« Reading List: Plastic Fantastic | Main | Reading List: Under The Cloud »

Wednesday, August 25, 2010

Effects of possible solar influence on radioactive decay rates on HotBits-generated random sequences

Since the publication of a Stanford press release on August 23rd, 2010 on a possible connection between solar phenomena and the rate of decay of radioactive isotopes in laboratories on Earth, several people have written to ask what consequences, if any, such an effect might have on the randomness of data generated by Fourmilab's HotBits radioactive random number generator. The “executive summary” is “none whatsoever”, but let's look into the matter in some more detail.

One doesn't do science via press releases, but rather scholarly publications. Those describing the supposed linkage which I have found on the arXiv.org preprint server are here, here, here, here, here, here, here, and here. Reading these papers, the first thing that's clear is that we're talking a very small effect, whose cause can only be inferred by analyses of its periodicity and suppositions of possible causes. One must exclude environmental effects which might affect the detector in order to argue that it is the intensity of the source which is being modulated. For example, any radiation detector illuminated by a local source will also detect cosmic rays, whose intensity is known to change due to atmospheric and heliospheric effects. One must demonstrate that the supposed effect is greater than can be explained by changes in the cosmic ray flux in order to argue a direct solar influence on radioactive decay rates is present.

The way to bet (and if there were a contract on this trading on Intrade, I'd put a thousand depreciating Yankee Dollars on it) is that all of this will go away in a few months. When you get to a certain age, you've seen it all before: a very weak effect, measurable only by deviations from an expected random result, quantified only by long-term experiments with difficult to exclude environmental effects. If you frequent this chronicle and site, you'll know that I'm anything but a “professional sceptic” who dismisses wild-ass ideas indiscriminately. But several decades of chasing unicorns teaches the teachable fellow with the unicorn net that most reported sightings come to nothing and that other pursuits may be a better way to expend one's limited world line.

Fine—so I doubt there's anything there. But what if there is?  Well, it doesn't matter. There is a well-known and perfectly quantified effect which does affect the decay rate of radioactive isotopes, and that's their half life. The Cæsium-137 source illuminating the HotBits detector will lose half its intensity in 30.17 years, and hence the duration of the second of the two pairs of decays used to generate each random bit is slightly more likely to be longer than the first, which would bias the results. Now, in fact, this bias is much smaller than the variation of clock rate in the time standards used to measure the intervals between decay events, and therefore is negligible. But just in case, HotBits reverses the sense of the comparison on each pair of events to push any bias up to higher order effects.

What matters is that even if the effect suggested in this paper turns out to be real (which I hope it to be, because we could use some new science about now, although as I noted above, that isn't how I'd bet), it is a much smaller modulation of radioactive decay rate than the well-understood process of half-life, and will be equally well compensated for by the alternation of inter-event interval tests in HotBits.

The bottom line is that any generator of purported random numbers should be evaluated only on its results, not by inferences from its methodology. I have published statistical tests of HotBits results, which you are welcome to verify with your own validation suites, either with the data used in these tests, or with new data downloaded from the generator for your exclusive use. I'd say, don't worry about solar effects—worry about much greater risks. How do you know that every random data request you make from HotBits hasn't been recorded at Fourmilab and provided to The Authorities along with the IP address which requested it? You don't. How do you know, even if Fourmilab is entirely on your side, that there isn't a back door somewhere in the SSL encryption you use to receive HotBits which allows The Powers That Be to decrypt your message and extract the key you've generated as the root key for your public key infrastructure? Again, you don't.

Getting paranoid now? Good—it'll keep you looking in all directions, especially behind you, because if you're a productive citizen, they are, really, out to get you. Getting back to specifics, what can you do in the interest of privacy? Easy—generate about 100 times the HotBits you need for your root key, then choose the bytes you use by some process The Man cannot surveil: throwing a dart at a printout, rolling dice to choose pages, lines, and items from a computer-generated list of random numbers, etc.

We're creative. That's how we'll beat The Man!

That's how we always have.

That's how we always will.

Posted at August 25, 2010 21:42