- Then computational complexity of whole problem is reduced by a
factor of 1000! The price, of course, is that error is increased by
variance (over
**l**) of . - This variance is added in to variance of anything else in
integrand including variance of over which is
already present.
- Thus, we can increase total number of points
**n**(where are chosen separately**n**times) by 1000 and reduce total error by if we generate**l**randomly rather than using full calculation of sum. - This strategy is often a big win!
- See following discussion for a related comment.

Geoffrey Fox, Northeast Parallel Architectures Center at Syracuse University, gcf@npac.syr.edu