[FRIAM] Fredkin/Toffoli, Reversibility and Adiabatic Computing.

[Santafe]

I see: 

> On Jan 12, 2025, at 8:13, steve smith <sasmyth at swcp.com> wrote:
> 
> I have imaginated that the value of reversibility in energy consumption is that to "clear a computation" (dispose o the slag) the obvious answer is to simply "uncompute" the computation... thereby (only?) *doubling* the computational time?   Of course the "readout" of the state of the  "halted" computation is it's own bit-burning exercise...  

That’s nice, and as I hear you say it, was probably the right way to think about it all along, which I managed to forget over the years since first being told. 

If answers have few bits compared to the questions from which they are distilled, even if there is dissipation in making a copy of the answer onto a new channel, leaving the original to support the undoing of the computation, that is still a much smaller cost than resetting a larger number of slag bits.

It also has the effect of leaving the Source of Questions ensemble unchanged, to be sampled by the next questioner.

There is probably a relation in there somewhere, between keeping a record to not ask the same question twice, and finding a recycling route for slag that doesn’t include the answer-output bits, so that in whatever sense we really conserve representation space, the same question isn’t available to be asked any more, and only different questions get asked in the future.

The resulting optimized cost should presumably be demonstrably the same by either of these protocols, if the concept of an ideal least-cost for a collection of questions is consistent.

Thank you for that Steve,

Eric