# A Better Random Number Generator?

In operations research, we often use “random” numbers.  Whether we are doing a simulation of a manufacturing facility, generating future economic scenarios in financial optimization, or creating “random” instances to test our algorithms, we use up lots of random numbers.  Terry Pratchett might wonder whether we are using up all the randomness in our need for these values!

For the most part, our “random” numbers are not that at all:  they are numbers generated by a particular formula, the results of which pass certain tests for randomness.    But any random number generator has flaws and patterns.    A “weak” random number generator (think of the “middle square approach” where a number is squared and the middle part is used as the random number, and seed for the next squaring) will quickly show its flaws, while a strong one will hide its patterns longer.

But what if you want true randomness?  Or, to avoid philsophical and physical discussions, as random as rolling a die?  The Dice-O-Matic can come to your rescue!  Able to generate more than a million die rolls per day, the Dice-O-Matic is a great addition to the offices of anyone doing work in simulation.  With a footprint of a mere 18″ by 18″ (albeit 7 feet tall), this certainly looks to be about as random as one could hope.  And with the periodic melting and squashing of dice (due to the ire of gamers annoyed at an outcome), any biases will quickly be eliminated.

I think we should all include this on our next NSF grant proposals.

## 2 thoughts on “A Better Random Number Generator?”

1. This will be great for my Monopoly simulation, but I may have trouble adapting it for my Monte Carlo simulations. My risk analysis might take too many months to execute and I won’t be able to dodge the credit crunch.

In all seriousness though… I tripped over this: http://www.fourmilab.ch/hotbits/ during my undergrad. Well and true random numbers from nuclear decay as nature intended it.

Interestingly, in a wyrd, Pratchett-like way, something is actually being used up in the production of these numbers…