1: \begin{abstract}

2:   We study the problem of decomposing a polynomial $p$ into a sum of $r$ squares

3:   by minimizing a quadratically penalized objective

4:   $f_p(\vecp{u}) = \norm{\sum_{i=1}^r u_i^2 - p}^2$. This objective is nonconvex

5:   and is equivalent to the rank-$r$ Burer--Monteiro factorization of a

6:   semidefinite program (SDP) encoding the sum of squares decomposition. We show

7:   that for all univariate polynomials $p$, if $r \ge 2$ then $f_p(\vecp{u})$ has

8:   no spurious second-order critical points, showing that all local optima are

9:   also global optima. This is in contrast to previous work showing that for

10:   general SDPs, in addition to genericity conditions, $r$ has to be roughly the

11:   square root of the number of constraints (the degree of $p$) for there to be

12:   no spurious second-order critical points. Our proof uses tools from

13:   computational algebraic geometry and can be interpreted as constructing a

14:   certificate using the first- and second-order necessary conditions. We also

15:   show that by choosing a norm based on sampling equally-spaced points on the

16:   circle, the gradient $\nabla f_p$ can be computed in nearly linear time using

17:   fast Fourier transforms. Experimentally we demonstrate that this method has

18:   very fast convergence using first-order optimization algorithms such as

19:   \mbox{L-BFGS}, with near-linear scaling to million-degree polynomials.

20: \end{abstract}

21: