1: \begin{abstract}

2: Semidefinite  Programs (SDPs)  are  an  important  class  of  optimization  problems with a variety of applications. For large-scale instances of these problems, memory becomes a key computational bottleneck. In this paper, we look at a way to represent an $\epsilon$-approximate solution of SDP with linear objective function and equality constraints in low dimensional space. This is done by generating random vectors with covariance that is positive semidefinite and an $\epsilon$-approximate solution to the SDP. Furthermore, we use the modified Frank-Wolfe algorithm, a first-order algorithm, to generate the \textit{sampled representation} of an approximate solution. This algorithm requires $\mathcal{O}(n^2/\epsilon^2)$ iterations for $\epsilon$-approximate convergence and linear additional memory at every step of the algorithm. With the advantage of low memory use throughout the algorithm, large-scale problems can be effectively solved on a small machine. One application of the framework is to use the algorithm to solve \textsc{MaxCut}. We provide a linear memory implementation of Goemans-Williamson rounding method so that the generated feasible solution has provable optimality guarantees.

3: \end{abstract}

4: