A note on semidefinite programming relaxations for polynomial optimization over a single sphere (Q341317)

This paper studies minimizing a polynomial function over a single sphere. The authors first compare recent two semidefinite programming (SDP) relaxations for computing the best rank-1 tensor approximation. It is shown that these two SDP relaxations are essentially equivalent. Then a specific example arising from Bose-Einstein condensates is considered, whose objective function is a summation of a quadratic and a quartic function. Since the two SDP relaxations for the best rank-1 tensor approximation usually are not suitable due to the large SDP matrix dimension, a much smaller quadratic SDP relaxation is proposed. Then both deterministic and randomized rounding procedures are developed and approximation ratios between function values at the global minimum and the solution constructed from rounding procedures are provided. Preliminary numerical examples illustrate the performance of these semidefinite relaxations.