The most successful solution approaches to the traveling salesman problem are based on lower bounds obtained from linear programming relaxations. We will discuss a number of open problems concerning this approach, with emphasis on topics that could improve the practical performance of LP-based algorithms.