Poking around ChessBase, I notice some folks have let their top engines crank on the initial position for a very long time. For example, one guy has let Stockfish run 54 ply from the initial position. The engine consensus indicates White’s natural edge to be between 10 and 18 centipawns. If we examine positions later on in the game, we’d regard such an edge as practical equality. But, among top GMs and engines, White scores significantly better than this computed edge would indicate. Wikipedia (en.wikipedia.org/wiki/First-mov … e_in_chess) claims about 55/45 for White. So, why does White tend to outplay his natural edge?
I’m not claiming any of this has practical significance, but it’s interesting.
Any of you statistics/rating jocks want to venture opinions ?
my two cents… i believe the human element comes into play here. people get tired during games. little mistakes get magnified and it is easier to play with a slight advantage, where a mistake could make the game more equal, while a slight mistake on the “inferior” side can be catastrophic.
I’ve heard an interesting tidbit, although I think I may actually have read it in the third edition of the rule book. (I don’t have my copy available right now, so I can’t check.) It was determined that, among top players, the advantage of having white was worth about 0.5 points in three games. That was the motivation for the absolute ban on assigning a player the same color for three consecutive rounds in the third edition.
That number would round to 58/43 which is pretty close to Wikipedia’s number for actual GM games. So, it looks as if their reasoning was based on actual results rather than engine analysis.
Given that the third edition of the Official Rules of Chess dates from the mid-1980s, I think it is a safe conclusion that engine analysis played no role whatsoever in the reasoning involved.
I tend to agree, but the Wikipedia article says the same 55/45 ratio holds for games between engines, where exhaustion plays no part.
I first tried resolving this by reference to two sorts of equality: absolute equality and practical equality. Of course, we can only approach knowing the former.
Take an extreme case of two hypothetical positions, “A” and “B”. Top engines crank for 50 ply on both and come up with 0.07 White edge for each. Pretty close to absolute equality. But if these same engines are allowed only 20 ply, position “A” comes up with five lines that all look pretty good, but given more time and computer power, we find four of the lines end up in the -2.00 range and only one of them reaches the equality nirvana. With position “B”, all the lines that look good at 20 ply come out in the range of 0.07 to 0.12 at 50 ply.
So, in an absolute (or as near as we can get to it) sense, BOTH positions are equal. But in a practical sense, they are not equal. Position “A” is full of pitfalls and, given our constraints and lack of knowledge, the choices are, in effect, random and White will end up losing most of the games.
We could speculate that the longer engines run, these inferior lines get discarded and we approach absolute equality. But the Wikipedia article says the 55/45 ratio holds in engine matches with both short and fast time controls, which shoots a hole in THAT theory.
A couple of comments: first, computers are notoriously poor at analyzing opening positions. Exhaustive analysis is just not that effective there. You and I know that 1…f5 is just not that effective against 1.e4, but engines have to chase it down all its rabbit holes. Second, 40/120 is not terribly slow. It is silly to say that an engine can determine the truth of a position in three minutes. No CC player would dare move based solely on what an engine (any engine) recommended after three minutes. Third, this data is based on tournaments from 2009. Six plus years is forever in computer terms.
I don’t see any inherent method for converting centipawns to win-loss percent, or vice versa. It might be reasonable to assume that a 300-centipawn advantage in the initial position would correspond to virtual 100% odds for the player with the advantage, at least among masters or higher. But does it vary linearly from 300 down to 0, as the centipawnage becomes 0? Or would 200 centipawns still be enough to ensure virtually 100%? – if so, now we would ask whether it is linear from 200 to 0 rather than 300 to 0.
An edge of “x” centipawns doesn’t mean much unless one knows the depth of the calculation from which it was derived.
I think it safe to say that, at least from a human perspective, an edge in the range of 15 centipawns (White’s edge is the initial position) has always seemed virtually meaningless. But from the starting position in games between strong players, White scores 55/45, a result which is NOT so meaningless (roughly, Komodo’s winning percentage over Stockfish or, from another perspective, 40 rating points). Why is this? How does practical equality slide into a solid advantage? Has this ratio persisted into the latest engine tournaments?
The advantage needed to ensure the likelihood of the stronger side winning approaches 100% seems a separate question.