1: \begin{abstract}

2: \noindent The problem of computing a representation for a real

3: polynomial as a sum of minimum number of squares of polynomials  can

4: be casted as finding  a symmetric positive semidefinite real  matrix

5: (Gram matrix) of minimum rank subject to linear equality

6: constraints.

7:  In this paper, we

8: propose  algorithms for solving the minimum-rank Gram matrix

9: completion problem, and show the convergence of these algorithms.

10: Our methods are based on the modified fixed point continuation (FPC)

11: method.  We  also use the  Barzilai-Borwein (BB) technique  and a

12: specific linear combination of two previous iterates to accelerate

13: the convergence of modified FPC algorithms. We demonstrate the

14: effectiveness of our algorithms for computing approximate and exact

15: rational sum of squares (SOS) decompositions of polynomials with

16: rational coefficients.

17: \end{abstract}