Statistical Sampling Methods

Spectre supports the following methods of sampling for Monte Carlo analysis:

Random
The Random sampling method takes the Brute Force approach of sequentially calling a random number generator without considering the samples generated previously. There is no sample selection or rejection. Therefore, it allows all samples to have an equal chance of getting selected from the population.
The Random algorithm has a convergence accuracy of 1/sqrt(N).
Latin Hypercube
The Latin Hypercube method is a quasi-random sampling algorithm with sample selection and rejection. The sample space is evenly divided into probable subspaces. All sample points are then chosen simultaneously making sure that the total ensemble of sample points is a Latin Hypercube sample and that each subspace is sampled with the same density. This method requires fewer samples to get accurate results. This method is recommended when you know how many simulation points you need to run.
The Latin Hypercube algorithm has a convergence accuracy of 1/pow(N,2/3).
Low-Discrepancy Sequence
The Low-Discrepancy Sequence (LDS) method uses a deterministic sequence to get a uniform coverage of the sampling space, which makes it better than the Random sampling method. In addition, LDS uses autostop features to generate samples, which is not supported by Latin Hypercube method. The convergence speed for LDS is faster than the Random sampling method and is comparable to the Latin Hypercube method. Therefore, overall, LDS method is better than both, Random and Latin Hypercube.