From: Alfredo Ferrari (alfredo.ferrari@cern.ch)
Date: Fri Nov 03 2006 - 09:40:17 CET
Joe
You completely misinterpret the double precision issue!!! I take issue
with your comment (I am kidding), you didn't think about!
Double precision is indeed fundamental to get a MC working and it has
nothing to do with the accuracy you can achieve on physics results.
Please try to do the following things in single precision:
a) a Lorentz transformation at 7 TeV for a pi
b) a minimum q^2 calculation in e- bremsstrahlung at 2 GeV
c) tracking with a particle incident at less than 1 mrad from a surface
d) ionizations (eV's) by heavy ions (tens of GeV in mass only)
and I could go on for four pages or even more.
BTW, some processes like b) don't even work in double precision unless
special series expansions are taken care of, instead of using naively
formulae. c) is the most critical one, even if you accept all kind of
inconsistencies from physics (ie particles below mass because of imprecise
Lorentz transforms, negative square roots everywhere for inaccurate
q^2's etc etc), YOU WILL NEVER SUCCESSFULLY TRACK A CHARGED PARTICLE
IN WHICHEVER GEOMETRY UNLESS YOU ARE SELFCONSISTENT AT LEAST UP TO
THE DOUBLE PRECISION LEVEL. Charged particle geometry tracking is all
about a proper management of rounding in the machine, it already requires
enormuous efforts to get it right in double precision, single would be
simply impossible (unless you limit yourself by switching off multiple
scattering and restricting to almost normal incidence)
BTW, I don't know of any modern MC which doesn't use double precision as
the default, including G4 which being in C++ has what we call "double
precision" as the default.
Ciao!
Alfredo
On Thu, 2 Nov 2006, Joseph Comfort wrote:
> Hi Francesco,
>
> Thank you for the suggestion. It worked. And I have a better
> understanding of how to prepare geometries.
>
> On another topic, the message about double precision that was sent out
> by someone earlier rather irritated me, and I can't get over it. We
> need to keep in mind that the data upon which the models are being built
> are seldom (if ever) better than 1% or so, absolute. If a model can get
> within a few percent (3-4 significant digits) of good data, it is doing
> well. This is just a comment and I don't mean to get into arguments,
> nor to lose any friendships.
>
> Thank you,
> Joe
>
-- +----------------------------------------------------------------------------+ | Alfredo Ferrari || Tel.: +41.22.767.6119 | | CERN-AB || Fax.: +41.22.767.7555 | | 1211 Geneva 23 || e-mail: Alfredo.Ferrari@cern.ch | | Switzerland || Alfredo.Ferrari@mi.infn.it | +----------------------------------------------------------------------------+
This archive was generated by hypermail 2.1.6 : Fri Nov 03 2006 - 10:00:26 CET