4 ms·

I remember, at the turn of the millennium, when lava lamps were actually used to produce entropy. An SGI project called lavarand had a battery of the things bub

by bostik 12y ago

I remember, at the turn of the millennium, when lava lamps were actually used to produce entropy. An SGI project called lavarand had a battery of the things bubbling up and down, used low-end digital cameras to capture images and, after some postprocessing, hashed the results to get what was effectively a fairly good PRNG. [0]http://en.wikipedia.org/wiki/Lavarand http://en.wikipedia.org/wiki/Lavarand

baddox 12y ago
Would that be considered a PRNG? I always thought that PRNG implied a completely deterministic sequence based on a relatively simple seed.
- tonysuper 12y ago
  Technically, a lava lamp is deterministic. Still, I'd say that it's "close enough" to be considered a real RNG. There's almost no way to see a pattern in it, and I'm pretty sure that it would be completely impractical to simulate the contents inside, even if you knew all the initial conditions.
  - baddox 12y ago
    Yeah, I meant "deterministic" as in "generated using standard computations starting from a simple seed." Whether the lava lamp itself is deterministic, or the Universe moreover, is another issue.
  - PhantomGremlin 12y ago
    > Technically, a lava lamp is deterministic. Technically, it's only "deterministic" if you can fully establish the initial state. That's not possible, if for no other reason than the Uncertainty Principle [1] prevents you from measuring the initial conditions carefully enough. Plus you're not really dealing with a closed system, so basically the entire Universe can influence your system going forward. Chaos theory [2] says that you're fighting a losing game if you think it's deterministic. [1] http://en.wikipedia.org/wiki/Uncertainty_principle http://en.wikipedia.org/wiki/Uncertainty_principle [2] http://en.wikipedia.org/wiki/Chaos_theory http://en.wikipedia.org/wiki/Chaos_theory
    - sopooneo 12y ago
      Even if you could determine the initial state, wouldn't quantum effects, aggregating over time, make it non-deterministic?
      - PhantomGremlin 12y ago
        That's a really good question. I'd love to hear the answer from a physicist in the know. Assume you could overcome the uncertainty principle or observer effect and you could fully determine the initial conditions of the entire system. You knew the position and momentum of every atom. You knew the spin of every electron. Etc. Etc. Etc. Doesn't the inherent nature of quantum mechanics say that it's still impossible to predict the state of that system at some future time? Interactions only occur probabilistically, and there is no way to predict them a priori.
        rprospero 12y ago
        Physicist here. In terms of predicting the "future state", it depends on what you want to call the "state". If I know the complete wave function of the system, then knowing the wave function at a future point in time is trivial. Just apply the time evolution operator. However, knowing the wave function at a given point in time doesn't tell me the position or momentum - it just tells me the probability with which I'll measure a given position or momentum. So knowing the "state" still means that my measurements will have random components. On a slightly different note, when you talk about knowing the exact position and momentum of every particle, you're not talking about overcoming a physical limitation, but a mathematical one. To put it differently, if I know that the momentum is exactly zero, I do know that the position is. The problem is that the position is NaN. If the position isn't NaN, then I know longer know the momentum isn't precisely defined.
        claudius 12y ago
        > If I know the complete wave function of the system, then knowing the wave function at a future point in time is trivial. Just apply the time evolution operator. But wouldn't the regular measurement cause a collapse onto a randomly-chosen eigenstate of the measured operator? That is, if we have a PRNG based on a regularly measuring the lava lamp, then to predict the state after N steps, we not only have the issue of the randomly chosen N-th measurement but also have to take into account the random results of the N-1 previous measurements, which can potentially evolve into entirely new directions. Overall, it's difficult to place a lava lamp over human time scales, as both are far from the usual quantum/classical limits: We know that even in thousands of years’ time, Earth will still revolve mostly deterministically (in the classical sense) around the sun. Similarly, electrons will hardly ever behave deterministically. Lava lamps and a couple of years are oddly in between.
michaelbuckbee 12y ago
Related: video of a real life dice rolling machine : http://m.youtube.com/watch?v=7n8LNxGbZbs http://m.youtube.com/watch?v=7n8LNxGbZbs IIRC it was built because the founders just could not get people to believe that the dice rolls the generated in software were sufficiently random.
ghshephard 12y ago
Didn't it turn out to be the case that SGI discovered that they got the same amount of randomness from the CCD from just taking a picture with the lens cap on?
- bostik 12y ago
  That sounds plausible... do you happen to have any references at hand to confirm it?
- wyager 12y ago
  > they got the same amount of randomness It's hard to quantify randomness. It's possible that whatever test(s) they were using reported similar results, even though the data from a lens-capped camera might be somewhat predictable if you know the physical properties of the device. I know individual models of camera frequently have very similar thermal and electrical noise profiles, so that wouldn't be too surprising to me.

by bostik 12y ago

baddox 12y ago

Would that be considered a PRNG? I always thought that PRNG implied a completely deterministic sequence based on a relatively simple seed.

tonysuper 12y ago
Technically, a lava lamp is deterministic. Still, I'd say that it's "close enough" to be considered a real RNG. There's almost no way to see a pattern in it, and I'm pretty sure that it would be completely impractical to simulate the contents inside, even if you knew all the initial conditions.
- baddox 12y ago
  Yeah, I meant "deterministic" as in "generated using standard computations starting from a simple seed." Whether the lava lamp itself is deterministic, or the Universe moreover, is another issue.
- PhantomGremlin 12y ago
  > Technically, a lava lamp is deterministic. Technically, it's only "deterministic" if you can fully establish the initial state. That's not possible, if for no other reason than the Uncertainty Principle [1] prevents you from measuring the initial conditions carefully enough. Plus you're not really dealing with a closed system, so basically the entire Universe can influence your system going forward. Chaos theory [2] says that you're fighting a losing game if you think it's deterministic. [1] http://en.wikipedia.org/wiki/Uncertainty_principle http://en.wikipedia.org/wiki/Uncertainty_principle [2] http://en.wikipedia.org/wiki/Chaos_theory http://en.wikipedia.org/wiki/Chaos_theory
  - sopooneo 12y ago
    Even if you could determine the initial state, wouldn't quantum effects, aggregating over time, make it non-deterministic?
    - PhantomGremlin 12y ago
      That's a really good question. I'd love to hear the answer from a physicist in the know. Assume you could overcome the uncertainty principle or observer effect and you could fully determine the initial conditions of the entire system. You knew the position and momentum of every atom. You knew the spin of every electron. Etc. Etc. Etc. Doesn't the inherent nature of quantum mechanics say that it's still impossible to predict the state of that system at some future time? Interactions only occur probabilistically, and there is no way to predict them a priori.
      - rprospero 12y ago
        Physicist here. In terms of predicting the "future state", it depends on what you want to call the "state". If I know the complete wave function of the system, then knowing the wave function at a future point in time is trivial. Just apply the time evolution operator. However, knowing the wave function at a given point in time doesn't tell me the position or momentum - it just tells me the probability with which I'll measure a given position or momentum. So knowing the "state" still means that my measurements will have random components. On a slightly different note, when you talk about knowing the exact position and momentum of every particle, you're not talking about overcoming a physical limitation, but a mathematical one. To put it differently, if I know that the momentum is exactly zero, I do know that the position is. The problem is that the position is NaN. If the position isn't NaN, then I know longer know the momentum isn't precisely defined.
        claudius 12y ago
        > If I know the complete wave function of the system, then knowing the wave function at a future point in time is trivial. Just apply the time evolution operator. But wouldn't the regular measurement cause a collapse onto a randomly-chosen eigenstate of the measured operator? That is, if we have a PRNG based on a regularly measuring the lava lamp, then to predict the state after N steps, we not only have the issue of the randomly chosen N-th measurement but also have to take into account the random results of the N-1 previous measurements, which can potentially evolve into entirely new directions. Overall, it's difficult to place a lava lamp over human time scales, as both are far from the usual quantum/classical limits: We know that even in thousands of years’ time, Earth will still revolve mostly deterministically (in the classical sense) around the sun. Similarly, electrons will hardly ever behave deterministically. Lava lamps and a couple of years are oddly in between.

michaelbuckbee 12y ago

Related: video of a real life dice rolling machine : http://m.youtube.com/watch?v=7n8LNxGbZbs http://m.youtube.com/watch?v=7n8LNxGbZbs IIRC it was built because the founders just could not get people to believe that the dice rolls the generated in software were sufficiently random.

ghshephard 12y ago

Didn't it turn out to be the case that SGI discovered that they got the same amount of randomness from the CCD from just taking a picture with the lens cap on?

bostik 12y ago
That sounds plausible... do you happen to have any references at hand to confirm it?
wyager 12y ago
> they got the same amount of randomness It's hard to quantify randomness. It's possible that whatever test(s) they were using reported similar results, even though the data from a lens-capped camera might be somewhat predictable if you know the physical properties of the device. I know individual models of camera frequently have very similar thermal and electrical noise profiles, so that wouldn't be too surprising to me.