[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] Physics in Go
Re: [gnugo-devel] Physics in Go
Mon, 7 Feb 2005 15:28:25 +0000
Le Vendredi 4 Février 2005 21:44, kai brommann a écrit :
> hello Alain and all GNU Go friends,
> first of all nice to hear that you like the XiStrat
Hmm, honestly, i read all the pages about physics ... but
i did not discover really XiStrat project (i do it know ;)
> > The authors says that Partial derivatives equations
> > should
> > give some interesting results... I m on the way :)
> well, we can share the Nobel prize then.
Yes, half a Noble Prize is enough for me, but it
will probably take a few weeks, maybe more ;)
Maybe should we begin by one paper ? In the optimistic case
that could help me to go back to university/research ... but
we ll see ...
> only how can
> we impress the venture capitalists enough, so they
> sponsor us now?!
Hmm that was not my problem: i m in Sénégal (so i have only
intermittent connection in cyber-centre), _but_ taking some time here
doesnt cost me a lot of money :) But if we can have some money for working on
, why not ? :)
> it's the problem of chicken and egg,
Maybe quantum physics will help us: we have a superposition of half an egg
+ half a chicken, that enough for cooking.
> somehow it has to start.
> > I reiterate my demand for some article making a
> > parallel between physics, and Go game
> > ... lot of usefull mathematical tools to get there
> if you find something, please don't hesitate to tell.
> actually once one starts to look at the mathematics
> and tries to use it, one sees its current limitations.
Yes. I m advancing in a "classical fluid physics" approach and
i know that my tools and knowledge are limited to that.
I have nearly finished a draft, which is in a quite good shape, much
far than my first hope, and i m progressing (on a paper sheet)
toward implementation. Things will be more clear in some days, but i
just discover the code of Gnu Go, so sometimes i get lost in weird variations
I have also some other qualitative thoughts , like what is on your site. I
of writing some complements and send them to you:
Go game is non-relativist: The information of a schicho breaker propagate
instantaneously at the other side of the universe.
Hmm, the Noble prize is still far ;)
> at the moment we (the XiStrat group) particularly
> focus on the Chess variant, the 2-body system (both
> pieces of type A for now) is maturing and the next
> release is only some months away (hopefully).
I don't know all that (i ll take a tour in your site)
For me N-Body problem is in astrophysics N body in Newton
gravitation interactions, with very bad chaotics equations
(all bodys interact with each other, no distance limit)
and it is a very huge number-crunching problem, solved
efficiently on clusters , for millions bodies , since some years now.
( at jpl or ncsa or gsfc or some other nice crazy top lab on huge
That give me hope for a future "parrallelisable/vectorisable" engine
in Gnugo, where we have only few hundreds stones interacting all
with each other, without distance limitation...
Lets says in 2010 size well on 10 000 CPU -10 000 GO Ram, and do
(like i read in TODO or Gunnar(?) mail) some automatic auto tunning
in crunching all the pro games from the old ages till now :)
> resemblance to Physics is existing in the game of Go
> no less.
> of course one cannot simply take complete physical
> theories 1:1 over to the strategy games though,
Yes, that's the problem.
> instead one probably has to invest more time to
> engineer a concise mathematical apparatus delivering
> results exactly at the necessary level of detail.
Yes I had some problems for modelling interactions,
and the shape of groups is very important (incredible isnt it ;)
For the moment i have found a nice simplication:
- in perfect fluid, atoms are tiny sphere
- in go game, groups are tiny sphere !!!!
This is classical approach, ~150 years old math, very well known
in both theory and quite efficient computer implementation.
For taking into account the shape of a group, i have
the "poetic feeling" that quantum mecanics descrition of a bunch
of particles _is_ a very good tools:
- shape resonnance, superposition of states, decoherency of
an intricated state
quantic geometry (the corner is a double border effect, if you cut the corner
and link 2-1 to 1-2, suppressing the 2lib-corner-point and having only
3lib-border-points, that change the topology and all the life and death)
Bose-Enstein condensed state ...
But for me that's poetry, intuition, i havent studied all the mathematical
> let's hope that there won't be a bad surprise in the
> end: all stuff translated into a question about
> algebraic topology, forms, finite groups of Lie type
> and whatsoever, only to realize that still not enough
> complexity has been replaced by insight. needless to
> say, it's really a very vast area to explore.
> in some sense the situation is like the topic of
> systems of linear equations 500 years ago, exemplaric
> case studies in place, but a general theory missing.
Yes that a good thing for a Noble Prize ;)
> in 100 years they will laugh at us.
Maybe, but i bet you one chicken they will remember the wonderfull
> well, GNU Go is designed to be of practical value as
> of today, having quite a remarkable strength already.
Yes That is very surprising for me: despite being nearly
completely stupid, it Gnu Go is quite strong, and all the
very tricky job seems done (ko, semeai...)
> theoretical models could deliver rules to follow in an
> engine, no question about that.
> there can be just the final word of wisdom:
> Es gibt nix gutes, ausser
> man tut es :-)
> in this sense: happy programming to all of us
Happy coding, thanks for your answer, and ...let's go on.