comic | episodes & e-books | store | about
< previous post | next post > | all news from May 2013 News archive | News search | RSS
  Quantum annealing I almost getMay 20, 2013 11:40 PM PDT | url
Added 2 new A* pages:An A* reader who actually knows stuff (unlike me who just reads stuff in articles that he thinks sounds cool) wrote me to clear up some of my confusion over that quantum annealing stuff that the new D-Wave computer NASA is getting is going to be doing with its qubits:

...quantum annealing is a variant of a regular computer optimization technique called simulated annealing.
Basically you start with a "landscape" (though it's multi-dimensional) with hills and valleys in it. In this you have the equivalent of a little ball that rolls downhill. The trick is a bit like one of those little puzzles with a ball in it where you shake it around and try to get the ball into a particular hole, except that the holes are different depths and you want to get the ball into the deepest one (which is your solution).
In regular simulated annealing, the algorithm "shakes" the box hard to start with, and less and less as it goes on, which *usually* results in the ball falling into the deepest valley, or at least one of the deeper ones. (In real metal annealing, the "shaking" is done by raising the temperature and slowly decreasing it (causing the atoms to settle into neat regular arrays, so in simulated annealing they refer to this as a temperature as well.)
In the quantum version, they don't make the ball travel *over* the hills in the landscape, it just tunnels through them, and they gradually reduce the area over which it tries to tunnel. But they're still looking to see which valley the ball landed in.
Okay, that's an oversimplification, but is roughly what's happening.
The main thing is that the D-Wave computer isn't a *real* quantum computer, exactly, it's only useful for solving this specific type of optimization problem, as far as I understand it.

And I had seen a chart of that over on Wikipedia. So I said all right but what is this fancy equationing actually doing, and the knowledgeable reader wrote back:

Oh, that's pretty simple. It's common optimization problems, with the X and Y (and other dimensions) representing the values of the parameters (e.g. airspeed and control surface position for a plane) that can be varied in the optimization process, and the Z dimension being the measure of how good the result is for a particular set of parameter values (for a plane design, that might be drag, or lift, or the size of vortices produced by a wing, or some combination of those).
I would guess NASA is interested in using the computer to optimize airflow over vehicles, or rocket engine thrust, or complex-but-efficient "slingshot" orbits, or similar problems. Since it's Ames that's getting the machine, I'd guess it's most likely for aerodynamics stuff, since that's been their specialty. Orbital mechanics would mostly likely be JPL, and engines would be Hunstsville, I think. Maybe they're trying to figure out a better scramjet design….
I don't really understand how the D-Wave machine works; my (limited) knowledge of quantum computing is of the kind of quantum computer that's a bit more like a general purpose computer, only it works by running all possible values through a computation at once, and then tries to pick out the "right" inputs that give the desired result. In other words, kinda like magic…. "Normal" quantum computers have devices that are sort of recognizably similar to logic gates in normal computers, I don't think the D-Wave machine does. Instead, I think it sets up a quantum physical simulation of the "landscape" somehow, and tries to measure where the "ball" comes to rest in it. Basically it's more like an old-style analog computer, I think.

Huzzah for putting it in layman's terms, I can almost wrap my head around that! I'm fortunate to have super-smart readers. : )
Original version of this page--kinda like the shading of the face but the facial features and body language were off:
< previous post | next post > | all news from May 2013 News archive | News search | RSS
© Copyright 2019 Ben Chamberlain. All rights reserved. | Privacy Policy