deletions | additions       diff --git a/Quasirandom_sampling.md b/Quasirandom_sampling.md  index 3cabd31..5a56049 100644  --- a/Quasirandom_sampling.md  +++ b/Quasirandom_sampling.md 
...
What makes quasi-random sequences superior to pseudo-random numbers for sampling is that one can construct quasi-random sequences which are biased toward "spreading out," i.e., covering more of the domain.  For example, the Halton sequence is a quasi-random sequence of numbers generated over a fixed interval by using a prime number as a base \cite{Halton_1964}.  For \(N \lt 5\), The \(N\)-dimensional Halton sequence is low-discrepancy, meaning it is very similar to the uniform distribution in terms of domain coverage over \((0,1)^N\).  For large larger  \(N\), a scheme for permuting the coefficients should be used to prevent linear correlations between dimensions \cite{hess2003performance}. \cite{hess2003performance, Chi2005}.  Note that the Halton sequence can also be calculated over arbitrary sub-regions of \((0,1)^N\), making it suitable for refinement schemes, though pycalphad does not implement refinement with this approach.  The Halton sequence alone does not satisfy our purposes for sampling composition