On the power of tests of multidimensional discrete uniformity used for statistical analysis of random number generators

Abstract

In this paper, we obtained the asymptotic power values for the statistical tests of multidimensional discrete uniformity under conditions of contiguous convergence of alternatives. Two versions of the test are considered, namely, with overlapping blocks (included in the NIST SP 800-22 test suit) and with non-overlapping blocks. The null hypothesis H0 is related to the so-called pure randomness of the observed sequence, i. e. independence and the same uniform distribution of its elements. An alternative H1 is assumed to be a Markov chain of some arbitrary fixed finite order.

Keywords

• power of a test
• serial test
• test of multidimensional discrete uniformity
• contiguous alternatives
• non-central chi-squared distribution
• random number generator
• markov chain