A course in computational algebraic number theory
A course in computational algebraic number theory
When are quasi-Monte Carlo algorithms efficient for high dimensional integrals?
Journal of Complexity
Concrete Math
Constructing Randomly Shifted Lattice Rules in Weighted Sobolev Spaces
SIAM Journal on Numerical Analysis
Component-by-component construction of good lattice rules
Mathematics of Computation
On the convergence rate of the component-by-component construction of good lattice rules
Journal of Complexity
Hi-index | 0.00 |
The component-by-component construction algorithm constructs the generating vector for a rank- 1 lattice one component at a time by minimizing the worst-case error in each step. This algorithm can be formulated elegantly as a repeated matrix-vector product, where the matrix-vector product expresses the calculation of the worst-case error in that step. As was shown in an earlier paper, this matrix-vector product can be done in time O(n log (n)) and with memory O(n) when the number of points n is prime. Here we extend this result to general n to obtain a total construction cost of O(sn log (n)) and memory of O(n) for a rank-1 lattice in s dimensions with n points. We thus obtain the same big-Oh result as for n prime.As was the case for n prime, the main calculation cost is significantly reduced by using fast Fourier transforms in the matrix-vector calculation. The number of fast Fourier transforms is dependent on the number of divisors of n and the number of prime factors of n. It is believed that the intrinsic structure present in rank-1 lattices and exploited by this fast construction method will deliver new insights in the applicability of these lattices.