there are positions, like this one after 13...Nf4, where the engine will give you no clue for well past that time that there is anything other than drawing chances with 14 Nd2-f3 Nxg2
How to identify those positions is really difficult. Here someone (Cheparinov?) went on the original voyage of discovery, and found that the eval changes when you slide forward with the moves ...a5 and ...Ra6
I would be really interested if Stockfish and Houdini register 14...Nxg2 immediately as a good move. That, as you suggested, would give a clue that the position was different than most, and would at least mark the position as worth looking at longer.
I noticed just analyzing with Komodo that it settled on 16 a4 very quickly and held to that opinion up to 28 plys with evals near 0.00 plus or minus .10, but then jumped to -.55 and rose. Was it because it took that long to see the effect of the last piece-the rook on f8- to become involved?
Or was it that 14...Nxg2 was only one of some 50 some moves it had to search?
That would be my two cents. If the engines disagree, and there are huge numbers of possible moves, most likely you should increase the time spent before taking the eval as worthwhile.
|Jan-22-15|| ||gokusano: Ivanchuk was an unsuspecting victim. The Nxg2 was hardly a move to cause trouble at first glance because the follow up a5 is really difficult to find. Even the engine could not evaluate it as a pressure packed move.|
|Jan-23-15|| ||Hand Of King: A devastating lose from the great Ivanchuk. I felt so bad after watching the end of the game. He is still my favorite player https://www.youtube.com/watch?v=jLe...|
|Jan-23-15|| ||kingscrusher: I video annotated this game here : https://www.youtube.com/watch?v=R4t...|
|Jan-25-15|| ||chancho: Poor Ivanchuk.
Caught like a trout in the opening.
Seeing him with that pained look... like someone whose prostate was heavily probed, was sad and yet comical at the same time.
|Apr-03-15|| ||wordfunph: "This is one example where the important g2-pawn is better than a knight."|
- Wesley So (after 14...Nxg2!)
Source: Chess Life 2015 April
|Sep-05-15|| ||offramp: I'm afraid that's not the greatest pun ever. It's a bit weak.|
|Sep-05-15|| ||Steve.Patzer: Awesome game!|
|Sep-05-15|| ||kevin86: A great game! So what?!|
|Sep-05-15|| ||thegoodanarchist: < offramp: I'm afraid that's not the greatest pun ever. It's a bit weak.>|
So be it.
|Sep-05-15|| ||offramp: <offramp: I'm afraid that's not the greatest pun ever. It's a bit weak.>|
I wrote that harsh criticism at 5:20am. I was working early on Saturday.
Now that I have had more than 20 minutes to mogitate upon that pun, I realise its superb merits.
In fact I think it might <be> the <greatest pun ever>!!
|Sep-05-15|| ||morfishine: FWIW: <offramp> & <thegoodanarchist> The game is sub-par and the pun sucks...Enough with the "So" puns...its nauseating around here|
|Sep-05-15|| ||thegoodanarchist: <morfishine: FWIW: <offramp> & <thegoodanarchist> The game is sub-par and the pun sucks...Enough with the "So" puns...its nauseating around here>|
So sue me! Sorry, couldn't resist one more. If you don't like So puns, maybe you will like this instead:
F Vallejo Pons vs Caruana, 2009
|Sep-05-15|| ||ajile: Note if 27.Be3
click for larger view
simply 27..Qxe3! since 28.fxe3 Rd2+ is fatal.
click for larger view
|Sep-06-15|| ||The Kings Domain: :-) Nice pun. Deep game, the kind of grandmaster chess one loves to rack his brain over. :-)|
|Sep-06-15|| ||ajile: https://www.youtube.com/watch?v=i_k...
|Sep-06-15|| ||morfishine: <thegoodanarchist> Sorry, keep up the good fight against lousing punning, just try not to contribute! |
<ajile> Very nice continuation that shows the base weaknesses in White's position
|Dec-20-17|| ||rainingpieces: 14...Nxg2 is the main line in Bologan's book, "Bologan's Ruy Lopez for Black"|
|Dec-21-17|| ||Clement Fraud: The result of this game is a far cry from this one Ivanchuk vs Adams, 2002 |
I must congratulate Wesley So, as it cannot be easy to keep up with all the latest theory of the ninety-nine year old Marshall Attack. I think the lessons learned here suggest that if white doesn't accept black's gambit pawn on e5 (like in Ivanchuk vs Adams, 2002), then black's idea of forcing e7-e5 is possibly refuted?!
< Earlier Kibitzing · PAGE 3 OF 3 ·