It is a well-known Fact that for any continuous scalar-valued function
phi on the unit circle there is a unique best approximation in the Ha
rdy class H infinity of bounded analytic functions. However, in the ma
trix-valued case a best approximation by bounded analytic functions is
almost never unique. To make it unique, one has to impose additional
assumptions. In 1986 N. J. Young introduced the so-called superoptimal
solution of the Nehari problem. The word superoptimal means that we a
re seeking F is an element of H infinity (matrix-valued) to minimize n
ot only sup{parallel to Phi(zeta)-F(zeta)parallel to:\zeta\=1} =sup(s(
0)(Phi(zeta)-F(zeta)):\zeta\=1) but also the suprema of further singul
ar values s(j)(Phi(zeta)-F(zeta)), j greater than or equal to 1. It wa
s proved by V.V. Peller and N.J. Young that for matrix-valued function
Phi in H infinity + C such superoptimal solution is unique. In this p
aper an alternative geometric approach to the problem is presented. It
allows us to obtain another proof of the uniqueness and some new resu
lts: uniqueness of the superoptimal approximation by meromorphic funct
ions (a generalization of the Adamyan-Arov-Krein result on approximati
on of scalar-valued functions), inequalities between s-numbers of a Ha
nkel operator, and superoptimal singular values. Our approach works as
well for operator-valued functions as for matrix-valued ones. (C) 199
5 Academic Press, Inc.