Mig 
Greengard's ChessNinja.com

Kramnik Leads Amber

| Permalink | 38 comments

After four rounds of the Amber rapid/blindfold event in Nice, Vladimir Kramnik is leading with 5.5/8 in what is shaping up to be another balanced event. His usual dominance in the blindfold, now 3.5/4, is the difference. He is pursued by Morozevich and Aronian at a half-point distance and then Radjabov, Carlsen, and Topalov on 4.5. Anand still isn't looking good and is on an even score. There have been the usual blind blunders, dropped queens, etc, as well as a few nice games that wouldn't have looked out of place in the 2500 section of Aeroflot. (Those who think the blindfold games are exciting should watch more games between IMs and the under-2600 crowd in general. They aren't as worried about Elo, are more speculative, and also make more mistakes, therefore leading to more tactical action. This is also why some people say elite women's chess is more exciting than elite "men's" chess. It may well be, but it's not because they are women. It's just that fans don't pay attention to 2500-rated men but the women are stars at that level. High stakes + more mistakes = exciting games.)

Instead of having them play each other, I really would like to know what the average Elo decrease is and have them play blindfold against sighted players. One thing you notice after looking at a lot of these blindfold games is the tendency toward long sequences of direct threat-and-react moves, which makes sense. It's hard to focus on full-board dynamics and planning when you are concerned about not hanging a piece.

I've enjoyed quite a few of the rapid games, which are like potato chips after the heavy meals of Corus and Linares. They are many worth a look. Aronian-Leko saw White win with a very nice rook sacrifice on g6. The quiet move 26.Rd1 is cool stuff. Wang Yue has been horrible so far, with or without his eyes. Ubiquity has its drawbacks and he's got to be exhausted after so many brutal events in a row. Apart from the blunders, his loss with white to Morozevich was truly a rare work of domination by Moro. Aronian was busted right out of the opening against Kamsky but battled along into a bad knight endgame. Kamsky sealed the deal with the non-obvious 43.Ne4 when both Nxg4 and e6 allow Black excellent drawing chances with the a-pawn.

Rapid play is unkind toward endgames in general, of course. Leko missed a chance to beat Carlsen by waiting too long to play f5, which lost its sting after 47..Bc8. A miracle save by Carlsen, with ..Bxf5 on tap if Bxf7. Even with White's dominating king position he can't make progress. Both 44.f5 and 47.f5 look good. Topalov, fresh from playing the Zaitsev and Caro-Kann against Kamsky, essayed the Petroff against Karjakin and was duly punished by the vengeful jilted gods of the Najdorf. Karjakin had something to do with it, I suppose. Petroff, hah! 20.f5 was nice and 31.e6! even nicer. The queen check ladder to pick off the black pawns is also cool, and not just for show. The immediate 39.Rf7 is only a draw because the white queen doesn't control a2, so it's a perpetual. Great stuff. Topalov got some of his own back against Morozevich today. The unexpected 21.g5! led to a brutal direct kingside attack. If 22..Qxg5? 23.Qxf7+ Kh7 24.Re3 is curtains.

38 Comments

Mig said "One thing you notice after looking at a lot of these blindfold games is the tendency toward long sequences of direct threat-and-react moves"

And you see the defensive counter-effect of this, a tendency to guard and doubly guard one's pieces.

Very seldom does a piece just stand unguarded for long, which prevents it from accidentally hanging, I guess. I have the feeling that a player's pieces are more interlocked in patterns of guard and co-guard, as though the pieces are stickier.

I see less likelihood of placing a piece on an unguarded outpost somewhere, disconnected from the others. Like a group of people feeling their way through the dark, the pieces huddle together, lightly touching one another. No one wants to venture apart from the group, alone.

"a tendency to guard and doubly guard one's pieces."
The same strategy (pieces should protect each other) is actually recommended - at least for weaker players - in time-trouble situations ... when you see the pieces, but may still be 'blind' as far as tactical shots are concerned.

Thomas: :-)

Impressive blind play from Kramnik. And just as an impressive rapid play by Topa against Moro. Anand has a lot to worry about ;)

Surprize! Businessman Kramnik does well when "high stakes" (big $$$) are in play. Just like Linares: no appearance feees, so Poker Pro Grischuk wins, with Cuban Dominguez in last in the prize fund hunt.

Hi, Mig---I can give a partial answer to your question:

"Instead of having them play each other, I really would like to know what the average Elo decrease is [for Blindfold]..."

At http://www.cse.buffalo.edu/~regan/chess/fidelity/ControlGames/, folders Blindfold and Fast, I've posted up Guid-Bratko style runs (but with Rybka 3 to depth 13, not Crafty to depth 12) of the 1994, 2007, and 2008 Ambers. Files ending in .rept3 summarize the move-matching (MM) and average-difference (AD) measures for all players and games, throwing out moves 1-8 and positions with one side judged ahead by more than 3 pawns, while those with .reptMG3 tally only moves 21--40 ("middlegames"). If your browser doesn't show all of my long filenames, mouse-over will show the full name at the bottom of the window. Files beginning "perfs..." with AD/MM and/or MG tags rank the players in each category. The same system is used in other directories, so in http://www.cse.buffalo.edu/~regan/chess/fidelity/ControlGames/Tournaments/ you can compare results with the same players having full sight of the board---and full time controls!

The answers so far seem to be: () the dropoff is overall not much, though noticeable when you think of what you'd want for world-championship standards, and () Kramnik is a cool operator! Indeed, his 66.0% matches to Rybka in the 2007 Amber Rapid, and 0.059 pawns/move average difference (in the 2008 Amber Blindfold (beaten by Gelfand's 0.056!), ***rank in the top 5% of human performances all time***, at least among the 800+ performances in top matches and tournaments at standard controls I've compiled so far. You may have the germ of an explanation here in your followup statement:

"One thing you notice after looking at a lot of these blindfold games is the tendency toward long sequences of direct threat-and-react moves..."

My statistical model based on multi-PV runs at higher depths [which take a heck of a lot more computer time to gather data for] will be able to quantify this tendency, and hence give a much more reliable indicator of the "intrinsic Elo level" than the Guid-Bratko measures. By the way, I ran Kurnosov's other games, and got no-more/no-less than the same conclusion you gave, on the Moiseenko game in particular. I haven't had time yet to properly unveil all this stuff---in the past 24 hours I've reviewed 3 papers including one on quantum automata, discussed new research with 2 professors with another on tap tomorrow morning, and this weekend I'm actually playing, in a big open up here in Quebec!

Ah, the comma got included in the URL for the ControlGames, so you have to manually delete it (or click the Tournaments link which is fine, then delete the "/Tournaments" part to explore other stuff). Or click this: http://www.cse.buffalo.edu/~regan/chess/fidelity/ControlGames/

As the name implies, this is originally motivated to see how much computer move-matching (to Rybka 3 in particular) is "normal" for non-cheating humans. However, since I haven't yet moved on to other uses for my quad-core's spare (overnight) cycles, this is becoming virtually the entire history of chess seen thru the eyes of Rybka 3...at a depth where 2700 seems to be the best estimate of its skill.

Regarding the Rybka-matching blindfold moves, the concept seems fatally flawed to me, Ken. With BOTH players blindfolded I wouldn't expect them to play significantly worse by computer standards. They are generally being very cautious and are playing a different sort of game. In fact, it wouldn't shock me that if you discarded the egregious blind blunders (say, drops of 3 pawns or more), that blindfold would match Rybka just as high or higher than sighted rapid! That's a very different situation from having one player sighted and the other blindfold.

It's similar to doing computer analysis of only Berlin or Petroff games and concluding the players are fantastically accurate. The types of positions reached just don't allow for many things the computer counts as a mistake. When this sort of thing was done by chess-tyro researchers a few years ago they decided, iirc, Kramnik and Capablanca were the most accurate. Of course any chess player would assume that result, since they both avoided complex, chaotic positions as much as possible. Which means fewer tactics and fewer serious mistakes. Which means it's an apples and oranges comparison.

This isn't to say I have a better idea for objectively judging the intrinsic strength of a game of chess taken in isolation. Across a field or time span blunders are a reasonable measure. But when comparing a game between two blindfolded players it doesn't seem to work to me. I used to hold little blindfold matches in my classes with kids in Argentina, none stronger than 1400. With practice they could play entire games. Mostly they were very correct Giuoco Pianos and such because they stuck to the patterns they knew well. Of course super-GMs can play real chess blind, but just going over a few dozen of these games in a row definitely gives the impression they aren't quite the same thing.

Intriguing. You would need to measure in the chaoticness of a position. Can you do this by measuring how often and how much the evaluations vary when the search depth goes deeper?

Then you could compare comparable chaotic apples with apples, and oranges with oranges.

Kramnik after his 4th round blindfold game against Wang Yue: "[T]he quality of my blindfold games is better than the quality of my rapid games, strangely enough, so I was even thinking to ask Geurt if he can make me play blindfold both games, I think I would have a better result..."

Interesting endgame in Topalov vs Kramnik.

Karjakin Moro has to be seen to be believed. Moro sac'd an exchange right out of the opening. Some pretty wild stuff.

Anyone know if Kramnik & Topalov shook hands either before or after either of their draws?! (I'm pretty sure they haven't since their match, but it's been awhile...) Maybe Chessvibes will have a video that'll show what occurred?!

They didn't according to TWIC and random commentators on ICC.
Nasty birthday present from Vishy to Ivanchuk.
Carlsen has not won a single rapid game so far.

Bartleby, indeed that is the objective of my computationally-intensive statistical model. In its full form it does a linear combination over evaluations at different depths, and thus quantifies and responds to "swings" in evaluations. The simple form, with multi-PV data from one fixed depth, already gives a good fit to practice, and is able to distinguish "different sorts of game" as per Mig's response. For example, my preliminary runs indicate that Karpov made things considerably tougher on Kasparov in their 1984--87 matches than Fischer's opponents in 1970--72 did for him, thus supporting some of Garry's assertions in his MGP series. Per what Mig said about Berlins and Petroffs, even my simple model should pick up that that they are less challenging. Both gathering and processing the data take scads of time, however---hence my using the quicker Rybka runs to scope the landscape first.

I see Mig's point---indeed saw it while quoting his remark about threat-and-react sequences---but my point remains: Even *including* the egregious blind blunders, the average dropoff (AD) isn't terribly worse than a standard super-GM event. And the move-matching (MM) is often as high---which is my main message to those conditioned to see anything pushing 70% as a smoking gun. Even if your opponent goes out for a smoke, he may be seeing the board well enough in his head to do what these guys are doing.

Acirce, Kramnik's remark may have been prompted by his 3.5-0.5 blindfold score (now 4-1), but my quick-&-dirty numbers from 2007-08 give it some support:

Rank, MM%, AD, #Moves, Player, Event
-----------------------------------------------
2 56.9% 0.059 348 Kramnik, V AmberBlindfold2007R3d13
7 56.4% 0.081 326 Kramnik, V AmberBlindfold2008R3d13

1 61.7% 0.068 491 Kramnik, V AmberRapid2007R3d13
8 66.0% 0.097 379 Kramnik, V AmberRapid2008R3d13

The ranks are out of 36 from the 1994, 2007, and 2008 Ambers (3 x 12 players each). In 2007 and 2008, Kramnik's AD was better in blindfold than in rapid, though his MM was worse. Neither is really a true measure of "quality" by itself, and MM less so than AD, but MM is what people are up in arms about so I'm trying to demystify it.

I can add that I'm currently occupied with an issue that is touched on by another of Mig's comments above:

"Those who think the blindfold games are exciting should watch more games between IMs and the under-2600 crowd in general. They aren't as worried about Elo, are more speculative, and also make more mistakes, therefore leading to more tactical action."

More tactical action ==> more clear-cut choices ==> my simple model will predict more matching (as it does for Tal vs. Karpov at Montreal 1979, for which I've run full data). Thus I'm looking to see whether Open events tend to have more high-matching performances than the top-GM events I've started with.

Our good old "paper" champion never ceases to amuse me! Is he declaring he is the king of blindfold? Or more than that? Obviously your quality would look better when your overall opposition quality is low. Mr Kramnik, clarify your statement please!!

Kramnik is the king of Blindfold. Check his record in Amber. He was half joking when he said he would prefer to play both games Blind. Even You should be able to understand that. Again a pathetic comment from another guy who criticise everything Kramnik do and say. pathetic.

I am not Kramnik, and actually there should be no need to 'clarify' his statement anyway. But as PircAlert misunderstood it, here I quote Kramnik again [as quoted by acirce, but I watched the same video ,:)]:
"[T]he quality of my blindfold games is better than the quality of my rapid games".
That's just a relative statement, he is comparing himself with "the other Kramnik", not with anyone else. And he gave an interesting and plausible explanation: in blindfold you are forced to be fully concentrated all the time, one slip of mind can be immediately fatal.
BTW, the statement is not true concerning his games against Topalov: Both games were drawn, but he was worse, maybe lost with black in the blindfold game, better and maybe winning with white in the rapid - presumably color distribution is also relevant ... .

Oh Thomas, thanks for clearing that up! And to think that all this time I thought you were Kramnik!

Oh Thomas, I'm confused now. You are saying his statement is true, then you say his statement is not true. Are you saying his statement regarding "Kramnik vs other Kramnik" comparison was true when he made that statement but it isn't true anymore?? Then you just proved my point - as usual his statements have only comedy value!! :)

Actually, this particular statement was probably meant to have some comedy value (though, once again, his argument in support makes sense to me). But anyway, based on my post his statement may be _generally_ true [actually I didn't even make this point], but it is not _absolutely_ true under all circumstances ... . Color distribution and opponent also play a role - and games against Topalov are sort of special for everyone, but (for obvious reasons) possibly even more so for Kramnik.
Or would you want Kramnik to play horribly in his rapid game against Topalov, just to confirm his statement [and to give Topalov fans a free ride on this blog for the next month ,:)]??
I could also say: My results in the Sicilian are better than in the Ruy Lopez - and one outlier (a good or bad game in either opening) doesn't immediately change the whole picture.

Looks like Vishy has joined the party.

This dispute on Kramnik's statement about his blindfold is perhaps a new low for all of the years I have been following the dirt. Would people mind keeping the chatter a cut above miserably inane?

Of course Kramnik did not say he was generally better in blindfold than in rapid. He probably genuinely thought he had played better that far in the blindfold games than in the rapid games, but he certainly did not say that it would always go on like that. So if his statement was true when he said it, it will remain true 1000 years from now. And just in case someone thought he was serious about preferring to play blindfold in both games, which would not be too surprising around here, I can assure you he wasn't.....

Not even close to a "new low" though, DP. I would say this is about par.

Kramnik is blindfold king?? hmm.. Carlsen has overtaken and Anand is about to.

Blindfold winners (including shared wins) in Amber tournament:

Kramnik: 1996, 1998, 1999, 2000, 2003, 2007, 2008.

Morozevich: 2002, 2004, 2006, 2008.

Anand: 1993, 1994, 1997, 2005.

Topalov: 1999, 2001, 2008.

Karpov 1993, 1995.

Shirov: 1999.

Aronian: 2008.

In one of Karpov's participations, not necessarily one he won, he tried repeatedly to make an illegal move in the blindfold, not realizing he didn't have a certain pawn anymore. By the time he figured it out and tried to enter the correct move he lost on time. He protested about a computer malfunction and the game was declared a draw. I think it was against Nunn. IIRC, Korchnoi was there and had some hilariously cutting remarks about the incident in a magazine somewhere.

Korchnoi and van Oosterom's favorite, Lev Polugaevsky, played in several early Ambers. Even Larsen played in the first in 1992, though there wasn't blindfold that year. He didn't do well, but I remember he beat Anand on the black side of an Accelerated Dragon, which I was devoted to. Had to look up the games in a thing called a "magazine" back then. It was a year or two later I copied van Wely's Tascbase database! Good old Tascbase.

Perhaps Kramnik should have waited till the end of the tournament before that quip.

comedy of errors! ok, I don't fault Kramnik for that. He is still the King. He only lost to the Chess King of Kings, the best ever!!

Which quip? As far as I know, Kramnik never claimed to be the "king of blindfold", at least it is not in the statement quoted by acirce (March 19 5:03AM). Only PircAlert put _those_ words into Kramnik's mouth; Rubinstein and greg koster subsequently stated that he would have the right to say so - based on long-term evidence which is not refuted by the result of one or two games, and wouldn't even be refuted if Kramnik loses all of his remaining blindfold games and drops (just) below 50% in Amber 2009.

To my knowledge, the only self-proclaimed king in the chess scene is/was "King Loek"(van Wely) - but that had to be taken with a grain or rather a barrel of salt even when he was rated above 2700 .... .

From ChessVibes: "..Topalov wanted to see the Anand-Kramnik blindfold game just before he had to play himself, “to get in a good mood”

They really hate each other :)

"Kramnik never claimed to be the "king of blindfold", at least it is not in the statement quoted by acirce (March 19 5:03AM). Only PircAlert put _those_ words into Kramnik's mouth; Rubinstein and greg koster subsequently stated that he would have the right to say so - based on long-term evidence which is not refuted by the result of one or two games, and wouldn't even be refuted if Kramnik loses all of his remaining blindfold games and drops (just) below 50% in Amber 2009"

that's becoz nothing can refute it :) moro starts competing 2002 and wins more blindfold than kramnik and scores the better results of the two, still kram is blindfold king. if carlsen wins six years in a row kramnik is still blindfold king since he once won 7, to those not wanting to see the truth :)

gg - while the entire discussion is about a statement that has never been made - in this context even kings can lose their title ... just as it seems that King Loek has left his throne.

Hilarious :) Carlsen is leading the blindfold with a TPR of 3013, Anand is a point behind him after beating Kramnik in a very fine effort - but these worthies are too busy proclaiming the 1000 year Reich to bother about such trivialities.

Can somebody pls explain why not 8.d4 in the Anand-Topalov rapid game?

sorry, guess it doesnt make that much of a difference.

Twitter Updates

    Follow me on Twitter

     

    Archives

    About this Entry

    This page contains a single entry by Mig published on March 18, 2009 12:09 AM.

    Clay-mate-tion was the previous entry in this blog.

    Tomashevsky Wins Euro Ch is the next entry in this blog.

    Find recent content on the main index or look in the archives to find all content.