A quantum view on convex optimization

In this dissertation we consider quantum algorithms for convex optimization. We start by considering a black-box setting of convex optimization. In this setting we show that quantum computers require exponentially fewer queries to a membership oracle for a convex set in order to implement a separation oracle for that set. We do so by proving that Jordan's quantum gradient algorithm can also be applied to find sub-gradients of convex Lipschitz functions, even though these functions might not even be differentiable. As a corollary we get a quadraticly faster algorithm for convex optimization using membership queries.
As a second set of results we give sub-linear time quantum algorithms for semidefinite optimization by speeding up the iterations of the Arora-Kale algorithm. For the problem of finding approximate Nash equilibria for zero-sum games we then give specific algorithms that improve the error-dependence and only depend on the sparsity of the game, not it's size. These last results yield improved algorithms for linear programming as a corollary.
We also show several lower bounds in these settings, matching the upper bounds in most or all parameters.