Melting and subsolidus experiments were carried out on Fe–Si–S alloys (2.2–2.7 wt% Si + 2.0–2.1 wt% S) up to 146 GPa in a laser-heated diamond-anvil cell (DAC). The melting and subsolidus phase relations were examined on the basis of in-situ synchrotron X-ray diffraction measurements and ex-situ textural and chemical characterizations of recovered samples. The subsolidus phase assemblage changed from Fe-rich hexagonal closed-packed (hcp) phase + Fe3S into a single phase of hcp Fe–Si–S alloy above 80 GPa at ~2500 K. The melting curve was obtained on the basis of the appearance of diffuse X-ray scattering and/or melting texture found in the cross section of a recovered sample. Microprobe analyses of quenched molten samples showed that liquid Fe–Si–S coexisted with Fe-alloy solid being depleted in sulfur but enriched in silicon compared to the liquid. This indicates that the liquid evolves toward a Si-poor and S-rich composition upon crystallization. Our data further suggest that the ternary eutectic liquid composition is Si-deficient and close to the tie line between the eutectic points in the Fe–Si and Fe–S binary systems at each pressure. The composition of Fe–Si–S liquid that accounts for the outer core density is outside the liquidus field of solid Fe at the inner core boundary (ICB) pressure. Accordingly, the solid alloy crystallizing from such outer core liquid must be more enriched in silicon/sulfur than the coexisting liquid and thus cannot form the denser inner core required from seismic observations. Furthermore, liquid Fe–Si–C nor Fe–Si–O does not crystallize a denser solid at the ICB. These reinforce the conclusion that silicon is not an important light element in the core.