Talk:Random number generation

Randomness is an observed entity

The article starts out with: "A random number generator (RNG) is a computational or physical device designed to generate a sequence of numbers or symbols that lack any pattern, i.e. appear random." But it should be the other way around: If an Observer find that a sequence lack any pattern, it appears random to him. Different observers may rate the same sequence differently. The randomness is not in the sequence.

Bo Domstedt http://www.trng98.se

Random number generation input through analogue computer filtering as an attractor which shapes the range of the output*

write about it please

True vs. pseudo-random numbers

The wording in this section (Random number generation#True vs. pseudo-random numbers) is misleading. It mixes together the theoretical underpinnings: true random numbers / pseudo random numbers and practical implementations. In the theoretical sense, the true- and pseudo-random numbers are two different beasts with their own benefits and drawbacks:

1. True random numbers are secure in cryptographic sense against future disclosures (cf. Forward secrecy)
2. The very predictability of the pseudorandom numbers is desirable for most applications (except cryptography and gaming machines, AFAIK).

In practical designs the true random generator actually includes a pseudorandom one. See, for example, NIST SP 800-90A, so there is no "vs.". Also, in the physical world, any entropy source is very fragile (for the simple reason that most of its non-catastrophic failures are extremely hard - or even impossible - to detect). Dimawik (talk) 22:04, 2 March 2024 (UTC)

Defects of simple algorithms

@Aezarebski: Thank you for adding a reference to Nishimura's work. However, I do not quite understand how this work is relevant to this section. [1] contains some language on p. 5, but I have hard time interpreting it to support the sentence "are unsuitable where high-quality randomness is required, such as ... statistics" (ellipsis is mine). Can you help me? Dimawik (talk) 06:13, 25 June 2024 (UTC)

Thanks for following up on this. If you use a poor quality PRNG, some statistical methods will produce poor results (e.g. over/underestimating variance). Many algorithms are not very sensitive to this, so often it doesn't matter. However, the cost of using a good one is not very high, so the rule of thumb is to use a decent one to be safe. I haven't been able to find a good reference spelling this out though. (Maybe this? http://www0.cs.ucl.ac.uk/staff/d.jones/GoodPracticeRNG.pdf)

I think the cleanest solution would be to remove the "statistics" part and just keep the "cryptography" part, for which a good quality PRNG is clearly important. Aezarebski (talk) 10:56, 25 June 2024 (UTC)