The present work is part of a broad effort toward predictive simulations of complex multi-physics flows at high Reynolds numbers. The main objective of the Stanford PSAAP Center (Predictive Science Academic Alliance Program) is to predict the reacting flow in the HyShot II scramjet experiment carried out in the High Enthalphy shock tunnel at Göttingen (HEG) facility of the German Aerospace Agency DLR (cf. Gardner et al. 2004; Laurence et al. 2012). Specifically, the objective is to predict the best-estimate of the flow, to estimate the uncertainties from a range of sources in this prediction and, finally, to quantify the margin to engine unstart. Scramjet combustors are designed to use the heat added from combustion to cause an increase in pressure, which, when expanded in a nozzle, causes thrust. At nominal conditions, the HyShot II is designed to have essentially attached flow throughout the combustor, with a smooth (not counting the many oblique shock waves) rise in pressure throughout. If too much fuel is injected, leading to excessive heat addition, the flow may become choked; this would initiate an unsteady " unstart " process, eventually resulting in subsonic flow throughout the combustor and complete loss of thrust. This failure mode is abrupt and catastrophic. The objective of the present work is to characterize the flow in the regime separating nominal operating conditions from catastrophic unstart. This is done using a combination of large eddy simulations (LES) and experimental data from a recent study by Laurence et al. (2012). Preliminary results of the LES study were reported in Larsson et al. (2011). The fact that LES directly resolves the largest turbulent structures implies that it, generally speaking, leads to more accurate predictions of mixing than RANS is capable of. Mixing is a key phenomenon in supersonic combustion; in many cases it is the rate-controlling process. LES can therefore be expected to yield more accurate and trustworthy predictions than RANS (cf. Fulton et al. 2012, for a direct comparison for a relevant flow). A nice summary of the current state-of-the-art and different approaches in LES of scramjet flows is provided in Fureby (2012). The complexity of the flow (turbulence, shocks, mixing, combustion) implies that a careful multi-stage validation plan is needed. Moreover, the available quantitative experimental data for the HyShot case is limited to measurements of pressure (mean and rms) and heat transfer (considered uncertain) along two lines in the combustor. Therefore, validation on the HyShot problem alone is not sufficient. The first step in the validation is to compute a supersonic flat plate boundary layer as a basic check on the LES method and, specifically, the modeling of the friction and heat transfer at the wall; this was reported in Larsson et al. (2011), where a full description of the wall-model was also provided. The next validation step is the duct flow with shock/boundary layer interactions studied experimentally by Helmer et al. (2012). This problem validates the capability of the wall-modeled LES to capture stress-induced secondary corner flows and, most importantly,