From: jim@mentat.com (Jim Gillogly)
Date: Mon, 18 May 1998 16:23:06 -0700 (PDT)
To: cypherpunks@toad.com
Subject: GFSR initialization? [Re: National Atomic Museum]
Message-ID: <9805182322.AA27897@mentat.com>
MIME-Version: 1.0
Content-Type: text/plain


[Bill Payne's's mailing list pruned back to cypherpunks only]
> Ted Lewis told me that they now use the gfsr at LANL for its nuclear
> bomb=20
> simulations.
> 
>   http://av.yahoo.com/bin/query?p=3Dgfsr&hc=3D0&hs=3D0=20
> 
> But this is how the mess got started.  Me making contact  with a
> Japanese=20
> professor who developed a new method for selecting the binary seed
> matrix
> for the gfsr.

Is that to speed it up, so you don't have to initialize it by spinning
through a bunch of initial iterations?  Turns out there's a time-memory
tradeoff, where if you do a recursion relationship on where you want it
to start, you can set it up in log2 time and come out with the same
starting point you would have by spinning: the recursion equations
cancel nicely and it boils down to something manageable.  I think other
people have initialized it with another PRNG with good success.

I <know> that's not classified, since I came up with the recursion
shortly after the Lewis & Payne paper on GFSR came out -- JACM, wasn't
it?  I never got around to writing it up.

	Jim Gillogly