In this talk I will describe efforts at NREL to use simulation optimization in renewable energy research, with particular emphasis on the search for better materials for organic and inorganic photovoltaic cells. Depending on how we formulate the problems, the search spaces and solvers come in a wide variety of "shapes and sizes": continuous, discrete, mixed, large, small, hierarchical, linear, nonlinear, analytic, black box, cheap, expensive, etc. A major theme is the idiosyncratic nature of the search spaces, which arises primarily because the natural representations include domain specific structure. In fact, the more canonical the representation, the less effective the algorithm, because often constraints can be built into the representation that can dramatically decrease the size of the search space. Unfortunately, this sometimes complicates the search (e.g., off-the-shelf methods are hard to apply). Another important characteristic is that due to the discrete nature of molecular space, there is often no natural distance metric to make use of. An additional challenge is that the simulations are often expensive, not particularly robust, and not well validated. These points will be illustrated via discussion of successes and failures in several domains, especially organic and inorganic semiconductors, but also including techno-economic analysis of wind energy, multi-scale battery simulation, building energy optimization, systems biology parameter estimation, and renewable energy deployment and grid modeling.