the great odds ?

Kingpatzer, you are correct that the % chance to win represents observed events, and some of those observed events are the rare ones, like patzers beating masters.

Fro example: http://en.wikipedia.org/wiki/Hetul_Shah

You state categorically that a 900 will never beat an expert. What about 1100? 1300? 1500?

A 1500 will, but a 900 won't? Where is the cutoff line? What about a 901?

The thing is, there is no cutoff, none at all. The rating just represents a bell shaped probability curve. There is no rating differential where you can just say its "impossible". It just becomes decreasingly likely.

 You are right.

The question is about winning a tournament or a match. Given the caveat above, it won't happen. It won't happen because it is not about random events. 

And a 900 beating an expert even in a single game (again, given the caveat above) won't happen because of the sort of mistakes a 900 represents. 

You wiki link demonstrats someone meeting the conditions of the caveat that I allowed for and in no way represents a counter factual to my claim.

 This thread is about all of it, games, matches, tournaments. And it doesn't matter, because a match is just a string of games, so the 900's tiny probability to win against a expert is just multiplied out by however many games we are talking about, making it increasingly extremely rare, but never ever impossible.

My link demonstrated one example even worse than what you mentioned, you said an expert would bever beat a GM, and this was a class A player defeating one (or whatever they call them in FIDE).

if you insist, then please let us know at what rating difference it becomes "impossible", and, more importantly, how you arrived at this conclusion.

You gave an example of a rapidly improving player, which pretty much definitionally meets the caveat I presented. 

The other problem this discussion represents is that human brains are in fact changable, and playing good players is pretty much the best way to learn how to play better. So you can't even really set up a valid experiment to test this as after the first game, the lower rated player has likely already learned something which will impact their rating to some degree. 

 Well, no I didn't, if you look at the FIDE report on the link http://en.wikipedia.org/wiki/Hetul_Shah

One year after this young player defeated a grandmaster, his rating only improved from 1817 to 1920, a large chunk of which was from that one victory, of course! In fact, later on in that same year, his rating even dropped 25 points, exactly what you would expect if he had some sort of improbable freak victory way above his head.

1920 is still not a GM, or even a candidate master. If they had a rematch today there would still be 500 points seperating them.

Anyway, maybe you guys go need to argue with Arpad Elo, he set up the system as a statistical measure of probabilities of success and not as some defnitive statement about who never ever ever has a possibility to beat whom.

http://www.chess.com/forum/view/game-showcase/omg-200-rated-pulls-off-unbelieveable-win-a-queen-down?ncc=5#first_new_comment this qualifys as great odds.

Where's my coke?

I would expect any of this to occur with less than 1% probability...

Well, there is impossible from the mathematical point of view, and then there is the event which is so unlikely that it has almost zero chance to happen in real life. From a practical point of view, I would assume in chess an event with odds < 0,1% could be considered as "impossible". For example, a class C player beating an IM in a match probably belongs to this category...

chessgdt,

The result you gave is dubious at best. The "P" after SPENCER TOY 's rating means his rating is provisional, i.e., not established.

You'll note the Mr. Spencer's rating changed much more for the game than his opponent's rating.

Let's get back to the orignal question. The question, was, at least to me, talking about established ratings.

That being said, a difference of 600+ rating points would mean the higher rated player would win over 99% of the time.

We'll work with the 1%.

The chances of an expert winning a six round tournament against all 2500-2660 rated player would be 1% x 6 games. Putting these figures in mathematical terms,

1/100 x 1/100 x 1/100 x 1/100 x 1/100 x 1/100 = 1/000000000000 or 1 in 10 to the 12 power.

A similar result would be expected for a Class C player in a match against an IM.

The safest bet would be a 900 player beating an expert.

I was a TD for years. I know this stuff - lol! 

the rating change was also because he won 2 games against higher than his rating players

If someone says impossible, then they better mean impossible, espescially on the internet.

I think Kingpatzer did mean impossible because he seems to think that weak players will inevitably make mistakes and there is no chance because they arrive at bad moves through flawed unrandom logic. This seems silly to me, but from my understanding of what he's trying to say, an algorithm playing completely randomly would beat a GM more often than a 900 player.