Another Number

Keywords: EndGame

The following position chills to -1/2 (1/2 point for White).

If Black plays the result is 0. That chills to -1, because it costs Black a point to make a play.

reverses. The result is -1, which chills to 0.

gauss: Why do we assume Black responds to White 1? Attempting to answer my own question: is it that White 1 turns this into the number -1 1/4 (it's a corridor; see parent), and so because the original situation was worth less than -1 1/4, we may assume there's nothing larger than -1 1/4 elsewhere on the board, so Black will do best to play here next?

Bill: gains 3/4 point, taking us to a value of -1 1/4. Then also gains 3/4 point, taking us back to a value of -1/2.

The chilled game looks like this:

                  { -1 | 0 }

In CGT that equals -1/2. That's no surprise. Go players already knew to count that as 1/2 point of White territory.

--BillSpight

gauss: This looks like 3/4 for white to me. Let me know what I'm doing wrong if I'm not calculating things correctly.

Actually, I've just realized this doesn't make sense, as it doesn't give the correct answer when added to -1/2 or to -1 1/4, using the corridors from numbers. Adding it to -1/2, for instance, shows that if it's a number, then it must be -1/2 (as it's clearly no worse for white than the smallest corridor, which is -1/2): just check that even with white sente, all white gets is 1 point (so it isn't worth more than 1 for white, as white sente is supposed to round up the number, according to the parent page).

It still seems slightly better than the smallest corridor, though. Maybe it's -1/2 plus an infinitesimal? Probably not actually: see my comment higher up on the page for my guess at why.

In any case, is there a more complete discussion of how to compute these somewhere?

My original (faulty) computation:

This situation is now worth -1 1/4 (it's a corridor), as mentioned on the parent page. We average^[101] this with:

This situation needs to be further analyzed (I claim it's -1/4):

This is worth zero.

This is worth -1/2 (it's a corridor), as mentioned on the parent page.

The average of these two is -1/4.

Finally, we average -1/4 and -1 1/4 and get -3/4, the value (or number? I don't know the terminology) of the original position.

Bill: gauss, there is nothing wrong with your approach, but you have to take it further. You start off assuming that each play is gote. Fine, but then you have to check whether that is the case.

For instance, let us assume that is gote. Then, as you say, the position before is worth -1/4. After it is worth -1/2, so gains 1/4 point.

But then plays to a position worth 0. gains 1/2 point, which is more than the 1/4 point that gained. Therefore is sente. It's a reduction ad absurdum. If is gote, it is sente.

[101]

Bill: No, we do not average it, because is sente.

Bill: After we are back to -1/2, which is what we started with. Sente gains nothing.

Bill: We can verify the value of -1/2 by the method of multiples. Two of these yield -1 (one point for White).

Bill: No matter who plays first, the result is -1. The two positions are miai.

Path: <= CGT path =>

Another Number last edited by Bill on November 11, 2007 - 06:27

RecentChanges · StartingPoints · About

Edit page ·Search · Related · Page info · Latest diff