1 Introduction

Liquid-to-liquid phase transitions (LLPT) belong to the group of polyamorphous phase transitions (PPT) and are driven by entropy, which refers to transitions between liquid phases with different locally favored structures (LFSs) and identical chemical composition [1,2,3]. LLPTs have been reported to exist in a number of liquids, such as SiO2 [4], Ce [5], Al2O3–Y2O3 [6], and triphenyl phosphite (TPP) [7]. High-pressure-induced LLPTs have been frequently reported. Katayama et al. [8] found two forms of liquids with different structures in phosphorus by in situ high-pressure X-ray diffraction (XRD). A similar finding was also observed by neutron diffraction of water [9] with the application of appropriate pressures. It was found that PPT occurred in Ce-Al metallic glasses (MGs) under high pressure, and 4f electron delocalization was proposed to explain the phenomenon [2, 10].

The structural origin of LLPT remains an ambiguous and highly controversial issue due to the interruption of crystallization in the undercooled metallic liquids upon heating/cooling [11]. LLPT was proposed to occur in some glass-forming metallic liquids above the melting point, such as La-based MGs [12]. However, the structural evolution induced by temperature-induced LLPT is more pronounced and may be overlooked in the undercooled liquids when crystallization occurs during cooling [8, 13]. An electrostatic levitation experiment was firstly adopted by Li et al. [14] to demonstrate the possible LLPT in MGs by measuring the change in specific volume during heating and cooling. Later, evidence for LLPT was also found in the glass-forming liquid of Vit1 (Zr42.6Ti12.4Cu11.2Ni10Be23.8) [15], Zr50Cu40Al10 [16], Vit106 (Zr57Nb5Al10Cu15.4Ni12.6) [13], etc.

Recently, Pham et al. [17] and Lan et al. [18] found that Pd–Ni–P glass alloys with an anomalous exothermic peak (AEP) in the supercooled liquid region underwent a hidden LLPT and subsequent reentrant behavior in the supercooled liquid region (SCL). By using in situ synchrotron XRD experiment, a noticeable change of the atomic structure of the Pd–Ni–P MG was observed in the medium-range scale at a transition temperature (TC). Moreover, by using small-angle neutron scattering (SANS), the existence of two different liquid states in the supercooled liquid region was proven. Later on, such evidence of hidden LLPT was also observed in Fe–Nb–Y–B MGs [19] and BAM 11 (Zr52.5Cu17.9Ni14.6Al10Ti5) [20]. Therefore, the controversy related to LLPT in MGs can be resolved by using state-of-the-art multiple-scale structure probing techniques on glassy alloys exhibiting an AEP. In addition to the above-mentioned alloy systems, AEP has been found in the calorimetric curves of many other MGs, such as Cu–Zr–Al–Y [21], Gd-Zr–Al–Ni [22] and Mg–Cu–Ag–Gd [23].

Mg-based MGs, as high-strength light alloys combined with excellent corrosion resistance [24,25,26,27], have been of great interest to researchers due to their potential applications in hydrogen storage [28] and biomedical [29,30,31]. Tailoring the atomic-to-nano-microstructure in Mg-based metallic glasses should be helpful in the above applications. However, the limited glass-forming ability (GFA) hindered the application of Mg-based glassy alloys [32]. Previous studies have reported that the GFA of Mg65Cu25Gd10 ternary alloys could be significantly improved [33, 34] with Ag substitution for Cu [35]. Although much effort has been made, the structural origin of the enhanced GFA by Ag addition remains unsettled. Moreover, a calorimetric anomaly could often be observed in some Mg–Cu–Ag–Gd MGs in the supercooled liquid region, e.g., Mg65Cu15Ag10Gd10 alloy, indicating the occurrence of structure evolution driven by entropy. The hidden liquid-state phase transition opens a possible window to reveal the correlation between the structure evolution and the GFA of Mg–Cu–Ag–Gd alloys.

In this paper, by combining in situ synchrotron XRD and small-angle X-ray scattering (SAXS), the multiscale structural changes were probed in Mg–Cu–Ag–Gd MGs of an AEP during heating. It has been found that LLPT occurs in the supercooled liquid region as temperature rises, accompanied by a change in packing density at TC. The synchrotron diffraction shows that the local order increases first and reenters into the initial states in the AEP temperature region upon heating. Moreover, the SAXS results indicate that the nanoscale heterogeneous structures appear first and diminish during the following heating process. A core–shell model was applied to analyze the new hidden amorphous structure after the occurrence of LLPT. There is no anomalous structure and property change for the reference alloys Mg–Cu–Gd without an AEP. It was also found that the square resistivity and mechanical properties could be tailored by applying appropriate heat-treatment in the AEP temperature region. Our results suggest that the anomalous multiscale structure evolution may be correlated with the excellent GFA for the Mg–Cu–Ag–Gd MGs of a calorimetric anomaly.

2 Experimental

In this paper, the Mg65Cu15Ag10Gd10 and Mg65Cu25Gd10 MGs ribbons were prepared using melt-spinning. The master alloys were prepared by induction melting Mg, Cu, Ag, Gd (purity > 99.9%) raw material particles under a high-purity Ti-gettered Argon (Ar) atmosphere. Subsequently, the ribbon samples were produced by the single-roller spinning technique. The thickness of the obtained ribbon is about 45 µm, and the width is 3 mm. XRD (Bruker-AXS D8 Advance) was used to check whether the as-prepared melt-spun ribbons are amorphous. The incident wavelength of the XRD is 0.154 nm, and the range of diffraction angle (2θ) is 20°–60°.

Transmission electron microscopy (TEM, FEI TECNAL G2 20) was also employed to observe the amorphous nature of the sample. A cryogenic ion-miller (Gatan 695C) was used to thin the ribbons using parameters at the beginning: 8° in angle, 8 keV in incident ion beam energy within 25 min in duration time. In order to acquire a better quality of the thin area, proper thinning parameters with a lower angle and lower energy were used: 3° in angle, 3 keV in energy, and 15 min in duration time.

The specific heat capacity experiment of the ribbons of 20 mg in mass was carried out by differential scanning calorimetry (DSC, METTLER TOLEDO) in the high-purity N2 atmosphere. The N2 flow rate was 50 ml·min−1. A sapphire reference sample was used for the specific heat capacity (Cp) measurements. The temperature range was set as 323–523 K at a heating rate of 10 K·min−1.

Time-resolved SAXS measurements were conducted at beamline 12-ID of the advanced photon source (APS), Argonne National Laboratory (ANL). The high energy X-ray beam size is 200 µm \(\times\)40 µm and wavelength is 0.0886 nm. A high-energy X-ray beam was employed in transmission geometry for data collection. The data rate is ~ 1 Hz. Small-angle scattering data were corrected for background scattering (empty sample holder), transmission and detector efficiency, and set to absolute units. The results were analyzed with the Igor Pro 6 equipped with an IRENA software package [36] to get information on their structure [37].

High-energy synchrotron X-ray diffraction measurements were performed at beamline 11-ID-C of the Advanced Photon Source (APS), Argonne National Laboratory. High energy X-rays with a beam size of 500 μm \(\times\) 500 μm and wavelength of 0.01173 nm were used for data collection. Two-dimensional (2D) diffraction patterns were obtained using a Perkin Elmer amorphous silicon detector. A Hz data rate was used and the sample to detector distance was 1800 mm. The static structure factor, S(Q), was derived from the 2D diffraction ring patterns recorded on the detector by masking bad pixels, integrating images, subtracting the appropriate background and correcting for oblique incidence, absorption, multiple scattering, and fluorescence. Compton scattering and Laue correction48 was conducted using Fit2D and PDFgetX2 [38].

For the in situ heating square resistance test, before experiments, the surface of the Mg65Cu15Ag10Gd10 melt-spun ribbons was polished and cleaned with alcohol to obtain an oxide-free sample surface. Using this procedure, detrimental effects on the resistance measurements could be avoided. A commercial four-point resistance tester with heating furnace equipment was used to measure the square resistance. Before the experiment, the oxide layer on the resistance tester probe has been carefully polished to reduce resistance measurement error. The annealing temperature range is from 303 to 503 K at a heating rate of 10 K·min−1. The resistance tester probe distance is 4 mm. Argon gas was introduced throughout the test to eliminate the error on the resistance value caused by oxidation of the sample during the heating process.

The nanoindentation tests were carried out at room temperature using an Agilent Nano Indenter G200 with a typical Berkovich diamond indenter [39]. Ribbons were made into 3 mm \(\times\) 3 mm square sheets and heat-treated at different temperatures in DSC. Before the actual experiments, calibrations were conducted on a fused silica standard. The nanoindentation tests were performed to measure the hardness and elastic modulus of Mg65Cu15Ag10Gd10 melt-spun ribbons. The load increased from 0 to 20 mN at a loading rate of 0.05 mN·s−1, and the peak load was held for 10 s and then was unloaded at a similar rate with the loading one. To ensure accuracy and eliminate scattering of the indentation data, at least 5 indents for each ribbon were performed during the experiment.

3 Results and discussion

3.1 Thermal-physical behavior and initial sample inspections

Figure 1a shows the specific heat capacity (Cp) curves of two glassy ribbons, Mg65Cu15Ag10Gd10 and Mg65Cu25Gd10, respectively. The corresponding thermophysical parameters, i.e., glass transition (Tg), crystallization temperature (Tx), have been labeled in DSC curve. The experiment determined the glass transition temperature of Tg ~ 441 K and crystallization temperature of Tx ~ 453 K for Mg65Cu15Ag10Gd10 and Tg ~ 412 K and Tx ~ 460 K For Mg65Cu25Gd10. There is an AEP in the DSC curve, which is ~ 12 K below the crystallization temperature (Tx), and the temperature is labeled as TC ~ 441 K for Mg65Cu15Ag10Gd10 MG. There is no such DSC anomaly for Mg65Cu25Gd10. The calculated GFA indicators, the reduced glass transition temperature (Trg, and Trg=Tg/T1) [40, 41] and Gamma value (γ) [42], for both MGs, were calculated and are summarized in Table 1. The upper temperature values of the melting endotherms were measured and are designated by Tl and summarized in Table 1. The values of GFA indicators for Mg65Cu15Ag10Gd10 glass are larger than those of Mg65Cu25Gd10 glass, and it is consistent with the larger critical casting thickness of Mg–Cu–Ag–Gd.

Fig. 1
figure 1

a Specific heat capacity curves Cp for Mg65Cu15Ag10Gd10 and Mg65Cu25Gd10 MGs and (inset) corresponding XRD patterns; b TEM result and SAED pattern (inset) for Mg65Cu15Ag10Gd10

Full size image
Table 1 Thermophysical parameters of Mg65Cu15Ag10Gd10 and Mg65Cu25Gd10 alloys
Full size table

The inset in Fig. 1a shows XRD patterns for both MGs. Only broad diffusely scattering peaks for both MGs can be observed, confirming the amorphous characteristic. Figure 1b is the bright-field TEM image for Mg65Cu15Ag10Gd10 ribbons, showing no nanoscale precipitate. Moreover, the inset in Fig. 1b is the corresponding selected area electron diffraction (SAED) pattern. The SAED pattern shows diffused halos, further confirming that the samples are amorphous.

3.2 In situ synchrotron X-ray diffraction and square resistance measurements results

Figure 2a, b shows the results of in situ synchrotron XRD results of two Mg-based MGs during heating. Both samples show essential amorphous characteristics, that is, diffused peak maxima, upon heating up to Tx. The inset in Fig. 2a is the 2D plot of structure factor S(Q) for Mg65Cu15Ag10Gd10. The S(Q) patterns during heating reveal the atomic-scale structural changes. One can observe that the peak position, peak width, and peak intensity for the first sharp diffraction peak between 20 and 30 nm−1, i.e., Q1, changes between Tg and Tx for the Mg65Cu15Ag10Gd10 MG, indicating the occurrence of liquid-state phase transition in the supercooled liquid region. It can be seen that the peak shifts back and forth, and the peak intensity increases first and decreases at a higher temperature around TC. In contrast, there is no noticeable structure change for the reference sample Mg65Cu25Gd10 up to heating above Tx.

Fig. 2
figure 2

Synchrotron XRD patterns during heating for a Mg65Cu15Ag10Gd10 and b Mg65Cu25Gd10; the first moment of Q1 (peak position) versus temperature curves for c Mg65Cu15Ag10Gd10 and d Mg65Cu25Gd10 MGs, where specific heat versus temperature curves were superimposed

Full size image

To further reveal the evidence of liquid-state phase transition for Mg65Cu15Ag10Gd10 MG, Fig. 2c, d shows the first moment vs. temperature curves during heating for both MGs. The first moment represents the peak position of Q1 [13, 18, 19]. The specific heat capacity (Cp) curves of two ribbons, Mg65Cu15Ag10Gd10 and Mg65Cu25Gd10 were superimposed. In Fig. 2c, one can find that the peak position firstly decreases and then increases back to the initial value around TC during the heating process, showing a good correlation with CP changing in Mg65Cu15Ag10Gd10. These anomalous structural changes may point to the LLPT, revealing the existence of hidden metastable amorphous phase in the supercooled liquid region of Mg65Cu15Ag10Gd10. The peak position is positively correlated with the packing density of amorphous alloys [43]. The peak intensity correlates positively with the local ordering for the amorphous matters [44]. Therefore, the new amorphous structure becomes more ordered after LLPT, and its packing density would be lower than that of the initial state. To distinguish it from supercooled liquid 1 (SCL1), which occurs around the supercooled liquid region, the new metastable phase is named SCL2. In Fig. 2d, as for the Mg65Cu25Gd10 MG, the first-moment curve of Q1 shows a sharply upward trend at the crystallization temperature (Tx). However, there is no anomalous structure change from Tg to Tx for Mg65Cu25Gd10 MG.

The synchrotron diffraction results suggest an LLPT in the supercooled liquid region of the Mg65Cu15Ag10Gd10 MG, and the new amorphous phase would dissolve when heated to temperatures higher than TC, thereby, reentering the supercooled liquid of the room temperature-glass phase. Reentrant has been used to describe this phenomenon in previous studies [45], which refers to a phenomenon in which matter changes back to the original phase after changing from one phase to another [17, 46, 47]. When Mg65Cu15Ag10Gd10 metallic glass is heated above the glass transition temperature (Tg), it reaches the initial liquid state SCL1. As the temperature rises, it transforms into a liquid SCL2 with a high degree of order and finally re-transforms to a supercooled liquid similar to the initial liquid SCL1. This process is similar to the reentrant phenomenon described above.

As shown in Fig. 2c, the square resistance curve of Mg65Cu15Ag10Gd10 also displays an abnormal evolution as temperature rises. It is observed that with temperature increasing from room temperature to Tg, the square resistance decreases linearly and slowly. Above the glass transition temperature, the Mg65Cu15Ag10Gd10 amorphous alloy undergoes a glass transition into a supercooled liquid state with the change in the relaxation mechanism causing an evolution of the electron transport properties. The square resistance deviates from the original linear relationship and enters another stage; a sharp increase can be observed, as shown in the blue line. As the temperature continues to rise near TC, the LLPT gradually occurs in the supercooled liquid region, and the square resistance exhibits a downward trend again. As the phase transition further evolves, the resistance rises subsequently. The square resistance curve and DSC curve show a remarkably consistent trend, implying that the in situ resistance method can detect the characteristic temperature of the structural transformation of Mg65Cu15Ag10Gd10 amorphous alloy during heating as well. It can be seen that the square resistance as a physical parameter can reflect the change in the amorphous structure from the electronic level [48,49,50,51,52,53].

3.3 Time-resolved SAXS

Figure 3a, b shows the results of time-resolved SAXS results of two Mg-based MGs during heating in order to reveal the evolution of nanoscale inhomogeneities and to provide further evidence for reentrant phenomena. The evolution of nanoscale structural heterogeneities matches well with atomic-scale structural changes, further indicating the occurrence of LLPT and the associated reentrant behavior in the supercooled liquid region for the Mg65Cu15Ag10Gd10 MG. Unlike Mg65Cu25Gd10 MG, one can clearly observe that the SAXS intensity within the range of 1.2–1.7 nm−1 shows a rise and fall around TC of Mg65Cu15Ag10Gd10 MGs. Moreover, as shown in Fig. 3c, the integration over the Q range of 1.42–1.53 nm−1 instead of the integration over the full Q range shows a rise and fall with TC temperature, suggesting that a hidden amorphous phase with size of ~ 4 nm first appears and then disappears upon heating. In contrast, Mg65Cu25Gd10 does not show abnormal SAXS intensity changes during the entire heating process, as shown in Fig. 3d.

Fig. 3
figure 3

SAXS profiles during heating for a Mg65Cu15Ag10Gd10 and b Mg65Cu25Gd10 MGs; integrated SAXS intensity versus temperature during heating for c Mg65Cu15Ag10Gd10 and d Mg65Cu25Gd10 MGs, where Cp curves have been superimposed

Full size image

Further analysis has been performed to reveal the additional information behind the SAXS profiles related to the interface, size, and morphology of the phases during the LLPT. An interference peak appears when approaching TC. There is an interference peak for the SAXS profile at TC as shown in Fig. 4a, indicating the formation of the second phase with diffused interfaces. The appearance of the interference peak usually correlates with the occurrence of core–shell nano-precipitation [54]. The core–shell structure model of small-angle scattering was given by the following equation [55]:

$$I(Q) = \frac{k}{{V_{{\text{S}}} }}[3V_{{\text{C}}} (\rho_{{\text{C}}} - \rho_{{\text{S}}} )\phi (QR_{{\text{C}}} ) + 3V_{{\text{S}}} (\rho_{{\text{S}}} - \rho_{{\text{M}}} )\phi (QR_{{\text{S}}} )]^{2}$$
(1)

where \(\phi (x) = (\sin x - x\cos x){/}x^{3}\), R is the radius, V is the volume, and ρ is the scattering length density. The subscripts C, S, and M denote the core, shell, and matrix, respectively. k is a scale factor and \(\phi (x)\) is a function that peaks at Q = 0 and falls monotonously at small Q values. The core–shell model fitting analysis is shown in Fig. 4b. The characteristic lengths, including mean radius (~ 2.6 nm), shell thickness (~ 1.0 nm), reach the maximum value at TC. The size decreases when heating above TC, indicating that core–shell structure diminishes upon continuous heating. Moreover, the slope of the SAXS profile at the high Q region could be correlated with the shape of the second phase. A characteristic feature of the small-angle scattering from fractal scatterers and many other disorder solids often obeys a “power-law” scattering in the magnitude for two types of fractal systems, including mass fractals and surface fractals. For mass fractals, I(q) ∝ q−D, where q is the scattering vector, and D is the mass fractal dimension and has a noninteger value for random or disorder objects. For rods, disks, and spheres, D is 1, 2, and 3, respectively. For surface fractal, \(I(q) \propto q^{{D_{{\text{S}}} - 6}}\), where DS is the three-dimensional surface fractal dimension which relates surface area [20, 56]. Figure 4c shows two power-law regimes [57], including a high q range (1.6–2.0 nm−1) and an intermediate q range (0.3–0.6 nm−1). At the intermediate q range, the scattering slope increases around TC during heating, representing the metastable phase with an elongated shape appear. The slope at the high q range shows a fall and rise around TC temperature, indicating that the surface is getting smooth with the formation of the SCL2 phase and then gets rough again, which tends to have a reentrant behavior. The slope vs. temperature curve shows slope change at TC, further confirming the occurrence of LLPT at TC.

Fig. 4
figure 4

a SAXS model fitting results for Mg65Cu15Ag10Gd10; b size distribution of the second phase after LLPT in supercooled liquid region of Mg65Cu15Ag10Gd10; c change of slope n values in Porod’s region as a function of temperature; d Porod’s analysis results for Mg65Cu15Ag10Gd10

Full size image

Figure 4d is the Porod’s analysis results for the SAXS profiles at different phase transition stages. According to the Porod’s law [55, 58], with the interface between the precipitates and the matrix becoming sharper, the value of Q4I(Q) [59] tends to be more constant as Q increases to a higher value. The Q4I(Q) at the higher Q4 region (> 6 nm−4) increases and tends to become constant as temperature increases, implying the increase in the electron density difference between the second phase and matrix, that is, suggesting that the diffused interfaces would become sharper during the LLPT and crystallization process.

3.4 Nanoindentation experiment results

Figure 5a shows the load–displacement curves of the Mg65Cu15Ag10Gd10 MG obtained from nanoindentation tests. It can be observed from Fig. 5a that the load–displacement curves obtained at different annealing temperatures are different. The displacement offset gradually decreases nonlinearly with annealing temperature increasing, indicating structure changes upon annealing. According to the statistical analysis of nanoindentation experiments, Fig. 5b summarizes modulus and hardness results at different temperatures for Mg65Cu15Ag10Gd10. As shown by the blue curve in Fig. 5b, the slope of the modulus varies differently at different temperatures. The modulus first increases and then decreases when heating from room temperature to Tg. However, the hardness continues to increase below Tg. It seems like the relaxation process dominates and affects the mechanical properties below Tg. Moreover, in the AEP temperature region close to TC, the modulus shows downward and upward changes. Simultaneously, there is a slope change of the hardness vs. temperature curve, indicating that the mechanical properties could also be tuned by the occurrence of LLPT and the following heat-treatment. The glass with the formation of new amorphous phases at TC would possess higher hardness and modulus than that of a relaxing glass at ~ Tg. It has been found that dissimilar nanoscaled structural heterogeneity would contribute to differing creep resistance during nanoindentation tests [60]. Therefore, structural heterogeneities resulting from LLPT may play a role in tailoring the relaxation and mechanical behaviors in Mg-based MGs.

Fig. 5
figure 5

Nanoindentation test results for Mg65Cu15Ag10Gd10 MGs: a load–displacement curves of Mg65Cu15Ag10Gd10 ribbon at different temperatures; b modulus and hardness as a function of annealing temperatures

Full size image

4 Conclusion

In conclusion, multiscale structural evidence has been revealed using in situ synchrotron diffraction and time-resolved SAXS measurements. An anomalous exothermic peak and thermophysical parameters have been determined, suggesting the occurrence of an LLPT and confirming that the GFA indicators of Mg–Cu–Ag–Gd MGs are higher than those of Mg–Cu–Gd MGs. Moreover, the peak position and integrated intensity for the first sharp diffraction peak in structure factors show abnormal changes at TC in the supercooled liquid regions of Mg–Cu–Ag–Gd MGs, indicating atomic structure evolution at medium-range order length scale when LLPT and the following reentrant supercooled liquid occurs. Square resistance measurements show consistent changes during the LLPT and the following reentrant supercooled liquid. Furthermore, the time-resolved SAXS experiment reveals a core–shell structure that emerges and diminishes when heating in the supercooled liquid region of Mg–Cu–Ag–Gd MGs, showing tunable nanoscale heterogeneity by LLPT. The morphology and interfacial information of the hidden amorphous phase resulting from LLPT have also been given. Nanoindentation experiments further confirmed that the mechanical properties could be tuned by applying appropriate heat-treatments around TC to enable the occurrence of the LLPT. Abnormal thermodynamic behavior and anomalous structure response could not be found for the reference alloy system, Mg–Cu–Gd. The above results may suggest that the AEP, LLPT, and the following reentrant behavior correlates with the good GFA of the Ag-addition Mg-based MGs. Furthermore, it would be interesting to determine the correlation between dynamic flow and thermodynamic phase transition in metallic glasses of an exothermic peak in the future [61]. Our results deepen the understanding of the theory of LLPT and its associated reentrant behavior and may help develop light-but-high strength Mg-based MGs of hierarchical structures from atomic-to-nanoscale.