In this work, we studied one of the very widely used approximations in the prediction of an enzyme reaction mechanism with computational methods, that is, fixing residues outside a given radius surrounding the active site. This avoids the unfolding of truncated models during MD calculations, avoids the expansion of the active site in cluster model calculations (albeit here only specific atoms are frozen), and prevents drifting between local minima when adiabatic mapping with large QM/MM models is used. To test this, we have used the first step of the reaction catalyzed by HIV-1 protease, as the detrimental effects of this approximation are expected to be large here. We calculated the PES with shells of frozen residues of different radii. Models with free regions under a 6.00 Å radius showed signs of being overconstrained. The QM/MM energy barrier for the remaining models was only slightly sensitive to this approximation (average of 0.8 kcal·mol, maximum of 1.6 kcal·mol-1). The influence over the energy of reaction was almost negligible. This widely used approximation seems safe and robust. The resulting error is on average below 1.6 kcal·mol, which is small when compared with others deriving from, for example, the choice of the density functional or semiempirical MO/SCC-DFTB method, the basis set used, or even the lack of sampling or incomplete sampling.