Approximation of the Shannon Capacity Via Matrix Cone Programming

This paper proposes a novel formulation using the matrix cone programming to compute an upper bound of the Shannon capacity of graphs,which is theoretically superior to the Lovász number.To achieve this,a sequence of matrix cones is constructed by adding certain co-positive matrices to the positive semi-definite matrix cones during the matrix cone programming.We require the sequence of matrix cones to have the weak product property so that the improved result of the matrix cone programming remains an upper bound of the Shannon capacity.Our result shows that the existence of a sequence of suitable matrix cones with the weak product property is equivalent to the existence of a co-positive matrix with testable conditions.Finally,we give some concrete examples with special structures to verify the existence of the matrix cone sequence.