[Beowulf] OT: public random numbers?

[Robert G. Brown]

On Fri, 12 Aug 2011, Lux, Jim (337C) wrote:

> All nice suggestions, but I wonder if they're truly random.
>
> Scores of games have underlying patterns from the "rules of the game"  (e.g. american football games tend to have scores that are tied to the 6,7,8 or 3 points.  basketball goals are 2 or 3 points, etc.)
>
> I'm sure someone has analyzed this.
>
> I suppose one could sum a large number of scores, which would give you something with Gaussian distribution, and then you could transform it into something with uniform distribution (sort of a inverse Box-Muller).
>
> What about using random.org and it being backed-up on archive.org?  Does that give you the "multiple independent sites" desired?

As I said and repeat, nothing like this is at all random.  Random is
stuff like thermal noise, shot noise, quantum noise, and even all of
those things are distributed and not flat and require massaging to make
into uniform deviates or random bits.  Unpredictable is easy, of course
-- flip a coin, roll some dice -- until you need to make it
>>rigorously<< unpredictable and >>rigorously<< uncorrelated, at which
point you need to not screw around with weather, scores, market closing
values, even "randomly sampled" ticks of a nanosecond clock aren't that
random without some work to make them so.

I liked the lavarnd site, and I like random.org.  Hell, tap into both of
their streams, they're both practically perfect as sources of random
numbers go, and it gives you your redundancy and you can xor their
streams together to get yet another irrelevant and probably unnecessary
degree of lack of correlation.  Even if one stream is subtley correlated
and the other is too, the chances of the correlations "matching" and
persisting through an xor process are astronomical.  But then, finding
correlations in the output of a properly seeded crypto prng is pretty
astronomically unlikely BEFORE you xor-fold it stream-wise a few dozen
times into a source of real entropy like atmospheric noise or
electro-optical noise.

If you want something better, you'll probably have to explain your
application in a bit more detail.  Do you need rigorously random and
flat numbers, or just something unpredictable?  The latter is cheap and
easy and can be done in the privacy of your own home by reading from
/dev/random or /dev/urandom (or perhaps from Intel's new on-CPU rngs).
The former requires theory and some work and some heavy duty empirical
testing.

Just remember, numbers are not random.  Numbers are numbers.  The
number 7 could be "random" or not not by its nature but by how the 7
was generated.

Processes, in other words, are (approximately, oxymoronically) random.
If you want random numbers, find a (mathematically provably) "random"
process, at least to some order and for some purposes...

    rgb

>
>
> ________________________________________
> From: beowulf-bounces at beowulf.org [beowulf-bounces at beowulf.org] On Behalf Of David Mathog [mathog at caltech.edu]
> Sent: Friday, August 12, 2011 08:58
> To: Peter St. John; beowulf at beowulf.org
> Subject: Re: [Beowulf] OT: public random numbers?
>
> Peter St. John wrote:
>> But at any given time, a group can agree on (say) the lowest significant
>> digits of the temperatures at time T in cities X, Y, and Z as reported at
>> time T2 by the NWS.
>
> Actually we don't know that, at least not reliably enough for this
> purpose.  It may be that the one web address is actually multiple
> servers, and if the NWS pushes out data revisions these could return
> different results for T:X,Y,Z at T2 if the servers were not strictly
> synchronized.  Never mind the caching problems that revisions like this
> would create on browsers.  I have no idea if the NWS revises their data
> files, but it would not be surprising if they did.
>
> After posting I thought of one other source of more or less random
> verifiable numbers - the scores of sporting events.  These are not
> always generated every day, and are seasonal for the various sports.
> They are however highly verifiable and when multiple events are grouped,
> pretty much impossible to "fix" to preselected digits.  For instance:
>
>  http://www.nfl.com/scores
>  http://mlb.mlb.com/mlb/scoreboard
>  http://scores.espn.go.com/nba/scoreboard?date=20110304
>
> These sites maintain historical records.  Even if they didn't the scores
> are widely published, and there are tens of thousands of witnesses to
> the original event, so it would be pretty much impossible to
> intentionally change a final score.  There could still be copying/typo
> errors from site to site though, but if such an error was discovered it
> would be easy enough to resolve.  There is no intrinsic order to the
> scores, and some scheduled games might be canceled, so it would have to
> be something like "sort the scores from all NBA teams who played on
> 4/4/11 into ascending order and concatenate the digits".
>
> Regards,
>
> David Mathog
> mathog at caltech.edu
> Manager, Sequence Analysis Facility, Biology Division, Caltech
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org sponsored by Penguin Computing
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>

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