A numerical scheme for stochastic differential equations with convex constr
aints is considered. The solutions to the SDEs are constrained to the domai
n of convex lower semicontinuous function through a multivalued monotone dr
ift component and a variational inequality. The projection scheme is a time
discrete version of the constrained SDE. In the particular case when the c
onstraining function is an indicator of a closed convex domain, the SDE is
reflected. Previous convergence results for the projection scheme applied t
o reflected SDEs are recovered. (C) 2000 Elsevier Science B.V. All rights r
eserved. MSC: 60H10; 60H20; 60H99; 60F25.