B. Moskowitz et Re. Caflisch, SMOOTHNESS AND DIMENSION REDUCTION IN QUASI-MONTE CARLO METHODS, Mathematical and computer modelling, 23(8-9), 1996, pp. 37-54
Monte Carlo integration using quasirandom sequences has theoretical er
ror bounds of size O(N-1 log(d) N) in dimension d, as opposed to the e
rror of size O(N--1/2) for random or pseudorandom sequences. In practi
ce, however, this improved performance fur quasirandom sequences is of
ten not observed. The degradation of performance is due to discontinui
ty or lack of smoothness in the integrand and to large dimension of th
e domain of integration, both of which often occur in Monte Carlo meth
ods. In this paper, modified Monte Carlo methods are developed, using
smoothing and dimension reduction, so that the convergence rate of nea
rly O(N-1) is regained. The standard rejection method, as used in impo
rtance sampling, involves discontinuities, corresponding to the decisi
on to accept or reject. A smoothed rejection method, as well as a meth
od of weighted uniform sampling, is formulated below and found to have
error size of almost O(N-1) in quasi-Monte Carlo. Quasi-Monte Carlo e
valuation of Feynman-Kac path integrals involves high dimension, one d
imension for each discrete time interval. Through an alternative discr
etization, the effective dimension of the integration domain is drasti
cally reduced, so that the error size close to O(N-1) is again regaine
d.