[Beowulf] The recently solved Lie Group problem E8

Peter St. John peter.st.john at gmail.com
Thu Mar 22 08:09:54 PDT 2007


First, a picky but pertinent point: Fermat's Last Theorem wasn't done by
machine. It was Andrew Wiles, at Princeton. The story is that he broke up a
major work into Least Publishable Units, and set to work on FLT in secret,
submitting a LPU once in awhile so his colleagues woudn't be suspicious.
Then after announcing the result at the Newton Institute in England,
wrinkles were discovered, and there was a couple years of intense labor,
speculation, and unprecedended (for us) media coverage. The final work is
two huge book-sized papers which would prerequire a couple years of
Arithmetic Algebraic Geometry to be able to read, that is, few of us other
than AAGeometers will ever be able to read it.

Yet this daunting thing was produced by humans, not machine. It's still
virtually intractable.

The Classification of Finite Simple Groups project is a modern legend. The
finished result is in **thousands** of papers published by hundreds of
algebarists, and last I heard some bits were still being combed over but
it's generally accepted as a Done Deal (there is a new project to try and
simplify it). The largest of the "Sporadic" Finite Simple Groups is the
Monster; it's order has factors of all the primes, 2, 3, 5, up to 31.
consecutively, then with 5 more prime factors skipping a bit among 37
through 67. The total order is too big a number to describe meaningfully in
casual email (the exponent of 2 is a double-digit number, for starters).

We don't need computers to do this kind of harrowing work; computers just
make it easier :-)

For centuries (since Euler) there was an open problem, The Four Color Map
Theorem (or Conjecture). How many colors do you need to color a map so that
countries that share a border (longer than a single point) have different
colors? The answer is 4. Cartographers knew this for millenia; they could
color maps that way, and it matters alot to printers. But Euler himself
could not prove it. He did classify all graphs (maps reduce to graphs; an
edge between nodes is a border between countries) into a short list of
special cases that would be sufficient to prove the theorem. Unfortunately
many of those special cases were so huge that nobody could color them by
hand. It took a computer, and this was done in the late 70's, at
Northwestern I think, when Robert and I were students, and maybe we both
attended that talk at Duke.

That result brought Robert's concern out in the open: do we believe a proof
that is too long for us to personally verify ourselves?  With Wiles and the
group classification thing and some other such work, we've realized that
machines aren't required. It is a practical impossibility for me to verify
Wiles' work; in principle, maybe I could in, say, a decade. To verify the 4
color theorem would take me centuries. But really that's just a matter of
degree.

What came out of the soul-searching is that we believe the process that
generated the result (which may involve code, hardware, peer-review,
reproducible measurements....) instead of the pile of generated paper. I can
read the source code at Northwestern, verify that, read over Euler's list,
verify that,. and then know that the correct algorithm applied to the
correct material will give a correct result.; Then I could run it on my own
machine to check hardware did not affect the result.

It made a lot of us very uncomfortable, but we left behind the day when
everyone could read everything for himself, sometime between Guttenburg and
Leibnitz's Monadology.

Peter



On 3/22/07, Robert G. Brown <rgb at phy.duke.edu> wrote:
>
> On Thu, 22 Mar 2007, Peter St. John wrote:
>
> > Well to me that's the point. My brain is too small for 500Kx500K
> matrices
> > over a ring of 22 degree polynomials, too. So we throw a 16-node
> computer at
> > it and crush it under the hobnailed jack-boots of Higher Mathematics.
> > I wish I know more about the SAGE (machine) that hosts the SAGE
> (software)
> > that was used for this, but apparently washington.edu's web server can't
> > handle the CNN exposure as well as their number cruncher can crunch
> numbers.
> > They are still down.
>
> In the case of E8, using a computer is probably necessary, although one
> would require a wetware interface to make the slightest bit of "sense"
> out of the results anyway.
>
> I do find this trend depressing, though.  Fermat's lost theorem proven
> using computing -- nothing elegant, just crush it underfoot, as you say.
> E8.  Next we'll hear that the Goldbach Conjecture is finally proven by
> virtue of solving 10^17 specific cases and exploiting a proof that once
> you have all of those cases proven you can iterate the result to
> infinity somehow, or we'll hear of the Riemann Hypothesis being solved
> this way -- nothing elegant, nothing that is (actually) of any USE.  We
> all know that these are true anyway, at least as well as we know that
> the theory of gravitation is true.  In neither case can they be proven
> (yet, in the case of the math, never in the case of gravity), in both
> cases they are known beyond any reasonable doubt via induction (see
> Polya's lovely books on induction and mathematical reasoning).  Proving
> these things by computer adds nothing to this -- in addition to the near
> impossibility of actually judging the computational results (deep bugs
> remaining a ubiquitous possibility in ALL complex computer code) which
> always leaves a sliver of doubt even then, that doubt (expressed as
> Jaynesian/Bayesian "degree of belief" on an information
> theoretic/entropic basis) is already so small that it hardly changes on
> a log scale from having done an exhaustive computation.
>
> In the SPECIFIC case of E8 that isn't quite true.  Since string theory
> as a theory of everything (TOE) may be covered by E8, and since string
> theory is reportedly insanely complex and so big that exploring it to
> find the RIGHT decomposition into whatever \times SU(whatever) by hand
> might take lifetimes, it is barely possible that being able to enumerate
> it even electronically will permit a systematic search to be performed
> that can eliminate huge blocks of the possibilities and home in on what
> we can at least HOPE is a small set of decompositions.  Ideally a
> single, unique decomposition.
>
> That would actually be pretty cool.  For the first time in pretty much
> forever, we'd have an actual CANDIDATE TOE, and yet another important
> step in "the end of physics" will have occurred.  (And note well the
> quotes, please -- I'm not suggesting that physics research will come
> close to ending with a TOE, only that it will finally have a firm known
> basic foundation.
>
>    rgb
>
> > Peter
> >
> >
> > On 3/22/07, Robert G. Brown <rgb at phy.duke.edu> wrote:
> >>
> >> On Wed, 21 Mar 2007, Peter St. John wrote:
> >>
> >> > Times have sure changed; with Wiles and Fermat's Last Theorm in
> >> newspapers
> >> > for over a year, then "A Beautiful Mind" from Hollywood; it's almost
> not
> >> > surprising that the solution of a difficult math problem is mentioned
> at
> >> > CNN.com.
> >> >
> >> > The Exceptional Lie Group E8 computation just got done (some info at
> >> > http://www.aimath.org/E8/computerdetails.html about the details of
> the
> >> > computation itself). Reference to the system SAGE is a bit ambiguous;
> >> it's
> >> > the name of a symbolic mathematics package and apparently also a
> 16-node
> >> > system at the same University of Washington. Natually I was curious
> >> about
> >> > the computer, but ironically, it seems that while they can handle a
> >> matrix
> >> > with half a million rows and colums each (and each entry is a
> polynomial
> >> of
> >> > degree up to 22, with 7 digit coeficients), their departmental web
> >> server
> >> > can't handle the load of all of CNN's readership browsing at once :-)
> >> >
> >> > The group E8 itself, together with some explanation of the recent
> news,
> >> is
> >> > in wiki, http://en.wikipedia.org/wiki/E8_%28mathematics%29
> >> >
> >> > Dr Brown might explain better than I could how sometimes the best way
> to
> >> > understand a thing is to break it down into simple groups of
> symmetries.
> >>
> >> Don't you be puttin' that off on me now.  I get off of that particular
> >> bus somewhere around the SU(N) stop, with rare excursions over into
> >> point groups on the other side of the tracks.  Unitary, yes.
> >> Orthogonal, why not.  SL(2,C) even.  Strictly UNexceptional.
> >>
> >> > Apparently, one of the funky things about E8 is that the "easiest way
> to
> >> > understand it" is itself.
> >>
> >> Yeah, and like I have a brain that can manage ~500,000x500,000
> >> complicated polynomial objects.  Thanks, I think... but not.
> >>
> >>    rgb
> >>
> >> >
> >> > Peter
> >> >
> >>
> >> --
> >> Robert G. Brown                        http://www.phy.duke.edu/~rgb/
> >> Duke University Dept. of Physics, Box 90305
> >> Durham, N.C. 27708-0305
> >> Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu
> >>
> >>
> >>
> >
>
> --
> Robert G. Brown                        http://www.phy.duke.edu/~rgb/
> Duke University Dept. of Physics, Box 90305
> Durham, N.C. 27708-0305
> Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.scyld.com/pipermail/beowulf/attachments/20070322/c49cd8ae/attachment.html


More information about the Beowulf mailing list