In monte carlo analysis, what is the need for the "monte carlo seeds" field.
Is there any result variation for the different number of seeds..
Setting the seed simply alters the starting point for the random number generator. It defaults to 12345 (if not filled in) which means that you will have a repeatable sequence of random numbers so if you run the same simulation twice (assuming nothing changed) you'll get the same results. If you set a different seed - the exact sequence of random numbers will be different, but will still be generated using the same distribution - and so your measurements should also have the same distribution (e.g. same mean and standard deviation), provided of course that you have enough samples.
Prior to implementing this, people often asked to be able to set the seed (it's been in spectre for as long as monte carlo has been available in spectre) - I'm never that convinced it matters that much though - the only time it might be useful is if your sample size is too small and hence your standard deviation is maybe not really "converged".
In reply to Andrew Beckett:
i need some clarification regarding "seeds".. so it will be helpful if i set some value to the seeds field and run the monte carlo analysis for small number of iterations.. for example..
i set seeds = 2 and run 200 iterations.
then i set seeds = 3 and run 100 iterations...
it means i will get same distribution and variation.. independent of iteration..
Is it correct..? and what is the maximum number of seeds field ?
on which basis this seeds are chosen?
Please reply as soon as possible..
In reply to kenambo:
Well, you won't get exactly the same distribution and variation, but you will get roughly the same shape. The fewer iterations you do, the less similar the shape will be. For example, in this picture I show the results from running a function that I have which generates random numbers with a normal (gaussian) distribution with two different seeds (this is not from monte carlo, but the principle is the same), with two different numbers of iterations. You can see that the mean and standard deviations are similar, but not identical.
Note that in the simulator, it may not be quite as lumpy as this - you typically have more random variables (this is just one random variable) - but the fewer iterations, the bigger variation you'll get. I'll post one with many more samples in a moment.
And here's a comparison between two different seeds for more iterations. Again, this is not a monte carlo simulation, but it should show you the idea that the number of iterations makes the mean and standard deviation more stable. Note that if I ran again with the same seed, I'd get the same answers. If I just ran the test again twice in a row, the two histograms (and means and standard deviations) would be slightly different.
Click on the picture to see it all in a new tab or window in your browser.
Hopefully that's a reasonable "Random Numbers 101" course?