Almost three quarters of a century ago, Charles Frank proposed that the deep supercooling observed in metallic liquids is due to icosahedral short-range order (ISRO), which is incompatible with the long-range order of crystal phases. Some evidence in support of this hypothesis has been published previously. However, those studies were based on a small population of maximum supercooling measurements before the onset of crystallization. Here, the results of a systematic statistical study of several hundred maximum supercooling measurements on Ti39.5Zr39.5Ni21 and Zr80Pt20 liquids are presented. Previous X-Ray and neutron scattering studies have shown that the structures of these liquid alloys contain significant amounts of ISRO. The results presented here show a small work of critical cluster formation (W* = 31 - 40 kBT) from the analysis of the supercooling data for the Ti39.5Zr39.5Ni21 liquid, which crystallizes to a metastable icosahedral quasicrystal. A much larger value (W* = 60 - 99 kBT) was obtained for the Zr80Pt20 liquid, which does not crystallize to an icosahedral quasicrystal. Taken together, these results significantly strengthen the validity of Frank's hypothesis.
