When using one of the Reynolds Stress turbulence models the CFX Solver can sometimes fail in the first few iterations. What's the best way to initialize a simulation when using these models?
The Reynolds Stress models (which include: k epsilon EARSM, BSL, BSL EARSM, BSL Reynolds Stress, SSG Reynolds Stress, LRR Reynolds Stress, QI Reynolds Stress and Omega Reynolds Stress) are regarded as the natural and most logical level of modeling within the framework of the Reynolds averaging approach at the cost of being computationally expensive. If you start directly from a Reynolds Stress model, the solver may fail at the very first iteration. The best route is to start with a SST solution, then use the BSL EARSM model (Explicit Algebraic Reynolds Stress Model, Beta feature in CFX11), then switch to the final Reynolds Stress model of choice. If you go directly from SST to a full Reynolds Stress model, the initialization of the six Reynolds Stresses cannot be done properly. This is because the SST turbulence model still assumes isotropic turbulence, and it uses one single constant value to apply to the six Reynolds stresses, these are: uu, uv, uw, vv, vw, and ww. The six Reynolds transport equations then have to rapidly adjust to the flow field and this can sometimes cause them to diverge. An Algebraic Reynolds stress model transforms the differential equations into algebraic ones and leads to direct expressions for the Reynolds stress components. Although this significant simplification usually doesn`t yield satisfactory results for the model itself, it can provide a good initial guess for the full Reynolds stress model. Some people like to start from a kepsilon solution first, but this cannot guarantee a smooth restart either, since the code only initialize the diagonal stresses with 2/3*k, and will nullify the offdiagonal stresses. A high quality mesh is also required when using a Reynolds Stress Model. 

