Heavy doping has been known to reduce the energy gap of semiconductors, affecting the characteristics of many devices. In the case of Si, theoretical work has so far focussed on calculating the reduced gaps. Despite the extensive calculations, however, no meaningful comparison of theory with experiment has been achieved because there exist major discrepancies in the values of the gap extracted from different experiments. In this paper, we point out that these discrepancies arose because the data were fitted by assuming simple models for different properties and report calculations of the photoluminescence spectra of n-type Si for various impurity concentrations. We are thus able, for the first time, to compare theory directly with experimental data. We find that intervalley scattering, previously considered negligible, is essential in order to reproduce the major experimental features and trends.