Post by Michaelhttp://www.bgonline.org/forums/webbbs_config.pl?noframes;read=203042#203042
Thanks for the pointer.
I did not read through the whole thread. However, I gather that your main
question is whether the standard "live cube takepoint" calculation is
"mathematically correct."
The term "mathematically correct" is more subtle than I think you recognize.
Let's start by stating something that is perhaps obvious but that needs to
be emphasized in any careful discussion of this type: Backgammon is not
mathematically solved. So if by "mathematically correct" we mean that the
live cube takepoint calculation yields an absolutely mathematically correct
rule for perfect backgammon play, then it is certainly not "mathematically
correct," for the simple reason that nothing is mathematically correct in
this very strong sense.
In fact, if you really want to speak precisely, even the concept of a "take
point" is not "mathematically correct." If we spell out the assumptions that
are usually left unstated, the term "take point" presupposes that if you are
given the following information---
1. the match score and the cube location, and
2. the percentages for wins, gammons, and backgammons for both sides
---then there exists a number, called the "take point," with the property
that if the win rate exceeds the take point then the correct cube action
is to take, and if the win rate falls below the take point then the correct
cube action is to drop. But this assumption is not "mathematically correct."
Whether the correct action is a take or a pass depends on the details of
the position, and not just on the items listed above. Mathematically speaking,
therefore, there is *no such thing as a take point*.
Take point calculations are *heuristics*. By a "heuristic" I mean a rule of
thumb that does not always hold, but that is relatively easy to apply and
that generally provides good advice about how to play well. We all know a
lot of backgammon heuristics---making the 5pt is usually better than making
the bar point, race when ahead in the race, double in a straight race when
you're 10% ahead in the pip count, etc. With the exception of some heuristics
for non-contact positions, none of these are mathematically proven to be 100%
correct. They're just rules of thumb.
Because take point calculations involve more arithmetic than most heuristics,
they may give the impression of being "mathematically correct." But don't
let the arithmetic fool you. Take point calculations are nothing more than
sophisticated rules of thumb, like JOH's rule of 9/36 market losing sequences.
My point is that asking whether a heuristic is "correct" is the wrong question.
The only appropriate question is whether the heuristic works well, and whether
the assumptions underlying the calculation are believable.
O.K., so let's say we understand all this, and we understand that the take
point calculation can't be "correct." We can still ask, who came up with
the calculation and why did they think it would yield a good heuristic?
The way people have approached this sort of question in the past is that
they have developed *models* of backgammon. That is, they invent some kind
of mathematical process that is simpler than real backgammon but that
resembles backgammon in certain ways. They then analyze the simplified
mathematical model, and use that to obtain a heuristic for real backgammon.
The simplest model for backgammon involves a *continuously* varying equity
that moves up and down randomly. In this model you get a 20% take point:
http://www.bkgm.com/articles/KeelerSpencer/OptimalDoublingInBackgammon/
As Keeler and Spencer discuss, this model is oversimplified in many ways.
So others have tried to come up with more complicated models that are closer
to real backgammon but that can still be analyzed. For a recent attempt of
this sort, see:
https://arxiv.org/pdf/1203.5692.pdf
I'm not sure where the particular formula that you asked about on BGO comes
from originally. It might be from Janowski---
http://www.bkgm.com/articles/Janowski/cubeformulae.pdf
---but I haven't read the paper carefully enough to tell whether it describes
exactly the formula you're interested in.
---
Tim Chow