Given Hermitian n-by-n matrices A and B, whose eigenvalues (and
multiplicities) are known, what can the eigenvalues of A+B be? In 1962,
A. Horn conjectured an answer, in terms of eigenvalue inequalities known
as Horn inequalities. His conjecture was proved less than a decade ago
due to work of Klyachko, Knutson and Tao. We consider the analogous
question in finite von Neumann algebras, and prove that Horn
inequalities (appropriately recast) hold in all finite von Neumann
algebras. The classical method of proving a Horn inequality involves
proving the existence of a projection satisfying certain properties with
respect to three arbitrary flags. In order to do so, and building on
ideas of Knutson, Tao and Woodward, we found methods of constructing
these projections in a II_1-factor. These work also in the finite
dimensional situation and give a new and constructive proof. This is
joint work with H. Bercovici, B. Collins, W.S. Li and D. Timotin.
In work with B. Collins, Connes's embedding problem is shown to be
equivalent to a version of the above question about the distribution of
sums of Hermitian elements in finite von Neumann algebras, but with
matrix coefficients.
|