1: \begin{definition}[$Low(\mathcal{I})$]

2:   For any instance $\mathcal{I}$ of AIALO (or more generally, for any linear program), we define $Low(\mathcal{I}) \in \R$ to be the optimal solution to the following convex program.

3:   		\begin{align}

4:            \min_\tau \quad & \sum_{i=1}^n \tau_i \label{low}\\

5:           s.t.\quad & \sum_{i=1}^n \frac{(s_i^{(k)} -x_i^*)^2}{\tau_i} \le \left( \V{c}^T (\V{x}^* - \V{s}^{(k)}) \right)^2, \forall k \notag \\

6:            &\tau_i\ge 0, \forall i \notag

7:           \end{align}

8:   Here $\V{x}^*$ is the optimal solution to the LP in $\mathcal{I}$ and $\V{s}^{(1)},\dots, \V{s}^{(k)}$ are the extreme points of the feasible region $\{\V{x}:\V{A}\V{x}\le \V{b}, \V{x}\ge \V{0}\}$.  % \footnote{It is actually not necessary to consider all the corner points. The theorem still holds if we define $\V{x}^{(1)},\dots, \V{x}^{(k)}$ to be $\V{x}^*$'s neighboring corner points on feasible region. We include all the corner points only for simplicity in the proof of Theorem~\ref{unknownc}}

9: \end{definition}

10: