We observe that polynomial evaluation and interpolation can be perform
ed fast over a multidimensional grid (lattice), and we apply this obse
rvation in order to devise a simple algorithm for multivariate polynom
ial multiplication. Surprisingly, this simple idea enables us to impro
ve the known algorithms for multivariate polynomial multiplication bas
ed on the forward and backward application of Kronecker's map; in part
icular, we decrease, by the factor log log N, the known upper bound on
the arithmetic time-complexity of this computation (over any field of
constants), provided that the degree d in each of the m variables is
fixed, m grows to the infinity, and N = (d + 1)m.