r250 --> r1270?

r250 --> r1270?

[H. G. Katzgraber]

Hello,

I have noticed that in the random number generator selection 
you list r250. Would it be possible to add r1279? It is 
almost as fast as r250, yet r250 is known to give wrong 
results.

Best regards, Helmut Katzgraber.
_____________________________________________________________
Helmut G. Katzgraber                    http://katzgraber.org
Theoretische Physik, ETH Hoenggerberg   Tel (+41) 44 633 3580
CH-8093 Zurich, Switzerland             Fax (+41) 44 633 1115


This email is brought to you by the letters D and K, and the
number 6.

Re: r250 --> r1270?

[Robert G. Brown]

nOn Sat, 11 Feb 2006, H. G. Katzgraber wrote:

>
> Hello,
>
> I have noticed that in the random number generator selection you list r250. 
> Would it be possible to add r1279? It is almost as fast as r250, yet r250 is 
> known to give wrong results.

It is very easy to wrap any generator you know of in the gsl template
for rng's and add it to the list of generators that are already there
for your OWN purposes.  However, "most" of the generators there are
there for historical purposes, for testing purposes, or to enable older
code to be ported, not because using them (especially in new code) is
supposed to be a "good" idea.  I believe that the current default is one
of the ranlux generators; the older one a mersenne twister, both are
excellent -- quite fast and very high quality (according to e.g. diehard
and sts tests).

That said, I think that it is good to add still more to the GSL >>for<<
testing purposes.  I personally would like to see the GSL have "all" the
significant (named, published, used) generators built in and selectable
at runtime.

I've been working on a GSL-integrated random number tester written in C
that incorporates many of the diehard tests, some of the sts tests, and
a few tests I've written myself.  Eventually I hope to have all of
diehard, sts, and knuth's tests that aren't already in one or the other
(plus a few of my own).  In it I added both the /dev/random and
/dev/urandom tests to diehard as described above, as well as an "empty"
random number generator that is there to aid in timing rngs.  If you
look at the code, you can see pretty easily how to do the wrapping.  I'd
like to see all of these added as well -- there are actually good
reasons to use the entropy-based generators in e.g. cryptographic
applications to generate a reliable seed for the diehard rngs.

This is a work in progress, but you can get what should be functional
code from the snapshot here:

   http://www.phy.duke.edu/~rgb/General/dieharder.php

     rgb

>
> Best regards, Helmut Katzgraber.
> _____________________________________________________________
> Helmut G. Katzgraber                    http://katzgraber.org
> Theoretische Physik, ETH Hoenggerberg   Tel (+41) 44 633 3580
> CH-8093 Zurich, Switzerland             Fax (+41) 44 633 1115
>
>
> This email is brought to you by the letters D and K, and the
> number 6.
>

-- 
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@phy.duke.edu

Re: r250 --> r1270?

[Heiko Bauke]

Hi,

On Sat, 11 Feb 2006 16:22:05 +0100 (CET)
"H. G. Katzgraber" <katzgraber@phys.ethz.ch> wrote:

> I have noticed that in the random number generator selection 
> you list r250. Would it be possible to add r1279? It is 
> almost as fast as r250, yet r250 is known to give wrong 
> results.

all generators of the lagged fibonacci type with xor, addition or
subtraction do have the same statistical defects [1]. For that reason
r1270 will give wrong results in all MC simulations where r250 is known
to give wrong results. It just takes a little bit longer (in terms of
MC steps).


	Heiko


[1] http://de.arxiv.org/abs/cond-mat/0307138 

-- 
-- Ich habe ein einfaches Rezept, um fit zu bleiben - 
-- Ich laufe jeden Tag Amok. (Hildegard Knef)
-- Supercomputing in Magdeburg @ http://tina.nat.uni-magdeburg.de
--                 Heiko Bauke @ http://www.uni-magdeburg.de/bauke

Re: r250 --> r1270?

[H. G. Katzgraber]

Hello,

Thanks a lot for the information. I looked also at dieharder 
which I will try to compile and play with.

Best wishes, Helmut.
_____________________________________________________________
Helmut G. Katzgraber                    http://katzgraber.org
Theoretische Physik, ETH Hoenggerberg   Tel (+41) 44 633 3580
CH-8093 Zurich, Switzerland             Fax (+41) 44 633 1115


This email is brought to you by the letters Y and I, and the
number 2.