Talk:Random number generator (included)

Remove draft status?

The entries so far are as I had hoped. --Paddy3118 15:03, 24 January 2010 (UTC)

A number of entries mention they invoke the rand() function from the C standard library. The wikipedia page for the linear congruential generator (whatever) mentions that the rand() in the library of many popular C implementations is of this type. There's also a link to a page that gives the spec. for the library's rand function. The spec is pretty loose. It doesn't specify the algorithm to be used by the PRNG. Only the interface, and some other properties. It does state that RAND_MAX must be 32767 or larger.

My question is - should there be a C entry that describes the situation with the rand() function of its standard library or not?

There should definitely be a C entry, because so many others defer to it, but, I thought that the C implementations were likely to vary on different platforms, (which, if true, should also be noted)? The (good), thing about C is that it is used everywhere, including in embedded systems, and many C environments only approach the spec of a standardization effort --Paddy3118 06:19, 30 January 2010 (UTC)

Main types of PRNG

This might be text sometime to transfer to the main task page.

The main types of PRNG that are in use are the Linear Congruential Generator and the Generalized Feedback Shift Register (of which the mersenne twister generator is a subclass). The last main type is where the output of one of the previous ones (typically a mersenne twister) is fed through a cryptographic hash function to maximize unpredictability of individual bits.

LCGs have the advantage of not requiring much state and being very fast to calculate, but produce random numbers with spectral problems. This makes them unsuitable for both Monte Carlo simulation and cryptography. By contrast, GFSRs (of which the Mersenne Twister is is particularly high quality version) require a lot more internal state and are considerably more expensive to compute and initialize (so much so that it is normal to use a LCG or simpler GFSR to drive the initialization); GFSRs tend to have much higher quality spectral properties than LCGs, and are suitable for use in Monte Carlo simulation. Neither LCGs nor GFSRs should be used for the most demanding applications (cryptography) without additional steps.

Probably needs more work. Feel free to massage! –Donal Fellows 16:17, 24 January 2010 (UTC)

Not much massaging done, just added some WP links. Ta! --Paddy3118 04:06, 25 January 2010 (UTC)