Re: update to sphere.c in randist

[James Theiler <jt@nis.lanl.gov>]

Hi Mark,
  I added a few more things to randist this weekend, a fast algorithm for
arbitrary discrete random variables and a subroutine to produce variates
only on the tail of a gaussian.  These are documented in the ChangeLog,
and in the cvs log. 
  I also updated the random.texi file, but I've still not got around to
setting things up so that I can tex those files, so I couldn't check that
the braces all match, etc.  My apologies if I "broke" the file, but it
should be easy to fix.  I have an old email that you sent that tells me
how to do that, I just haven't gotten around to it.  (When Larry Wall
said that laziness was a virtue in a programmer, I don't think this is
what he was talking about...)
  
]     James> interface.  i'm not sure i'll always be this disciplined,
]     James> but when adding a new routine to gsl, i try to be generous
]     James> with comments, but mostly i try to have a reference that i
]     James> can cite, so the algorithm is less of a black box.
] 
] That is very important.  In some cases it is maybe better to not use a
] source in the first place, but rather develop the details of the
] algorithm from scratch.

Absolutely.  Original code is important for copyright purposes, but
pointers to the numerical literature is important for intellectual
reasons.  One, people will trust the algorithm more, and two,
people who don't trust us will have something to check against, 
and are more likely to correct our mistakes.

] 
] For example in the case of numerical recipes where they agressively
] enforce their copyright (even unreasonably) and their algorithms are
] frequently expressed as C or FORTRAN code.  It's a pity Knuth didn't
] cover the whole numerical field :-)

Knuth's book is a funny one.  The 3rd ed was published in 1997, and though
it is technically up to date (insofar as I'm qualified to judge), there is
an old-fashioned feel to it. 

] 
]     James> my vision of gsl is not so much a comprehensive math
]     James> library (though it is rapidly heading in that direction, to
]     James> the credit of its developers) as a place to go to get some
]     James> good numerical source code, unencumbered by restrictive
]     James> copyright, which can then be modified to the particular
]     James> project you are working on.
] 
] That is the original motivation, but Brian also suggested way back
] that GSL should also be a ready-to-drop-in package for free
] mathematical software so that they can shed the maintainance burden

yes, both.  and so far, and in my opinion quite impressively, it does do
both.  we need to keep both goals in mind as we maintain it though, and
for instance strive to minimize the interdependencies. 

] onto us.  Just a couple of days ago someone wrote to tell me that the
] Octave team is seriously considering putting GSL into Octave
] (presumably replacing some of their infrastructure).

cool!

regards,
jt

---------------------------------------------
James Theiler                     jt@lanl.gov
MS-D436, NIS-2, LANL        tel: 505/665-5682
Los Alamos, NM 87545        fax: 505/665-4414
----- Space and Remote Sensing Sciences -----