Hidden amorphous phase and reentrant supercooled liquid in Pd-Ni-P metallic glasses

An anomaly in differential scanning calorimetry has been reported in a number of metallic glass materials in which a broad exothermal peak was observed between the glass and crystallization temperatures. The mystery surrounding this calorimetric anomaly is epitomized by four decades long studies of Pd-Ni-P metallic glasses, arguably the best glass-forming alloys. Here we show, using a suite of in situ experimental techniques, that Pd-Ni-P alloys have a hidden amorphous phase in the supercooled liquid region. The anomalous exothermal peak is the consequence of a polyamorphous phase transition between two supercooled liquids, involving a change in the packing of atomic clusters over medium-range length scales as large as 18 Å. With further temperature increase, the alloy reenters the supercooled liquid phase, which forms the room-temperature glass phase on quenching. The outcome of this study raises a possibility to manipulate the structure and hence the stability of metallic glasses through heat treatment.


Supplementary Note 1: Slope changes of peak position Q 1 at T g and T c
It can be seen in Supplementary Figure 2 that at low temperatures, the position of Q 1 decreases with increasing temperature for both samples. The slope changes at T g , indicating a structure cross-over. In Pd 41.25 Ni 41.25 P 17.5 , however, another transition is seen at T C . Moreover, above T C (and below T x ), the slope resumes its value for T< T g . This result once again confirms the reentrant behavior for Pd 41.25 Ni 41.25 P 17.5 .

Supplementary Note 2: Sample preparation for Pd-Ni-P BMGs with different cooling rates
Pd-Ni-P bulk metallic glasses (BMGs) with different thermal history were prepared to study the effect the cooling rates. The air-cooled BMG was prepared using the fluxing technique as mentioned in the experiment section of the main text. The total time for cooling the molten liquid (~ 1473 K) to the glass transition temperature (~ 573 K) in air was recorded to be ~ 180 s, yielding an average cooling rate of ~ 5 K s -1 .
Slow-cooled samples were prepared by cooling the alloys coated with B 2 O 3 flux with a quartz tube in a furnace with a preset temperature ~ 568 K (5 K below the T g in the DSC curve using 20 K min -1 heating rate). The recorded cooling time from high temperature (~ 1473) to the preset furnace temperature (~ 568 K) is ~ 300 s, which corresponds to an average cooling rate ~ 2 K s -1 . As shown in Supplementary Figure 4, the DSC measurement for the Pd 41.25 Ni 41.25 P 17.5 metallic glass with a slower cooling rate exhibits a shallower exothermic peak at T C ~ 612 K compared to that of the air-cooled sample.

9
Glassy ribbon samples were also prepared to illustrate the influence of fast cooling rate on the anomalous exothermic peak above T g . Ingots of alloys were prepared by fluxing techniques first and then ribbons were prepared using melt-spinning. The speed of melt-spinning was ~ 30 m s -1 corresponding to a cooling rate ~ 10 6 K s -1 . As shown in Supplementary Figure 4, a DSC measurement at a 20 K min -1 heating rate confirmed that there is a similar exothermic peak at T C ~ 612 K. (not shown here). The above features for samples with different thermal histories illustrated that the structure of the metallic glasses could be controlled using heat-treatment around T C .

Data reduction procedure for S(Q)
The basic theory for obtaining S(Q) from synchrotron or neutron scattering has been given by Billinge and Egami 1 . In practice, S(Q) was reduced using software PDFgetX2 2,3 .
The incident beam flux was normalized using the ion-chamber counts.

(a) Raw data reduction and polarization correction in Fit2D
The raw data are 2-D images collected using a GE detector plate, as shown by the inset in Supplementary Figure 6. Fit2D was employed for Φ integration of the diffraction pattern images, where Φ is the azimuthal angle of the scattering vector. A polarization correction factor is also applied in this step.

(b) Intensity Correction
The background intensity mainly comes from the environmental (e.g., air) scattering, which has been subtracted at the beginning of the intensity corrections. The sample scattering intensity, I samp , of the as-cast Pd 41.25 Ni 41.25 P 17.5 bulk metallic glass can therefore be expressed as samp = ( corr + multiple + Fluor ) where, I samp -the sample intensity after background subtraction;α -the absorption factor, which depends on the sample geometry and attenuation coefficients of relevant elements; I multiplethe multiple scattering contribution; I Fluor -the isotropic fluorescence correction.
I corr is obtained after corrections outlined in Eq. (1).
Step-by-step corrections, including absorption correction, multiple scattering (two orders only) correction, and oblique incidence correction (for flat detector plate), as well as fluorescence correction, were applied in sequence and illustrated in Supplementary Figure 6.
(c) Normalization, correction for Compton scattering The measured scattering intensity is then corrected for the X-ray scattering form factor, calculated from tabulated values. By integrating over a finite and selected range of the high-q data, we can calculate the overall normalization factor N in equation (2). Compton scattering correction was applied to reduce the incoherent scattering. The tabulated data were used to describe the Compton scattering profile in PDFgetX2.
where, N -a normalization factor; I coh -the coherent, elastic scattering contribution; I incoh -the incoherent or Compton scattering contribution.
Step-by-step correction to obtain the X-ray structure factor S(Q) is shown in Supplementary Figure 6. The reduced S(Q) and G(r) are shown in Supplementary Figure 7.