Life In 19x19

If I was faced with such a position in an OGS game where I was doing detailed yose analysis and such a position was not obviously double sente, I would classify that position as:
black 4 points gote with 32 points follow-up,
white 4 points gote with 30 points follow-up.

Then I can easily look at the tree for answer and tenuki, and decide the best time to play it.

Click Here To Show Diagram Code

[go]$$Bc A - +5 (for Black)
$$ --------------------
$$ | . . . . . . . O X .
$$ | . . . X O . . O X .
$$ | . . . X O . O O X .
$$ | . . . X X O . O X ,
$$ | X X X X O O O O X .
$$ | O O O O X X X X X .
$$ | . . . O O O . . . .
$$ | . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc B - +25
$$ --------------------
$$ | . . . . B . . O X .
$$ | . . . X O . . O X .
$$ | . . . X O . O O X .
$$ | . . . X X O . O X ,
$$ | X X X X O O O O X .
$$ | O O O O X X X X X .
$$ | . . . O O O . . . .
$$ | . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc C - +42
$$ --------------------
$$ | . . . . B . . O X .
$$ | . . . X O B . O X .
$$ | . . . X O . O O X .
$$ | . . . X X O . O X ,
$$ | X X X X O O O O X .
$$ | O O O O X X X X X .
$$ | . . . O O O . . . .
$$ | . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc D - +8
$$ --------------------
$$ | . . . B B W . O X .
$$ | . . . X O W . O X .
$$ | . . . X O . O O X .
$$ | . . . X X O . O X ,
$$ | X X X X O O O O X .
$$ | O O O O X X X X X .
$$ | . . . O O O . . . .
$$ | . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc E - -15
$$ --------------------
$$ | . . . W . . . O X .
$$ | . . . X O . . O X .
$$ | . . . X O . O O X .
$$ | . . . X X O . O X ,
$$ | X X X X O O O O X .
$$ | O O O O X X X X X .
$$ | . . . O O O . . . .
$$ | . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc F - +4
$$ --------------------
$$ | . . B W W . . O X .
$$ | . . B X O . . O X .
$$ | . . . X O . O O X .
$$ | . . . X X O . O X ,
$$ | X X X X O O O O X .
$$ | O O O O X X X X X .
$$ | . . . O O O . . . .
$$ | . . . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc G - -34
$$ --------------------
$$ | . . . W . . . O X .
$$ | . . W X O . . O X .
$$ | . . . X O . O O X .
$$ | . . . X X O . O X ,
$$ | X X X X O O O O X .
$$ | O O O O X X X X X .
$$ | . . . O O O . . . .
$$ | . . . . . . . . . .[/go]

It is plain that the gain for the next play after Black plays first is 17 points, and that the average value of the position is +25. It is also plain that after White plays first the gain of the next play is 19 points, and that the average value of the position is -15. That means that the first play gains 20 points, more than the next play by either side. Therefore this is a gote.

To add what might be a further helpful comment in understanding this for people who are confused:

Bill is defining a "gote" move in a particular way - namely one where the urgency of play is decreased after the move.

Except for ko threats and such, this position is exactly equivalent to any position with the following properties:
* If black plays first, black gains 20 points and leaves behind one more move that either player can make that gains 17 points.
* If white plays first, white gains 20 points and leaves behind one more move that either player can make that gains 19 points.

In particular, if there are a wide variety of moves on the board of all different values, including plenty of moves with values between 17 and 20 points, and even moves with values between 19 and 20 points (ex: a position where a player can connect 19 dead-weight stones and make zero territory and where the other player can cut and kill them and also make 1 point of empty territory is worth 19.5 points under this counting method), then once the value of a move drops to 20 points, it will become correct for one player to play the hane, and for the other player not to respond. Not because of some weird tactical reason, not because of some weird thing involving tedomari, not because the endgame was crafted specially somehow, but simply because the position literally does become less valuable to play after a hane, and other moves whose values are in between will be worth more.

Morally, the position is no different than the following:

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------
$$ . . . . . . . .
$$ . . . . . . . .
$$ X X X X O O O O
$$ . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Bc
$$ ---------------
$$ . . . . . . . .
$$ . . . 3 1 2 . .
$$ X X X X O O O O
$$ . . . . . . . .[/go]

Click Here To Show Diagram Code

[go]$$Wc
$$ ---------------
$$ . . . . . . . .
$$ . . 2 1 3 . . .
$$ X X X X O O O O
$$ . . . . . . . .[/go]

Where:
* If black plays first, black gains 3 points and leaves behind one more move that either player can make that gains 1 point.
* If white plays first, white gains 3 points and leaves behind one more move that either player can make that gains 1 point.

It's just that the numbers involved are larger and it's harder to see at a glance that the urgency does actually drop.

The way I understand Bill's point is that if you define "gote move" as "move that decreases the local temperature" and "sente move" as "move that increases the local temperature", which are sensible purely local definitions, then every position is either gote for both sides, sente for black only, sente for white only, or settled (ex: such as a seki, where a move by either side is "sente" but is a losing move).

In particular, with these local definitions there should be no such thing as a genuine "double-sente" position - every position that people would normally think of as double sente, just like the one discussed here, is actually secretly gote for both sides, sente for black only, or sente for white only. The reason that these positions seem like double sente is due to a more global fact, such as the absence of other moves on the board with values between 17 and 20 points, as well as the implicit meta-knowledge that moves with such values are rare (so that if one side plays, the other side will respond because it's the next biggest move). Without these extra global assumptions, locally they are no different than ordinary gotes and one-sided sentes, it's just that the numbers involved are larger. This is what Bill means when he claims that you can define "sente" and "gote" both locally and globally in different ways, but "double sente" only makes sense defined globally.

Bill Spight wrote:It is plain that the gain for the next play after Black plays first is 17 points, and that the average value of the position is +25. It is also plain that after White plays first the gain of the next play is 19 points, and that the average value of the position is -15. That means that the first play gains 20 points, more than the next play by either side. Therefore this is a gote.

Those values are based on calculating the "average" value of the position, under the assumption that it is equally probable for either player to get the next move. Would it not make more sense to calculate the "likely" value of the position, using a better estimate of the probability for each branch of the decision tree? Classifying this position as "sente" for B means I assign a probability of 1 to the branch where W defends (because I consider it highly likely that on the rest of the board there will be nothing larger for W to play).

Both of these evaluations of the local position are uncertain approximations, but if we cannot solve the full-board problem, we rely on them to guide our play. In a real game, which weighting of the decision tree is more likely to be represent optimum play?

Calling a gote that gains 20 points with a reply that gains 17 points for one player and one that gains 19 points for the other a double sente does little harm. In practice such moves appear on the go board and disappear in a few moves, with sente. The problem comes with plays that are not played right away because they are too small.

Click Here To Show Diagram Code

[go]$$Bc Double sente -- Not!
$$ --------------------
$$ | . . . . . . . O X .
$$ | . . . X O . . O X .
$$ | X X . X O . O O X .
$$ | . . . X X O . O X ,
$$ | X X X X O O O O X .
$$ | O O O O X X X X X .
$$ | . . . O O O . . . .
$$ | . . . . . . . . . .[/go]

This is not a double sente. Nor is the double kosumi on the second line, nor are any number of plays identified as such. Thinking of them as double sente leads to confusion, at best.



Look at , for instance. How many amateurs would answer such a play because they have been taught that it was double sente?

Bill Spight wrote:How many amateurs would answer such a play because they have been taught that it was double sente?

Not me. It's middle game not endgame. It doesn't make territory as 3-3 still has a problem. White's purpose was to allow q10 peep to be answered at q9 and I'm not letting you do that in sente. I do perhaps have more fighting spirit than average, sometimes to my own detriment, but are there really many dan players who would answer that?

How many amateurs would answer such a play because they have been taught that it was double sente?

But where are amateurs being taught this is double sente? Surely only in a localised problem where there is no global override.

Agreed that you can see many amateurs answer, but I suspect the reason is not because of double sente. It's more likely a problem of a bigger magnitude, such as being unable to find a better move. If they could see one (e.g. killing a big group at one stroke) they'd make it, double sente or no.

My experience is that even DDK learn for themselves for principle that an opponent's move is only sente if you answer it.

What they are crying out for is advice on finding good moves. In hat connection I was interested to read last night, as I continue my investigations into Soltis's book on studying chess, his observation that even weak players find it easy to identify many moves as bad. However, they find it very hard to identify good moves. The explanation seems to be that you just need one reason to identify a move as bad and then you can prune it off at once. But for good moves there are usually several competing reasons to identify them as good and you have to juggle them all around and prioritise them in some way. Not easy. Anyway, I think this adequately explains the double sente problem without any need for deriving endgame theorems. Such moves get a high priority in the confused mind simply because you can see you get at least something out of them.

Uberdude wrote:

Bill Spight wrote:How many amateurs would answer such a play because they have been taught that it was double sente?

Not me. It's middle game not endgame.

Unless there is some special tactical aspect to the local situation, it is even less likely to be sente in the endgame. The whole point of teaching that the double kosumi on the second line is double sente is to encourage amateurs to follow the proverb and make the play early (i. e., not to wait for the endgame) because it is sente. And not only sente, but double sente, because if you don't play it (with sente) early, your opponent will.

Edit: Is it bigger than it looks? Yes. And that is what should be taught.

John Fairbairn wrote:

How many amateurs would answer such a play because they have been taught that it was double sente?

But where are amateurs being taught this is double sente? Surely only in a localised problem where there is no global override.

But then it is even worse. Sorry to pick on Kano, but he is representative. From p. 31 of the 1974 edition of his Yose Dictionary. Edit: BTW, it is obvious from the diagrams that Kano is following the usual convention and treating the Black and White stones as unconditionally alive.

Click Here To Show Diagram Code

[go]$$B Double sente -- Not!
$$ -----------------------
$$ | . . . . . . . . . O .
$$ | . . . . . a . . . O .
$$ | X X X X X . O O O O .
$$ | . . . , . . . . . , .[/go]

"a" is indeed the right play for each player. And it is a middling endgame gote. However, Kano says that it is double sente and gives only play diagrams where the opponent answers. He goes on to say that because each side makes a profit with sente, a double sente is the largest kind of boundary play.

See http://senseis.xmp.net/DoubleSenteIsRelative for a discussion of this position.

My experience is that even DDK learn for themselves for principle that an opponent's move is only sente if you answer it.

That is one meaning of sente. As they advance, they need to learn more about sente.

What they are crying out for is advice on finding good moves.

Yes. And they should be taught that the double kosumi on the second line is bigger than it looks. They should not be confused with talk about double sente.

Bill

Thanks for taking the time to go over what you must have had to repeat many times before, but I confess it's entirely over my head. But is that entirely due to me? I have to admit that I think you are giving Kano too much of a bashing. He does say the position you quote is double sente, but nowhere can I see any statement that tells me I must rush to play double sentes willy nilly. He does cite the proverb ryosente yuzurubekarazu, I know, but that and the position cited are firmly embedded in a section entitled "two-point moves" (as opposed to other sections on one-point moves, three-point moves, five point moves and so on). Japanese writers, as you know, are used to expecting readers to think for themselves (which may be a nuisance: I know I always prefer to get an American "For Dummies" style book), and in this case it doesn't seem too onerous even for me to recognise that his double sente applies only in that context.

Similarly, the excellent Mokusu Shojiten makes it plain that even in a position described as a double sente it is often the case that one side cannot necessarily expect the other to answer, adding that such plays are typically decided as part of the middle game negotiations.

I really do think the mutual damage kind of kosumi is in a different category altogether, and I personally find the attempt to drag it into the same box as Kano's kind of move just to have a pop at Kano (and then kick it out again!) is what causes confusion, at least for me.

Maybe the history of the term "double sente" in English has compromised it for the western audience, but it seems well enough presented in Japanese, to me, and there have been enough mathematically minded and well qualified Japanese amateurs contributing to Kido and the like to make me wonder why they don't see it as a problem. Maybe the difference is that the Japanese use the term as a description; westerners want to use it as a definition (or something like that).

I find the western agonising over the meaning of 'forcing moves' and sente misplaced for similar reasons.

Am I speaking for other confused souls, or is it just me?

John Fairbairn wrote:Am I speaking for other confused souls, or is it just me?

I am not confuesed, but I consider this discussion about "Double Sente" to be superfluous.

In my opinion, the typical Westerner is looking for a definition of a term that will be true for 100 % of all application cases.

And -- her comes the crux -- at no time having in mind some implicit implications, which I suppose are typical for Japanese thinking / writing.

Sente is Sente, only when there is no larger spot on the board for the opponent's move. As a matter of course, this self-evidence is NOT stated in every third or fourth sentence in a Japanese book on Go during a discussion of "Sente-related" issues.

The author supposes that the reader is able to think on his own, at least partially.

John Fairbairn wrote:Am I speaking for other confused souls, or is it just me?

I don't yet understand Bill's/lightvector's posts about a play in that example position being 20 points, but then I didn't put much effort into trying to understand it. But neither do I see much value in it: the approach I use of simple swing not miai counting of "4 points with 30 followup for white, but 4 points with 32 followup for black" has worked well enough for me so far and I understand it. Maybe there are some positions it will make me play sub-optimally if I play yose experts like Bill or professionals. In timed tournament games I rarely make detailed yose calculations because I am in byo-yomi, but on OGS my approach worked well enough that I usually made significant gains in yose (against mid-dan opponents, against Alexander Dinerstein 3p I did lose some points in yose but not too many to lose the lead, and those mistakes were mostly missing clever tactical sequences rather than miscounting sizes). So if I want to play better yose in my games I see this kind of thing as irrelevant: most important is playing the opening and middle game faster so I have time to play the endgame with my current skill instead of blitz skill, next is better reading to find better sequences, and this kind of counting is way down the list. Of course an appreciation that sente is relative is very important, but I've known that for ages and so do much weaker players: I remember around 20k(?) people would play 1st line yose "sente" hane and connects in the opening or middlegame but they soon realised they weren't sente at that stage of the game because tenuki was a good answer.

Uberdude wrote:I don't yet understand Bill's/lightvector's posts about a play in that example position being 20 points, but then I didn't put much effort into trying to understand it. But neither do I see much value in it: the approach I use of simple swing not miai counting of "4 points with 30 followup for white, but 4 points with 32 followup for black" has worked well enough for me so far and I understand it

I do understand their counting method, but I prefer your simpler approach and consider it more practical. Still, you can see that mathematically your value of 4 points gote plus followup opportunity does not make sense. The only way the true value will be 4 points is if your opponent invests another move here. If he does so immediately, your move was 4 points sente. If he does so later, your original move was perhaps gote, but it was clearly worth more than 4 points pure gote, since it earned you an extra move at some later time. Maybe we should call it temporary gote or delayed sente. Bill's counting attempts to include this added opportunity value up front, at the time the original move is played.

In my opinion, the typical Westerner is looking for a definition of a term that will be true for 100 % of all application cases.

And -- her comes the crux -- at no time having in mind some implicit implications, which I suppose are typical for Japanese thinking / writing.

To misquote Stalin:

A mathematician exist, a problem exists.
No mathematician exists, no problem exists.

For kyu players, value-dependent endgame theory is overrated; valueless endgame theory is ten times more relevant. Dan players should already be applying the valueless endgame theory so, for them, value-dependent endgame theory becomes relevant for avoiding most of their remaining mistakes. Different value-dependent endgame theories exist:

1) Territorial positional judgement: imagine a sequence of a few endgame moves, then make a (at least local) territorial positional judgement for the created, imagined position (or at least for its changes), then compare to other imagined sequences.

2) Swing (deire) counting: this works as long as considerations do not involve kos and all local endgames are either sente or gote. As soon as sente, reverse sente and gote (and possibly kos) need to be considered together, the theory fails.

3) Per excess move (teire) counting: values can always be compared but occasional exceptions exist nevertheless. If so, (1) helps.

When I hear somebody (especially a mid dan) boasting to need only (2), he is wrong and can easily become stronger by applying also (1) or (3). When I hear somebody (especially a dan) complaining that all value-dependent endgame theory (i.e., (1), (2) and (3)) would be hated applied mathematics and so better disregarded, he only confirms to have no interest whatsoever in improving. Go is a scored game and therefore value-dependent. One must not avoid value-dependent endgame theory, but one can only postpone it until one has understood and applies all the valueless endgame theory.