I wonder what is the thinking game that engines earn most elo from doubling the speed

[Uri Blass]

In chess the gain from doubling the speed was in the past something like 70 elo but engines became stronger and today top engines earn almost 0 elo points from doubling the speed.

I wonder if top engines earn today more from doubling the speed in go
I guess that engines may earn more elo from doubling the speed in games that are similiar to chess if you change the rules to have no draws in chess or if you increase the number of ranks in the board.

Is this even a question? Obviously those games that become more complex with depth. Shogi would be a good example.

I am not sure and in theory a complex game may be so complex that even when you double the time engines still have no idea who is going to win.

This is kind of ridiculous word salads.
If you double the time IT IS MORE LIKELY the engine is able to a better eval of the status of the position.

For example engines play 20 chess moves and white is the winner if perft(50) of the final position is even and black is winning if perft(50) of the final position is odd.

This is kind of pointless.
Perft(50) gives a result only if there were no pawns/captures. Otherwise you are left with the eval. We all know this already.

Here we even practically cannot know the winner after the game is finished ,
but it is possible with a similiar idea that we can find the winner, but finding the winner may take 24 hours of search that the engines do not have during the game.

And, so, your point is?

My point is that it is possible to have a game when the engines know the rules of the game but calculating the result of the game takes so much time that practically the time is not important for rating for a reasonable time control(if the engines know only that the game is finished after 20 moves and calculating who won take 24 hours then they can earn no rating points from the fact that they have 2 hours for all moves instead of 1 hour for all moves).

The earning of rating points is a statistical thing. More time = more likely to find a better move. Sometimes will and sometimes won’t, but more likely will with more time. Isn’t this kind of obvious?

1)More time is more likely to find a better move only if the time is enough to decide which move is better.
It is easy to have a game when engines have not enough time to know which move is better if clculation of the identity of the winner after you know the game is finished takes more time than playing the game.

2)You can say that the example in 1 is not a practical problem but even if we talk only about practical games the question is about rating improvement from doubling the time for games that we do not know the solution.
I am not sure if more complex game means more rating improvement from doubling the time and I am not sure if go with a bigger board is always going to get more rating improvement from doubling the time.

I do not know if people tried to test it with boards of different dimensions to say for example 39*39 gives more rating points improvement from doubling relative to 19*19 and 79*79 gives more rating points relative to 39*39.

[chrisw]

In chess the gain from doubling the speed was in the past something like 70 elo but engines became stronger and today top engines earn almost 0 elo points from doubling the speed.

I wonder if top engines earn today more from doubling the speed in go
I guess that engines may earn more elo from doubling the speed in games that are similiar to chess if you change the rules to have no draws in chess or if you increase the number of ranks in the board.

Is this even a question? Obviously those games that become more complex with depth. Shogi would be a good example.

I am not sure and in theory a complex game may be so complex that even when you double the time engines still have no idea who is going to win.

This is kind of ridiculous word salads.
If you double the time IT IS MORE LIKELY the engine is able to a better eval of the status of the position.

For example engines play 20 chess moves and white is the winner if perft(50) of the final position is even and black is winning if perft(50) of the final position is odd.

This is kind of pointless.
Perft(50) gives a result only if there were no pawns/captures. Otherwise you are left with the eval. We all know this already.

Here we even practically cannot know the winner after the game is finished ,
but it is possible with a similiar idea that we can find the winner, but finding the winner may take 24 hours of search that the engines do not have during the game.

And, so, your point is?

My point is that it is possible to have a game when the engines know the rules of the game but calculating the result of the game takes so much time that practically the time is not important for rating for a reasonable time control(if the engines know only that the game is finished after 20 moves and calculating who won take 24 hours then they can earn no rating points from the fact that they have 2 hours for all moves instead of 1 hour for all moves).

The earning of rating points is a statistical thing. More time = more likely to find a better move. Sometimes will and sometimes won’t, but more likely will with more time. Isn’t this kind of obvious?

1)More time is more likely to find a better move only if the time is enough to decide which move is better.
It is easy to have a game when engines have not enough time to know which move is better if clculation of the identity of the winner after you know the game is finished takes more time than playing the game.

2)You can say that the example in 1 is not a practical problem but even if we talk only about practical games the question is about rating improvement from doubling the time for games that we do not know the solution.
I am not sure if more complex game means more rating improvement from doubling the time and I am not sure if go with a bigger board is always going to get more rating improvement from doubling the time.

I do not know if people tried to test it with boards of different dimensions to say for example 39*39 gives more rating points improvement from doubling relative to 19*19 and 79*79 gives more rating points relative to 39*39.

Yes, it was that original question that I tried to respond to (the complexity argument).
If Go (more complex than chess?) or Shogi (definitely more complex than chess and larger game space) computer communities have rating lists and gazillions of published games then it should be possible to answer your question with data based observations.

[towforce]

I guess that my not very interesting answer is "the game with the lowest branching factor" and that is just a guess since I did not test it.

I'm going the opposite way: "the game with the highest branching factor"

A game with a low branching factor is likely to get solved (or almost solved) on the first move.

[chrisw]

I guess that my not very interesting answer is "the game with the lowest branching factor" and that is just a guess since I did not test it.

I'm going the opposite way: "the game with the highest branching factor"

A game with a low branching factor is likely to get solved (or almost solved) on the first move.

Er, not exactly, you’re thinking of games with a branching factor less than or equal to one.