bob wrote:I do not follow your reasoning regarding either collision or repetition.
Yes, I can see that. We're having a serious communication problem. I'm tired of trying to explain my self. So lets cut to the chase!
bob wrote:Path 1: I start with a capture, then play 10 other moves. Let's call these moves A through J.
Path 2. I start with move B then go through J, and then finally play move A. That is a classic transpositions. Doesn't matter whether we have just one capture (move A) or several. All that matters is that we change the order of the moves but reach the same position.
That's a nice fantasy story you have going on there. You have what amounts to an infinite size TT, and there is no search that can interfere with you finding the transposition by overwriting your TT entries. There is no engine telling you, “oops, sorry but your transposition didn't make it into the TT because the moves are so flawed that LMR, Null move, or futility pruned them away.” Your A, B, C, moves aren't any more real than the transposition they're meant to be a part of.
So let's get real.
Here is a real position:
[d]1q2r1k1/1b1nr1bp/pp1np1p1/3p1pN1/PPpN1PP1/2PPB1QP/4PRB1/2R3K1 b - - 0 26
You make it out like finding 10-ply transpositions is a piece of cake. Let's forget the fact that the ones we've been talking about up till now have been buried at the bottom of a deep search. You have claimed 10-ply transpositions are SO EASY to find, so you should have no problem finding tons of them. You can use as large a TT as you like. There are only two restrictions. 1.) The transposition has to start and end with at least half the material still on the board. (i.e. you can't reduce the material to an endgame position to make them easier to find.) 2.) At least 2 variations that lead to the transposed position have to be playable. (i.e. no blunders that drop material just for the sake of completing the transposition.) This task is at least 7 orders of magnitude less hard than if it were at the bottom of a long search.
The real test:
You get 16 gigabytes of TT space not one byte more. You have to conduct a “real” search with a real engine that will write real entries all over you carefully prepared but flawed transposition. It will have all of it's normal pruning methods turned on so it can put the ax to all those theoretical transpositions you seem to think will make it into the TT . Your job is to report how many 10-ply transposition are found by a normal TT with the first move of the transposition being at a MINIMUM depth of 33-plies into the search. The line of play leading to the first ply of the transpositions CAN NOT contain blunders that drop material.
I have no doubt that your going to find out that all those fantasy transpositions either don't make into the TT to start with and the few that actually pass muster will be quickly overwritten.
The point I'm making is that you are comparing a “fantasy” TT to a real one. When you compare a “real” TT that is highly over-subscribed to a “real” one that isn't the shoe will be on the other foot.
Feel free to prove me wrong.
We could argue this point, or the point about split TT's in general, to death. But I see little point in it. It's complicated enough that arguments alone are insufficient to be convincing.
Regards,
Zen
Only 2 defining forces have ever offered to die for you.....Jesus Christ and the American Soldier. One died for your soul, the other for your freedom.