Greengard's ChessNinja.com

Dresden Olympiad r2

| Permalink | 46 comments

Still in progress. Curiously, all four games on the first table, Poland-Russia, are still in progress when most of the day's games are finished. Probably not due to tension over the missile defense shield. All four games are up in the air. Btw, if you are watching at the official site, you can toggle between the single-game viewer and all four tiny boards of a match by just clicking the board. Of the completed matches, favorites China, Armenia, and Azerbaijan scored 3-1 over Belarus, Moldova, and Slovakia, respectively. Nice win by Ivanchuk with the cool 25.Bf5! More later, plus podcast.

Update: ICC Chess.FM podcast for round 2 now up here. Results, some on-site tidbits culled from the web and email, Hero of the Day, etc. 19 minutes. 4.3mb MP3. Right-click and save file locally for best results.


What happened in the Akobian game? He was hopelessly lost until white erred with 56.Kc7?? Just when it appeared that he might get a desperation perpetual, the game abruptly ended. I don't see a trivial win... certainly not as trivial as a few moves earlier. What happened? Did Akobian lose on time?

What the heck? On the official site, the Grischuk game is shown as 1-0, although it seems to be either a draw or a win for black. In fact, wouldn't the 50 move rule imply that a draw was reached after 112 moves...? Can someone clarify?

In Mastrovasilis - Akobian, white is winning after 56.Kc7, it is no blunder. However, I don't know about Kd8 - perhaps the final position is drawn.

It seesm like Grischuk- Wojtaszek are going for an olympiad record of the longest game (there at move 113)- its been over 50 moves since a piece was taken or a pawn moved but I guess Grischuk had already stopped writing his move down before the last pawn move was made (move 53)as he has not claimed the draw. - its a theoretical draw anyway. Now the site shows 1-0 for grischuk presumably its a draw Russia have won as the Jakovenko game has been over for ages but the site never recorded blacks resignation

Mamedjarova vs Tilenbaeva early 1-0,
as Tilenbaeva was at the board ca. 16.03
rules are strict apparently..

The draw has to be claimed and then verified by the arbiter and Grischuk only has to tick his scoresheet once he gets below 5 minutes on the clock so he couldnt tell I suppose

I thought it was just a warning until round 3?

Andy, are you sure about the scoresheet. As far as I know when you have increments you have to write on the scoresheet till the very end.


Yes perhaps you are right - I dont know but it would make semse - in which case Grischuk just forgot the rule

The Olympiad provides an interesting opportunity to test whether 2700 GMs are really that stronger than lesser GMs. I always felt that the difference between a 2700 and say a 2600 GM was due to the ability (luck?!) to get invitations to strong round robins consistently. The problem is 2700 GMs do not play 2600 GMs very often so it is difficult to make comparisons and have a sample large enough to run the numbers. But the Olympiad is an exception. Take all the games played by 2700 GMs against GMs rated below 2700 and calculate the rating variation. This should be a large enough sample and I bet the 2700 GMs will lose much more rating points than would be expected by the rating differential. I wonder if there is anybody out there with the statistical knowledge to run the numbers and share the results with us.

I checked the performance of 2700 players in the last 10 Olympiads. Some performed lower and some higher than expected by their ratings. The average difference between actual and expected results is not significant, meaning they performed more or less according to their ratings.

Notice that to perform according to rating expectation does not require spectacular scores. A 100 points difference only requires about 65 % score. So a 2700 playing against 2600s needs approximately plus 2 from 10 games, which does the job and keeps his rating.

A 200 points difference requires about 75 % score to keep the rating. So if a 2700 playing 2500s, or Kasparov playing 2600s, wins half of the games and draws the the rest, he will keep his rating. But of course the 2700s will not face only 2500s, some of his opponents will be 2600s, and perhaps a few 2700s - so the required score will be lower than 75 % to keep his rating.

More about the performance of 2700 players.

As a concrete example, let's take Ivanchuk. He plays in all the recent Olympiads, and tends to play in almost every round (often all the way from 1st to last round), so he played all levels of players, and his data is good.

Take Ivanchuk's performances in the last 9 Olympiads, since 1990 (he also played in 1988, but he wasn't top yet, so lets skip that year).

These are the differences between his performance rating and actual rating (plus meaning he performed better then expected), for each Olympiad.

1990 +31
1992 -21
1994 +28
1996 +79
1998 -33
2000 +9
2002 -31
2004 +114
2006 -15

The average is only slightly positive. Basically, he more or less performed as expected.

Notice that these are NOT points he gained or lost, but differences in performance and actual ratings. The actual rating points gained or lost is only a fraction of these differences.

Sorry I meant

"... a 2700 playing against 2600s needs approximately plus 3 (THREE) from 10 games ... "

It is true 2600 rateds are not often invited to elite tournaments. But there are other places where 2700s meet lower rateds, besides the Olympiad. There are club championships in various european countries (in which even Anand and Kramnik play), the bigger open tournaments, national and regional championships.

After all, the 2700s cannot just play 2700s all the time, simply because there are not that many of them :-). Even they don't always get invited at elite tournaments, so they need to play in some opens, etc.

Thank you for your analysis Henry, I appreciate it. I must confess I am surprised by the results. I just calculated the statiscs for round 1 of Dresden Olympiad and came up with results similar to what you describe. There were 12 games of 2700s vs lesser GMs in round 1. The average rating differential between the two groups was 178 points (2735 vs 2557) and the 2700s scored 71%, consistent with what you said. But of course the sample size is too small to draw any conclusion.

I have a question for you: what is the formula to calculate the expected score given a certain difference in rating points? Say, 50, 100, 150, 200 points difference?

Actually my numbers for round 1 were wrong. There were 20 games of 2700s vs lower rated GMs with 183 points difference (2736 vs. 2553). The 2700s score 70% (14 points out of 20), consistent with your results.

For rounds 1 and 2 combined I get 42 games with 164 points difference (2736 vs 2572). The 2700s scored 69% (29 out of 42).

Ed, there is a formula (that I don't fully understand - I believe exp has to do with logarithms)here: http://www.flyordie.com/games/help/chess/en/rating_chess.html
If you go to page 25 of the following document, you will see a table, where you can look up the expected scores, given the difference in rating points.

Some of the pairings in 3rd round seem absurd:

No. 13 India vs No. 91 El Salvador
No. 10 USA vs No. 84 South Africa (highest rated player is 2369)
No. 6 Bulgaria vs No. 102 nicaragua (Topa playing 2289 rating player)


On the olympiad site, where can the player-vs-player pairings be found?

Rubert, these "absurd" pairings always show up in the first round or two when you switch from an accelerated system (used so far) to a normal Swiss (as it will be from now on).
Personally I'm not a great fan of this new rule (and even less of the one reducing the number of rounds, while I can live with lowering the number of reserves from 2 to 1), but let's wait for the final standings before judging if FIDE's latest changes seriously alter the results or not.

Here is a link to the Rd3 pairings.

Go to home page -> Click English flag -> Click on the pairing, results link -> Click on the Rd3 on Team... something in duetche on the chart!

In individual tournaments, it often seems that accelerated systems have the event of merely switching the first round for the 1st non-accelerated round. This usually means that the accelerated rounds did not do a good job of dinging the top players (or here, teams).

Maybe two accel. rounds isn't enough....

OTOH, Nicaragua, El Salvador, & South Africa earned their right to get prestige pairings: gens una sumus.

Very interesting rd 3 game Wang Yue v Magnus which I think is a game between 2 future world champions In an equal endgame the chinese No 1 outmanouevred magnus and at the critical point 31 Rg8+ wins a pawn with excellent winning chances instead Wang Yue chose gxh4+ and magnus just held with great defence.

IM Watu Kobese of South Africa beats GM Hikaru Nakamura! I thought Kamsky was a better matchup, but I'm sure he was dealing with his match negotiations. Ten years ago Kobese was considered a remarkable young talent. He has a universal style and has also beaten Peter Leko and Judit Polgar. However, South Africa lost by a point.

Watu deserves all the credit by giving Hikaru a run for the the money, but Nakamura did win the game at the end. You just have to pay attention to the position on the board instead of going by results on the official site (which were wrong in this case). Yuri Shulman did get held to a draw, and the final position is very tough for him, I think, but he had to content with distractions of his clock and electronic board malfunction: his time was running even after he pressed the clock.

what are you talking about? Nakamura won the game.

Wonder where is Daaim and his predictions.

The results and games are a mess. I looked at that game earlier and it was clear only Nakamura was playing for a win, so barring some spectacular blunder on his part or time forfeit, I assume he did win. There is also confusion about the 4th board of Georgia-Ukraine. It's reported as a win for Arutinian on the official site and as a win for Volokitin on the chess-results site. In the live viewer and the TWIC PGN, the score is the exact same as board three. The PGN file on the official site has a different score, but no result. Black is winning in the position, but then again, the score the PGN file has for Jobava-Ivanchuk is totally different from the 95-move game on the live page. Sigh. So much for German efficiency.

Can't see how Arutinian - Volokitin can be a win for Volkitin. Ivantchouk was black on #1, therefore it makes sense that Volokitin on #4 is white, the pgn file and the "live" display both end with 45...Re7 and both give 0-1 which is reasonable given that the position is an ending with one more knight for black and no tricks for white.

Got it, the game logs of Arutinian - Volokitin are copycats of the Mchedlishvili - Efimenko game, played on board #3 of the same match.

Probably someone connected twice the #3 sensory board's output, and the game #3 was displayed twice (while the actual game #4 is now lost for posterity)

Anyway, so much for German efficiency.

So Nakamura wins! OK... my mistake, but these problems are pervasive and if you're not physically there, there is no way to know. Many folks here posted one full hour after I posted and certainly got the correct result. Thanks.

Don't understand your question.

It's so strange now. I saw the final position and assumed there must've been some time scramble or some malfunction or even incorrect score, but Nakamura does not do that. It was strange. So OK... don't pay attention to the results. We're screwed.


"The problem is 2700 GMs do not play 2600 GMs very often so it is difficult to make comparisons and have a sample large enough to run the numbers."

it depends on what you mean with "not very often". here's the fact:

sometime 6-12 months ago (don't remember exactly when) i created some statistics in this respect, and it turned out that 2700+ players on average played more than 50% of their games against sub-2700 players.

another thing that some might find surprising, but which actually is quite logical - on average, the 2700+ players gain more rating points from their games against sub-2700 players, than in their games against fellow 2700+ players.

but i agree that a nice thing about the olympiad is how 2500-2600 players get some shots against super-gms.

As Frogbert said, it is not true that 2700s play mostly against 2600s.

I checked the average opponent of each 2700 players since 2002. Almost always, the average is less than 2700, which means they do play more against lower than 2700s.

Even for Anand, one of the most elite, the average opponent is almost exactly 2700, which means about half his opponents are lower than 2700. Other top players have average opponents lower than that.

The news covers almost exclusively Anand's appearences in 2700 level events, giving the impression that's all he does. But he also plays many games in the German Bundesliga, Olympiads, Europan Cup, against much lower opponents.

Come to think about it, it should be obvious that 2700 players play many games against lower rated opponents. There's simply not that many 2700 level tournaments to play in. To make a living they have to play in lower rated events as well. Anand and Kramnik are regularly invited to Corus and Linares, but somebody even as high as 15th in the world might not get elite invitations that often. They just have to play in lower rated tournaments.


I made a typo again. The first sentence must be:
" ...it is not true that 2700s play mostly against 2700s". (not 2600s).

Yeah, the Olympiads give rare chances for low (even very low) rated players to get a shot against super-GMs.

Just today, a 2200 rated guy was sitting opposite of Topalov. He lost, but it might still be one of the highlights of his whole career.

As for Watu Kobese,
He might be a rising star in the past but, perhaps partly due to lack of opportunity, the hopes are not fulfilled. He is still only 2300s, is now 35, and last time I checked, his Elo is declining.


Given the rating difference, the expected score is calculated as follows:

ExpectedScore = 1 / (1 + 10^(-difference/400))

The number you get will be a ratio, such as 0.6, which is the proportion expected to be achieved.

You could also reverse the formula to get the Rating Performance, given the score:

RatingPerformance = CurrentRating + 400 * log ((1-score)/score)

Again, score is the proportion of win.

The last formula should be:

RatingPerformance = OpponentRating - 400*log ((1-score)/score)

You're right. It's very unfortunate. It just shows that there aren't any viable programs to harness the talent of players showing promise from unheralded regions of the world. It's tough.

henry, for an average of 2700, anand just has to play say one opponent at 2500 and 8 opponents at 2725, for example. i am guessing that the average of the above 2700 players in the world is very roughly around 2725...

using the median, or even say the 10th percentile would be a much better measure. so topalov played this 2200 guy here, and now, if he plays just over 2700s for the next say 20 games (avg of 2725), his average opponent ELO is still 2700.


Theoretically, those possibilities are there. But that's not the case here. I actually checked Anand's opponents. For instance in the Bundesliga, he did play a good number of sub-2700 opponents.

By the way, notice that to measure the number of 2700 opponents, it's not as simply as averaging the ratings of 2700 players. He could be playing certain opponents, say Kramnik or Topalov, more often than other 2700s, and these might have 2750+ ratings.

The case of 2700 playing 2200 is very rare :-).

For those who say that 2700 players avoid playing sub-2700 because of the risks involved, just look at Topalov.

Because his team is only giving middle class performance, he keeps getting much lower opposition (average even lower than 2600). But he keeps playing. And winning. So far only 1 draw in 6 games. With a performance that's even better than expected from his rating. What risk?

Twitter Updates

    Follow me on Twitter



    About this Entry

    This page contains a single entry by Mig published on November 14, 2008 2:30 PM.

    Chernenko: You Show Me Yours was the previous entry in this blog.

    Dresden Olympiad r3: So Much for German Efficiency is the next entry in this blog.

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