Lie group E8 mapped

[ewmayer]

Alright, who's the wise guy? I think we can narrow it down quite quickly:

1) David Hasselhoff admirer - that's pretty much me, akruppa and Xyzzy;
2) Nothing better to do - that rules out akruppa;
3) Super-secret "invisible edit" forum privileges - uh huh, that's what I thought.

-Ernst, a.k.a. E5

"Lie groups, damned Lie groups, and E8"

[mfgoode]

Originally Posted by Xyzzy

We thought it had something to do with chess.

I'm sorry! I'm hard of hearing - jet engines, Pratt and Whitney/Rolls Royce , you know !

yeah did you say chess or chest ?

Mally

[bdodson]

Originally Posted by ewmayer

Was watching a piece on this on ...
Neat result, but c'mon folks, no need to pull phony "practical applications" out of your collective rumps just to please the news media. It's embarrassing to watch.

Glad to have the Baywatch episode info clarified (previous, more recent post).

I've gotten through to some of the more informative links; and have a much
better idea of what they think they were doing --- not sure you'll be happier.

Several of the links (from Baez's blog entry, maxal's post) give nice colored
pictures of the Dynkin diagram I was verbalizing about. First, on the "super"
part of the supercomputer computation, they had some help from Elkies -
otherwise they'd still be waiting for a "larger supercomputer". Bob might
be pleased to hear this one, the key idea seems to have been to run the
program four times to find the solution mod m, four moduli, then patched
the solution together via the Chinese remainder thm. I was just typing
"mod p, for primes", but the actual moduli used were 251 (prime), then
253 = (11)(23), then 255 = (3)(5)(17) and, finally 256 --- sure enough,
these are relatively prime, and Chin.Rem applies. So that 77 cpu hours
was distributed over 4 computations, just over 19 hrs each. The super
part appears to have been that they had 64Gb of ram to use --- ah,
on a "FusionA8" from Western Electric. Ooops, that was supposed to have
been "nearly 3 days solid processing" on the machine, with AMD chips.
Just checked with our grad student that has an account on the machine -
claim is that at the time of the computation Red Hat was only recognizing
8 of the 16 cpus (fixed since then, with a re-installation of the Red Hat).
That would be 3*24 = 72, so a bit more than 3 "solid days" times 8 cpus,
for 616 cpu hours, err ... four computations, average 154 cpu hours each.
Ah, yeah, 8 dual cores, "DDR2 667/533/400 Memory"; 8000 series.
With 10/100/1000 ethernet onboard "networking", infiniband interconnect
support. Sounding like super memory access/data exchange to people
more familiar? The 616 >= 3*24*8 amd cpu hours would be a slow day
for ecm here, otherwise.

Anyway, Adams did an update yesterday on "what we really did". Claim
is that they expect to get the data needed to be able to check an Arthur
conjecture on residues of Eisenstein series --- they still need to identify
which "perhaps 200 of the 453,060" satisfy Arthur's condition. -Bruce

ref: http://www.liegroups.org/AIM_E8/technicaldetails.html