Re: Reversible Computation and Experience

From: Robert J. Bradbury (bradbury@aeiveos.com)
Date: Thu May 10 2001 - 23:07:42 MDT


Hal wrote:
> The model of reversible computation I was using is due to Bennett and is
> what he calls "Brownian" computation. A good article by him is on
> Robert Bradbury's site at [snip]

I don't think that is a good model. The Brownian motion aspect
of reversible computation is a way to use the latent heat in
material to drive forward a computation. For example all chemical
reactions are inherently reversible. You can eliminate adding extra
heat to a reaction by using the latent heat in the environment and the
fact that the probability distribution of the energy of individual molecules
will mean that some of them will have more than the required activation
energy to drive the reaction forward. However you may have to wait
a long time to get significant numbers of molecules to go through
the transition state. You still lose energy here because the
product molecules have to give the transition-state energy back
to the environment.

The Fredkin/Toffoli "billiard ball" reversible computation model is
a better analogy. The real point you have to understand is that
it isn't how you do the computation (you can do the computing for *free*
(e.g. Brownian motion as a driver)) -- its whether or not you erase
bits! The is the very counterintuitive result of all of the physics
of computation work done by Landauer, Bennett, et al.

So when you do reversible computation you must save all the intermediate
states and then roll them back to their original state after you have
computed your result. The majority of the heat you produce using this method
is produced in the erased bits of the final result (there are of course
some losses in all the intermediate steps and the trick will be getting these
to the smallest levels possible).

I believe Michael Frank has a paper that determines the size a reversible
computer must be to trump any non-reversible computer (its *very* big).

What isn't clear is whether you are always going to pay a "real time"
penalty by shifting to reversible computation (note that Eric's
rod-logic reversible computers only function ~1 GHz and we will probably
be pushing 10-15GHz in non-reversible architectures before we make
the transition to reversible architectures). Of course what you
sacrifice in "real time" you get back (several times over) in parallelism.

Now, of course the interesting thing about this is that if you are
always saving the "intermediate" states of your consciousness in this
very long reversible computation and you get the results and do a
roll back it seems in some strange way that you have killed all the
copies...

Robert

(Also, if anyone tries to get to my site (www.aeiveos.com) and ends up
at a site that seems to be in France, could they please let me know
off-list; I seem to be having some very strange DNS problems and can't
quite figure out if its me or the net.)



This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:07:34 MST