Let f P C 1Y1 À1Y 1 2 , d 2 f adxdy^0. We characterize the unique best onesided L 1 -approximant h à to f from above (resp. h à from below) with respect to the subspace B 1Y1 which consists of all bivariate functions which are sums of univariate functions. h à resp. h à are constructed by a Hermite type interpolation on the diagonal ftY t X t P Ig resp. the anti-diagonal ftY Àt X t P Ig, where I X À1Y 1.
Introduction and main result.In various situations the characterization of best L 1 -approximants can be given in terms of interpolation with respect to canonical point sets.Interpolation with respect to such sets yields best L 1 -approximants for a large class of functions. As classical examples we mention the Markov Theorem (see p. 87]) and the best one-sided L 1 -approximation of monomials by polynomials of lower degree. These examples deal with univariate approximation by finite dimensional spaces of approximating functions.On the other hand, there are characterization results for multivariate L 1 -approximants by interpolation with respect to canonical point sets where the spaces of approximating functions are infinite-dimensional. For example, if the space of approximating functions consists of harmonic functions on the unit ball in R n , then we have a concentric sphere as a canonical point set (see [2] for details). If one considers L 1 -approximation by blending functions defined by