LETTER doi:10.

1038/nature14674

Rejuvenation of metallic glasses by non-affine

thermal strain
S. V. Ketov1*, Y. H. Sun2*, S. Nachum2, Z. Lu3, A. Checchi2,4, A. R. Beraldin2,4, H. Y. Bai3, W. H. Wang3,
D. V. Louzguine-Luzgin1, M. A. Carpenter5 & A. L. Greer1,2

When a spatially uniform temperature change is imposed on a energy dissipation11 and of elastic modulus (,30%)12 with correlation
solid with more than one phase, or on a polycrystal of a single, lengths of 2.520 nm.
non-cubic phase (showing anisotropic expansioncontraction), In metallic glasses under elastic loading, non-affine strains lead to
the resulting thermal strain is inhomogeneous (non-affine). local structural change; this led us to question whether thermal strains
Thermal cycling induces internal stresses, leading to structural show analogous effects. In elastic deformation, the effects of non-affine
and property changes that are usually deleterious. Glasses are the strain and modulus variation are strongest for shear. For thermal
solids that form on cooling a liquid if crystallization is avoided expansion/contraction of a glass, however, the macroscopic strain
they might be considered the ultimate, uniform solids, without the is hydrostatic. In simulations of amorphous iron, the local shear
microstructural features and defects associated with polycrystals. modulus G shows a relative standard deviation of 27%; the local bulk
Here we explore the effects of cryogenic thermal cycling on glasses, modulus B shows a smaller, but still substantial, value of 18%13. This
specifically metallic glasses. We show that, contrary to the null variation is likely to be greater for the range of bonding found in a
effect expected from uniformity, thermal cycling induces rejuvena- multicomponent system14. As the coefficient of thermal expansion
tion, reaching less relaxed states of higher energy. We interpret (CTE) isother factors being constantinversely proportional to B
these findings in the context that the dynamics in liquids become (ref. 15), we expect that it also shows significant variation. Noting that
heterogeneous on cooling towards the glass transition1, and that neighbouring regions of different CTE must satisfy elastic compatibil-
there may be consequent heterogeneities in the resulting glasses. ity, atomic-level shears must develop on temperature change, and
For example, the vibrational dynamics of glassy silica at long wave- given the existence of soft spots, local structural change may occur.
lengths are those of an elastic continuum, but at wavelengths less On annealing, metallic glasses undergo thermally activated structural
than approximately three nanometres the vibrational dynamics are relaxation to states of higher density and lower enthalpy. To avoid such
similar to those of a polycrystal with anisotropic grains2. Thermal effects, thermal cycling in the present work is only from near room
cycling of metallic glasses is easily applied, and gives improvements temperature (293343 K) to lower temperatures, always far below
in compressive plasticity. The fact that such effects can be achieved the glass-transition temperature Tg near which thermal relaxation
is attributed to intrinsic non-uniformity of the glass structure, occurs (Fig. 1)16.
giving a non-uniform coefficient of thermal expansion. While As a metallic glass is heated towards Tg, there is exothermic struc-
metallic glasses may be particularly suitable for thermal cycling, tural relaxation (Fig. 2a). The relaxation spectrum (heat release rate as
the non-affine nature of strains in glasses in general deserves a function of temperature) can be detected by differential scanning
further study, whether they are induced by applied stresses or by calorimetry (DSC). While the shape of the spectrum may be of inter-
temperature change. est17, we focus here only on the overall heat of relaxation (DHrel, shaded
In glasses there are no equivalent atoms, and elastic deformation areas in Fig. 2b). Changes in DHrel have been used to characterize the
must be non-affine. For metallic glasses this has been explored through effects of thermal and mechanical treatments on metallic glasses; for
atomistic simulations3,4 and diffraction-based measurements5,6 of the example, heavy plastic deformation (shot-peening17) can induce age-
pair distribution function (see, for example, the review in ref. 7). ing (that is, relaxation to a lower-energy state) or rejuvenation.
Compared to their crystalline counterparts, metallic glasses have lower For lanthanum (La)-based metallic glasses, cycling between room
elastic moduli, associated with local atomic rearrangements occurring temperature (293 K) and liquid-nitrogen temperature (77 K) leads
even well within the nominally elastic regime8. While these rearrange- to increases in DHrel, that is, rejuvenation (Fig. 2b). For melt-spun
ments (shears) contribute to apparently elastic strain, they are not ribbons, DHrel peaks at ,10 cycles; for bulk glass, DHrel is smaller,
individually reversible7, and there is net structural change. In atomistic peaking at ,15 cycles; in each case, the maximum increase in DHrel
simulations, shear events are localized in ,1 nm soft spots with low- is ,50% relative to the as-cast sample.
frequency vibrational modes4, and can be associated with the range of We next addressed whether such changes are due to cycling, or to
density and stiffness of nearest-neighbour clusters (that is, the shell the time spent at 77 K. Ribbons were held at 77 K for 4 h, much longer
around a central atom)3,4. than the sum of the times spent at 77 K in cycling treatments, but this
Measurements of metallic glass properties do not have such fine anneal did not give a detectable change in DHrel (see Methods and
spatial resolution, but non-uniformity is still clearly detected. Strongly Extended Data Fig. 1).
and weakly bonded regions have been postulated to account for ,5 nm Because surfaces respond more rapidly than interiors, sample tem-
zones of accelerated crystallization9. The initial yield load (in nano- perature is not spatially uniform during thermal cycling; surfaces must
indentation) shows a dispersion of values associated with structural be in tension during cooling and in compression during heating. In
heterogeneity10. Mapping of heterogeneity by atomic-force micro- nano-indentation, stress cycling can harden metallic glasses18, raising
scopy, over areas much larger than in simulation, shows variations of the question of whether the DHrel changes (Fig. 2b) are due to the
1
WPI Advanced Institute for Materials Research (WPI-AIMR), Tohoku University, 2-1-1 Katahira, Aoba-ku, Sendai 980-8577, Japan. 2University of Cambridge, Department of Materials Science & Metallurgy,
27 Charles Babbage Road, Cambridge CB3 0FS, UK. 3Institute of Physics, Chinese Academy of Sciences, Beijing 100080, China. 4Department of Management and Engineering, University of Padua,
3 Stradella San Nicola, Vicenza 36100, Italy. 5University of Cambridge, Department of Earth Sciences, Cambridge CB2 3EQ, UK.
*These authors contributed equally to this work.

2 0 0 | N AT U R E | VO L 5 2 4 | 1 3 AU G U S T 2 0 1 5
G2015 Macmillan Publishers Limited. All rights reserved
 LETTER RESEARCH

Loss modulus a b
1.0
a c

0.8 As-cast (rod)

Heat flow
15 cycles (rod)

Cp
400 W mol1
Tg As-cast
(ribbon)
0.6 10 cycles
d
(ribbon)
RT 8 kW mol1
T/Tg

Exo
TX 4 J mol1 K1

0.4 100 150 200 250 300 100 150 200

Temperature (C) Temperature (C)
e c 1.2

0.2 30 J mol1
1.0 Ribbon
77 K

Hrel (kJ mol1)

b
0.8
0
0.6
Time
Rod
Figure 1 | Thermal cycling of metallic glasses. a, At high fractions of the 0.4
glass-transition temperature, Tg, to as low as 0.6 Tg, metallic glasses can
undergo a and b relaxations, giving peaks in loss modulus as a function of T/Tg 0.2
as shown by the curve16. b, The thermal cycling in the present work explores 0 5 10 15 20 25
lower temperatures. Samples are cycled from near room temperature, Number of cycles
293343 K (RT), to liquid-nitrogen temperature (77 K). The ranges of T/Tg
Figure 2 | Differential scanning calorimetry (DSC) of melt-spun ribbons of
reflect both range of temperature and the range of Tg for the metallic glasses,
La55Ni20Al25 and bulk rods of La55Ni10Al35 metallic glasses. a, DSC traces
from 474 K for La55Ni10Al35 (at.%) up to 700 K for Cu46Zr46Al7Gd1.
on heating show an exotherm before the glass transition at Tg (here for ribbon).
ce, Schematic depictions of the degree of heterogeneity in a metallic glass in
b, By subtraction of successive traces of specific heat Cp (see Methods), the
the as-cast state (c), and after increasing numbers of thermal cycles (d, e). The
heat of relaxation, DHrel, associated with the exotherm can be determined
population and intensity of soft spots (dark), with lower elastic stiffness and
(shaded areas) and is increased after room temperature77 K thermal cycles. In
higher CTE, increases with cycling. The scale of these heterogeneities is not
both a and b, vertical bars give units and scales for the y axes. c, For ribbons,
given by the present results (so no scale bars are given), but mapping of elastic
DHrel peaks at ,10 room temperature77 K cycles; for discs (250500 mm
properties11,12 suggests characteristic lengths of ,10 nm.
thick) cut from the bulk rod, DHrel is lower and peaks at ,15 cycles; the data
points are for individual measurements and the changes lie outside the error
temperature cycling itself, or to the associated stresses. As-cast rods of range of 630 J mol21 (standard deviation estimated as in Methods).
La55Ni10Al35 bulk metallic glass (BMG) were subjected to room tem-
perature77 K cycles at the same time as thin discs cut from the rods.
As detailed in Methods, the changes in DHrel are indepen- less sensitive to local heterogeneities than is Py. The Young modulus,
dent of sample size, showing that stresses generated during cycling determined on unloading, shows similar behaviour to H (Extended
have a negligible effect (Extended Data Fig. 2). Thermal calculations Data Fig. 4).
(Extended Data Table 1) suggest that the stresses themselves are The distribution curves of Py and H confirm changes related to the
negligible, consistent with this conclusion. number of cycles rather than the holding time at 77 K. The marked
Many aspects of thermal cycling could be tailored: upper and lower reduction in Py induced by cycling is accompanied by a reduction in
temperatures, holding times at these temperatures, rates of cooling initial yield increments (Dh, Extended Data Fig. 4a).
and heating, number of cycles. We report preliminary studies in Metallic glasses show exceptionally high yield stress and yield
Methods; our focus here, however, is on property changes induced strain19, and can be very tough20, but there is a clear need to improve
by thermal cycling. their plasticity. Annealing reduces their plasticity, often inducing brit-
In instrumented indentation (nano-indentation) of metallic tleness. Figures 2 and 3 show property changes opposite to those
glasses10,18, initial yielding is indicated by a sharp pop-in (increase expected from annealing, suggesting that thermal cycling gives reju-
in indentation depth h) on the load-h curve (Extended Data Fig. 3), venation. The effects on macroscopic plasticity are explored for two
corresponding to shear-banding onset. Cumulative distributions of BMG compositions tested in uniaxial compression (Fig. 4). The
initial yield pressure Py (Fig. 3a) for La55Ni20Al25 glass ribbon in three plastic strain of 1.5-mm-diameter samples of Zr62Cu24Fe5Al9 is
states permit the effects of a longer hold at 77 K to be compared with improved (from 4.9% to 7.6%) by successive thermal cycles (Fig. 4a).
cycling. The median value of Py, 2.98 GPa in the as-cast sample, Improvements in plasticity, smaller in absolute terms (for reasons
decreases by 3% after 10-min hold at 77 K, and by a further 17% after given in Methods), but larger relatively, are seen in larger-diameter
ten room temperature77 K (1 min) cycles. After the cycling treat- samples of the same glass (Fig. 4b), which also show a reduction
ments the relative width (1st to 9th decile) of the Py distribution in microhardness (by 4%, inset), similar to that for the median
(615%) is greater than in the as-cast sample (67%), suggesting that hardness of the La-based glass in Fig. 3b. Cu46Zr46Al7Gd1 BMG
cycling induces greater heterogeneity. (1.5 3 1.5 3 3.0 mm3 cuboid) shows an increase in plastic strain from
The corresponding distributions of hardness H (Fig. 3b), deter- 1.4% in the as-cast state to 5.1% after ten room temperature77 K
mined from h at full load, show similar trends. The median H of the cycles (Fig. 4c). A sample fully relaxed by annealing is brittle and is
as-cast sample, 2.65 GPa, is 1% lower after 10 min at 77 K, and a not successfully rejuvenated by cycling, but a strong rejuvenation effect
further 4% lower after ten room temperature77 K cycles. For a given is possible after partial thermal relaxation, when thermal cycling gives
sample, the H distribution is narrower than that for Py; for example, a plastic strain similarly improved to that of as-cast samples.
after cycling, the relative width is only 63% (one-fifth that of Py). Well below Tg, plastic flow in metallic glasses is localized into thin
Evidently H, reflecting conditions for continuance of flow, is much shear bands. Macroscopic plasticity is limited by premature failure on
1 3 AU G U S T 2 0 1 5 | VO L 5 2 4 | N AT U R E | 2 0 1
G2015 Macmillan Publishers Limited. All rights reserved
 RESEARCH LETTER

a 100 a b
8.0
As-cast
Plastic strain 1.5 mm rod
10-min hold 60 cycles
6.0

Engineering stress

Plastic strain (%)

460
30 cycles

Microhardness (HV)
80 10-min hold 450 2 mm rod
and 10 cycles 10 cycles
440
5 cycles 4.0 430
As-cast 420
0 10 20 30 40 50 60
60
Fraction (%)

Number of cycles
2.0 2 mm rod

1% 2.5 mm rod

40 0
Engineering strain 0 10 20 30 40 50 60
Number of cycles
c d
20
Fully relaxed + 10 cycles

Engineering stress
Partially relaxed + 10 cycles

0 10 cycles
1.5 2.0 2.5 3.0 3.5
Initial yield pressure, Py (GPa) As-cast
b 100
As-cast 1% 5 m
10-min hold Engineering strain
80
10-min hold
and 10 cycles
Figure 4 | Improved plasticity after thermal cycling. a, 1.5-mm-diameter rod
samples of Zr62Cu24Fe5Al9 BMG (yield stress 1.68 6 0.05 GPa) show an
increase in plastic strain under uniaxial compression when treated with 338 K
60 to 77 K thermal cycles. b, The compressive plasticity of this glass is strongly
Fraction (%)

dependent on sample dimension, but for all the diameters tested, thermal
cycling leads to an increase in plastic strain, accompanied (inset) by a reduction
in microhardness (in kg mm2; each data point is the average of 20
40 measurements). c, 1.5 3 1.5 3 3.0 mm3 cuboids of Cu46Zr46Al7Gd1 BMG (yield
stress 1.61 6 0.04 GPa), both as-cast and annealed (1.0 h at 400 uC) to
partial relaxation, show an increase in plastic strain under uniaxial compression
after ten room temperature77 K cycles; for full relaxation (1.5 h at 400 uC), the
20 sample remains brittle. d, For the Cu46Zr46Al7Gd1 BMG samples, scanning
electron microscopy shows that the population density of shear bands near the
dominant shear band is lower in an as-cast sample (top) than in a similar
sample tested after ten room temperature77 K cycles (bottom). In a and
0
2.4 2.5 2.6 2.7
c, horizontal bars indicate the scale and units for the x axes; quoted yield stress
gives a guide to the y axes.
Hardness, H (GPa)

Figure 3 | Cumulative distributions of (a) initial yield pressure Py and glassy after cycling; Extended Data Fig. 6), and the effects are more
(b) hardness H. La55Ni20Al25 metallic glass ribbon is tested in instrumented analogous to those induced in single-phase polycrystalline non-cubic
indentation, to a maximum load of 40 mN, in three states: as-cast, after a metals, for example, thermal-cycling growth in uranium23 and
10-min hold at 77 K, and after a further ten room temperature77 K cycles each
with 1 min hold. The cycling reduces Py and H, and widens the distribution
increased dislocation density in zirconium24. The widespread use of
of their values. The widening is taken to represent increasing heterogeneity of DCT25 does, however, suggest that cycling to cryogenic temperatures
the glass structure (Fig. 1ce). may be a practicable process.
The changes in DHrel (Fig. 2c) show rejuvenation followed by age-
ing, opposing trends known in processes such as irradiation or plastic
one or a few dominant shear bands. The plasticity can be improved by deformation, which introduce both damage (raising the energy of the
proliferation of shear bands, making the overall flow more uniform21. system) and mobility (enabling relaxation to lower energy). Shot-
The shear-band spacing of ,10 mm in the as-cast sample (Fig. 4d) is peening increases DHrel in an annealed BMG, but for an as-cast
reduced to ,2.5 mm in the thermally cycled sample, suggesting greater BMG even this intense plastic deformation reduces DHrel (ref. 17).
ease of shear-band initiation, consistent with reductions in Py (Fig. 3a) Thermal cycling increases DHrel even for a rapidly quenched metallic
and in initial yield displacements (Extended Data Fig. 4a). Resonant glass. The maximum increase in DHrel (Fig. 2c), 340 J mol1, is (cor-
ultrasound spectroscopy (RUS) was used to measure the elastic moduli recting for differing DSC heating rates) ,60% of the increase in DHrel
of Cu46Zr46Al7Gd1 BMG, which show no detectable change on ther- in a Zr-based BMG induced by one rotation in high-pressure torsion26,
mal cycling (Extended Data Fig. 5). and reductions in hardness (Figs 3b and 4b and ref. 26) are, pro rata,
Across a range of metallic glass compositions, melt-spun and bulk, similar. The increases in DHrel (Fig. 2c) are very similar to those seen in
thermal cycling induces changes in properties. As the cycling is down elastostatic loading27; it is remarkable that both these apparently mild
to 77 K, these changes can be compared with the property improve- processes give effects comparable with heavy plastic deformation, sug-
ments (hardness, wear resistance) obtained by deep cryogenic treat- gesting that non-affine strains in the nominally elastic regime are
ment (DCT) of steels22. DCT, however, works through phase changes efficient in generating structural damage (disordering).
(transformation of residual austenite, precipitation of fine carbides) on Although continued cycling leads to reversal in the DHrel increase
holding at low temperature. In the present case, there is no discernible (Fig. 2c), the microhardness reduction appears to saturate (Fig. 4b
phase change (X-ray diffraction confirms that the samples remain fully inset) while the plasticity continues to improve (Fig. 4a, b); the onset
2 0 2 | N AT U R E | VO L 5 2 4 | 1 3 AU G U S T 2 0 1 5
G2015 Macmillan Publishers Limited. All rights reserved
 LETTER RESEARCH

of yield (Fig. 3a) is facilitated more than general yielding (Fig. 3b) and 13. Egami, T. & Srolovitz, D. Local structural fluctuations in amorphous and liquid
metals: a simple theory of the glass transition. J. Phys. F Met. Phys. 12, 21412163
the effect on high-frequency (MHz in RUS) macroscopic elastic mod- (1982).
uli is negligible (Extended Data Fig. 5). These contrasts for different 14. Mendelev, M. I., Ott, R. T., Kramer, M. J. & Sordelet, D. J. Determining strain in
properties lead us to suggest that thermal cycling introduces hetero- amorphous alloys: Uncertainties with analyzing structural changes during
deformation. J. Appl. Phys. 105, 023509 (2009).
geneities (soft spots)4 that are particularly effective in initiating flow
15. Gruneisen, E. Theorie des festen Zustandes ein-atomiger element. Ann. Phys. 344,
and improving plasticity, with lesser effects on the average structure of 257306 (1912).
the metallic glass. That the effects (at least for DHrel, Fig. 2c) are greater 16. Yu, H. B., Wang, W. H. & Samwer, K. The b relaxation in metallic glasses: an
in melt-spun ribbon than in BMG of similar composition suggests that overview. Mater. Today 16, 183191 (2013).
17. Concustell, A., Mear, F. O., Surinach, S., Baro, M. D. & Greer, A. L. Structural
pre-existing heterogeneity (greater in a less-relaxed metallic glass) relaxation and rejuvenation in a metallic glass induced by shot-peening. Philos.
helps cycling to introduce more heterogeneity; this is expected if, as Mag. Lett. 89, 831840 (2009).
we speculate, the effect is due to non-affine thermal strains arising 18. Packard, C. E., Homer, E. R., Al-Aqeeli, N. & Schuh, C. A. Cyclic hardening of metallic
from the heterogeneity itself. The present results, however, do not glasses under Hertzian contacts: Experiments and STZ dynamics simulations.
Phil. Mag. 90, 13731390 (2010).
identify any characteristic length scale for the heterogeneity. 19. Ashby, M. F. & Greer, A. L. Metallic glasses as structural materials. Scr. Mater. 54,
Rejuvenation of metallic glasses, with improved plasticity, has 321326 (2006).
been achieved by elastostatic loading27,28, ion irradiation29 and plastic 20. Demetriou, M. D. et al. A damage-tolerant glass. Nature Mater. 10, 123128 (2011).
deformation21. In comparison with these methods, thermal cycling is 21. Greer, A. L., Cheng, Y. Q. & Ma, E. Shear bands in metallic glasses. Mater. Sci. Eng.
Rep. 74, 71132 (2013).
attractive: it is non-destructive, not involving shape change; unlike an 22. Preciado, M. & Pellizarri, M. Influence of deep cryogenic treatment on the thermal
applied elastostatic stress, it is isotropic and cannot introduce aniso- decomposition of FeC martensite. J. Mater. Sci. 49, 81838191 (2014).
tropy; it can be applied to any sample (thin film, ribbon or bulk) and 23. Ma, B. M. Nuclear Reactor Materials and Applications Sect. 6.7.2., 149151 (Van
Nostrand, 1983).
repeated as necessary; it changes the whole sample (not just the surface 24. Yuan, C. et al. Effects of deep cryogenic treatment on the microstructure and
as, for example, in shot-peening or ion irradiation, or only inside shear mechanical properties of commercial pure zirconium. J. Alloy. Comp. 619,
bands); and it is controllable, can be performed in situ, induces no 513519 (2015).
macroscopic residual stresses, involves strains far below the elastic 25. Kalsi, N. S., Sehgal, R. & Sharma, V. S. Cryogenic treatment of tool materials:
a review. Mater. Manuf. Process. 25, 10771100 (2010).
limit, and induces no macroscopic plastic flow. 26. Meng, F., Tsuchiya, K., Seiichiro, I. & Yokoyama, Y. Reversible transition of
deformation mode by structural rejuvenation and relaxation in bulk metallic glass.
Online Content Methods, along with any additional Extended Data display items
Appl. Phys. Lett. 101, 121914 (2012).
and Source Data, are available in the online version of the paper; references unique
27. Lee, S. C., Lee, C. M., Yang, J. W. & Lee, J. C. Microstructural evolution of an
to these sections appear only in the online paper.
elastostatically compressed amorphous alloy and its influence on the mechanical
properties. Scr. Mater. 58, 591594 (2008).
Received 26 January; accepted 8 June 2015. 28. Ke, H. B., Wen, P., Peng, H. L., Wang, W. H. & Greer, A. L. Homogeneous deformation
of metallic glass at room temperature reveals large dilatation. Scr. Mater. 64,
1. Ediger, M. D. Spatially heterogeneous dynamics in supercooled liquids. Annu. Rev. 966969 (2011).
Phys. Chem. 51, 99128 (2000). 29. Magagnosc, D. J. et al. Effect of ion irradiation on tensile ductility, strength and
2. Baldi, G. et al. Emergence of crystal-like atomic dynamics in glasses at the fictive temperature in metallic glass nanowires. Acta Mater. 74, 165182 (2014).
nanometer scale. Phys. Rev. Lett. 110, 185503 (2013).
3. Ding, J., Cheng, Y. Q. & Ma, E. Correlating local structure with inhomogeneous Acknowledgements This research was supported by the World Premier International
elastic deformation in a metallic glass. Appl. Phys. Lett. 101, 121917 (2012). Research Center Initiative (WPI), MEXT, Japan, by NSF China and MOST 973 China,
4. Ding, J., Patinet, S., Falk, M. L., Cheng, Y. & Ma, E. Soft spots and their structural and by the Engineering and the Engineering and Physical Sciences Research Council,
signature in a metallic glass. Proc. Natl Acad. Sci. USA 111, 1405214056 (2014). UK (Materials World Network project). Y.H.S. acknowledges support from a China
5. Poulsen, H. F., Wert, J. A., Neuefeind, J., Honkimaki, V. & Daymond, M. Measuring Scholarship Council (CSC) scholarship. M. L. Falk, T. C. Hufnagel, E. Ma, D. B. Miracle
strain distributions in amorphous materials. Nature Mater. 4, 3336 (2005). and J. Orava are thanked for discussions. All data accompanying this publication are
6. Hufnagel, T. C., Ott, R. T. & Almer, J. Structural aspects of elastic deformation of a directly available within the publication and the accompanying Extended Data figures
metallic glass. Phys. Rev. B 73, 064204 (2006). and tables.
7. Egami, T., Iwashita, T. & Dmowski, W. Mechanical properties of metallic glasses.
Metals 3, 77113 (2013). Author Contributions Preparation of metallic glasses was by S.V.K. and Z.L. The
8. Weaire, D., Ashby, M. F., Logan, J. & Weins, M. J. On the use of pair potentials to thermal-cycling treatments were performed by Y.H.S., S.V.K., A.R.B. and A.C.;
calculate the properties of amorphous metals. Acta Metall. 19, 779788 (1971). calorimetry by Y.H.S., S.V.K., Z.L., A.C. and A.R.B.; nano-indentation by S.N. and A.R.B.;
9. Ichitsubo, T. et al. Microstructure of fragile metallic glasses inferred from compression and microhardness tests by S.V.K., Y.H.S., A.R.B. and A.C.; X-ray diffraction
ultrasound-accelerated crystallization in Pd-based metallic glasses. Phys. Rev. Lett. by S.V.K. and Y.H.S.; and resonant ultrasound spectroscopy by Y.H.S., M.A.C., A.R.B. and
95, 245501 (2005). A.C. Direction of the work was by D.V.L.-L. (Sendai), H.Y.B. and W.H.W. (Beijing), and
10. Packard, C. E., Franke, O., Homer, E. R. & Schuh, C. A. Nanoscale strength A.L.G (Cambridge). A.L.G. led the project and the writing of the paper. All authors
distribution in amorphous versus crystalline metals. J. Mater. Res. 25, 22512263 contributed to interpretation and presentation of the results.
(2010).
11. Liu, Y. H. et al. Characterization of nanoscale mechanical heterogeneity in a Author Information Reprints and permissions information is available at
metallic glass by dynamic force microscopy. Phys. Rev. Lett. 106, 125504 (2011). www.nature.com/reprints. The authors declare no competing financial interests.
12. Wagner, H. et al. Local elastic properties of a metallic glass. Nature Mater. 10, Readers are welcome to comment on the online version of the paper. Correspondence
439442 (2011). and requests for materials should be addressed to A.L.G. ([email protected]).

1 3 AU G U S T 2 0 1 5 | VO L 5 2 4 | N AT U R E | 2 0 3
G2015 Macmillan Publishers Limited. All rights reserved
 RESEARCH LETTER

METHODS of geometry comparable to the samples in the present work32; it occurs at tem-
Metallic glasses in this study. Sample preparation and property measurements peratures ranging from 150 to 100 K. Thus a room-temperature sample immersed
took place in three laboratories. Kinetic phenomena in glasses are best compared in liquid nitrogen spends around 75% of the temperature range down to 77 K with
on a homologous temperature scale, using the glass-transition temperature Tg as the relatively poor heat transfer associated with the film-boiling regime.
the normalizing parameter. The selected glasses allow studies over a range of T/Tg: We used materials parameters determined for Vitreloy 1; Zr-based BMGs of
for example, room temperature (293 K) covers the range of T/Tg 5 0.420.62 for this kind are among those for which thermal-cycling treatments might be
these glasses. (The values of Tg were determined using differential scanning calori- put into practice. For Vitreloy 1, the thermal conductivity decreases with decreas-
metry at standard heating rates of 20 and 40 K min1.) The four selected glasses are: ing temperature33 and starts to level off below room temperature. At 300 K,
(1) Cu46Zr46Al7Gd1 rod, 3 mm diameter (Tg 5 700 K (ref. 30)); (2) La55Ni20Al25 k 5 4.59 W K21 m21; we use a rough estimate of k 5 4 W K21 m21 for the tem-
(at.%) melt-spun ribbon 40 mm thick (Tg 5 475 K); (3) La55Ni10Al35 rod 3 mm perature range in the present work. Extended Data Table 1 gives the estimates of Bi
diameter (Tg 5 474 K); and (4) Zr62Cu24Fe5Al9 rods, 1.5, 2.0 or 2.5 mm diameter for the various sample geometries of interest.
(Tg 5 658 K). Master alloys were prepared by arc-melting 34N pure elements. We can see that for nearly all of these cases Bi , 0.1, and therefore temperature
Glassy ribbons, 40 mm thick, were produced by melt-spinning, under argon atmo- gradients (and thermal stresses) in the samples are negligible. Only for the largest-
sphere, on to a single copper wheel at 4,000 r.p.m. Bulk metallic glass (BMG) rods, diameter cylinders in the final (nucleate boiling) stages of cooling is there any
with diameter 1.53.0 mm and length up to 80 mm, were cast, under Ti-gettered significant deviation from Newtonian cooling. These calculations support the
argon atmosphere, into water-cooled Cu-moulds, using induction melting and experimental work (Extended Data Fig. 2 and associated discussion, see below)
suction casting. suggesting that sample size is not a significant factor in the thermal-cycling effect,
Thermal cycling. The La-based and Cu46Zr46Al7Gd1 samples were inserted into and reinforce the conclusion that the effect is due to the temperature changes
liquid nitrogen for 1 min, then heated by a hair-dryer set at room temperature, and themselves and not due to any stresses arising from non-uniform temperature
held for 1 min; samples were treated with up to 25 such cycles. The Zr62Cu24Fe5Al9 profiles in samples during cooling and heating.
samples were inserted into liquid nitrogen for 1 min, then into ethanol at It is of interest to consider the possible extent of thermal stresses associated
333343 K for 1 min; samples were treated with up to 60 such cycles. with non-uniform sample temperature. Taking the value34 determined for the
The paper reports only results from cycling between near room temperature Zr-based BMG Vitreloy 106, the linear coefficient of thermal expansion is
and liquid-nitrogen temperature (77 K). Keeping the upper temperature fixed a 5 8.7 3 1026 K21. For a room-temperature sample immersed in liquid nitrogen,
(near room temperature), it is expected that there would be an optimum lower the maximum possible temperature difference during cooling is between 293 K
temperature for maximum effect of thermal cycling (at given cooling and heating (room temperature, possible at the sample centre) and 77 K (possible at the sample
rates, and hold times at the upper and lower temperatures): for higher tempera- surface). For this temperature difference, the linear thermal strain is 0.19%, which
tures closer to room temperature, there would be less driving force for structural would give internal stresses of the order of 10% of the macroscopic yield stress. In
change; for lower temperatures further from room temperature, the rate of struc- practice, as highlighted by the low values for the Biot number in Extended Data
tural change near the lower temperature would be lower. Table 1, any thermal stresses would be much lower than this value. Rejuvenation of
In addition to cycling down to 77 K, melt-spun ribbons of La55Ni20Al25 glass metallic glasses by elastostatic loading28 typically involves much higher stresses (up
were cycled down to dry-ice (CO2) temperature (195 K) and to liquid-helium to 95% of the yield stress) for much longer times (tens of hours, rather than a few
temperature (4.2 K) to explore the role of the lower temperature. The state of minutes). Again, it seems that stress effects in the current thermal cycling should
the samples was assessed using DSC (with heating rates from 20 to 40 K min1). be negligible.
For a given type of sample, the heat of relaxation DHrel is lower for higher heating Differential scanning calorimetry. This was performed (as in Fig. 2) using a
rate; it is therefore appropriate to normalize the DHrel values relative to that for the Q2000 DSC (TA Instruments). Samples were heated at 20 K min1 from room
as-cast glass measured at the same DSC heating rate. Measurements were made for temperature to 503 K (into the supercooled liquid state), held for 2 min, then
samples: (1) held at 195 K for 100 min, for which the ratio of DHrel to that of the as- cooled to room temperature at 20 K min1. A second cycle using the same pro-
cast glass is 1.23; (2) treated with ten cycles of 1 min at 195 K, for which the ratio is cedure was used as the baseline for subtraction from the first cycle. The heat of
1.24; (3) treated with ten cycles of 10 min at 195 K (ratio 5 1.02); (4) treated with relaxation DHrel is calculated from the area between the two curves, from the onset
ten cycles of 1 min at 77 K (ratio 5 1.55); and (5) treated with 13 cycles to 4.2 K of relaxation to the glass transition35. Additional data were acquired using a Perkin
(ratio is 1.37). Given the low number of samples tested, there is significant uncer- Elmer DSC 8500 at a heating rate of 40 K min1.
tainty in these relative values (615%); nevertheless, it seems that, within the Extended Data Figure 1a shows examples of DSC traces used to assess the
limited range of these tests, cycling down to 77 K produces the strongest effect. reproducibility of measurements of the heat of relaxation DHrel (using the base-
A Ti-based BMG cycled up to 400 times between 77 K and 423 K (to simulate line-subtraction method outlined above). Six samples of as-cast ribbons were
conditions on the surface of a spacecraft in low earth orbit) showed slight relaxa- tested, giving an average value of DHrel 5 740 J mol1, with a standard deviation
tion rather than rejuvenation31. This suggests that there is a significant annealing of 30 J mol1 (that is, 64%).
effect during the holds (25 min each) at the upper temperature, which itself is As noted in the paper, it is important to know whether the effects of thermal
higher than in our work (and corresponds to T/Tg 5 0.68, inside the range assoc- cycling between near room temperature and 77 K are due just to the time spent at
iated with b relaxation, see Fig. 1a in main text). 77 K. Samples of La55Ni20Al25 glassy ribbons were held at 77 K for 4 h (a time much
Stresses associated with thermal cycling. As noted in the paper, it is important to longer than the total spent at 77 K in the course of ten room temperature77 K
understand whether the effects of thermal cycling are due to the temperature cycles). Within experimental error (Extended Data Fig. 1b), such holds at 77 K
changes themselves or to stresses arising from non-uniform temperature in the appear to have no effect on DHrel. Of course, a hold at 77 K represents a single
samples during cooling and heating. Whether the temperature is significantly non- cycle. As shown by the nano-indentation results in Fig. 3 in the main paper, even
uniform can be assessed using the dimensionless Biot number, one cycle can have a measurable effect on properties. In the present case of DHrel,
however, any change as a result of one cycle is within the experimental uncertainty.
hL As noted in the main paper and discussed above in the Methods, it is necessary
Bi~ 1
k to examine whether the effects of thermal cycling are dependent on sample size.
where h is the heat-transfer coefficient between the sample and the surrounding The non-uniform temperature in a larger sample could give rise to stresses and the
medium, L is a characteristic sample dimension (usually taken to be the ratio of cycling of those stresses could have effects. As-cast rods (3 mm diameter) of
sample volume to surface area), and k is the thermal conductivity of the sample La55Ni10Al35 BMG were subjected to room temperature77 K cycles at the same
material. For a flat plate, L is the half-thickness; for a cylinder the half-radius. If time as thin (250500 mm thick) discs pre-cut from the rods. After cycling, discs
Bi , 0.1, the temperature differences within the sample are negligible compared to were cut from the centres of the bulk rods, so that samples of the same glass and
the step in temperature between the sample and its surroundings (Newtonian same DSC-sample size could be compared after thermal cycling in two different
cooling). Conversely, if Bi . 1, there are significant temperature gradients in the geometries (thin disc and bulk). Extended Data Figure 2 shows that the changes in
sample. We examine the conditions for samples immersed in liquid nitrogen. DHrel are essentially the same for the samples thermally cycled in disc and in bulk
Cooling in liquid nitrogen has been widely studied, particularly in connection form, suggesting that the stresses generated during cycling have a negligible effect.
with cryopreservation32. The French straw sample containers used are cylinders The two sets of data are combined in Fig. 2c.
of 2.8 mm diameter, and so are very similar in dimension to the cast BMG rods in Instrumented indentation. This was performed using an XP Nanoindenter
the current work. On first immersion, there is film boiling, in which a film of (MTS Systems Corp.) at room temperature under load control up to a maximum
vapour limits the heat transfer; in this regime32 h 5 148 W m22 s21. As the sample load (Fmax) of 40 mN. A diamond spherical indenter of tip radius 8.074 mm was
cools, there is a transition to the nucleate-boiling regime32 in which h 5 1,355 W used, its area function calibrated using fused silica and sapphire standards.
m22 s21. The transition point has been measured for a variety of metallic samples Loading and unloading rates were 0.5 mN s1; the drift rate was maintained below

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

0.07 nm s1. Samples were loaded to Fmax, then unloaded to 0.05Fmax and main- underestimate the Youngs modulus of the materials. The curves are used only
tained at that load for 60 s for measurement of thermal drift. The cumulative to extract values for the plastic strain before catastrophic failure. Figure 4b collects
distribution curves (Fig. 3) include 3948 data points. Initial yield events data for rod samples of Zr62Cu24Fe5Al9 BMG of three diameters: 1.5 mm, 2.0 mm
(pop-ins) were identified directly from the Fh curve and the associated spike and 2.5 mm, and corresponding heights 3.0 mm, 4.0 mm and 5.0 mm. The stress
in indenter tip velocity, or from the first deviation from the Hertzian elastic train curves for 1.5 mm samples are shown in Fig. 4a; those for 2.0 mm and 2.5 mm
solution fitted to the Fh curve. Hardness was determined by the standard samples in Extended Data Fig. 7.
Oliver and Pharr method36, on unloading at Fmax. As has been much discussed21,39, larger samples are more susceptible to cata-
Tests were conducted on melt-spun ribbons of La55Ni20Al25 glass under the strophic failure through shear localization. Such size effects are extrinsic, and relate
conditions noted above. Typical load-displacement curves (Extended Data Fig. 3) only to the mechanical tests themselves. In addition, there is an intrinsic effect:
show pop-ins (that is, the transition from purely elastic to elastic-plastic deforma- larger samples have a lower cooling rate during casting and so are more relaxed
tion) with a moderate sharpness that is not affected by thermal cycling. The pop- and tend to be more brittle in the as-cast state.
ins are characterized by the values of initial yield load Fy and initial yield displace- Resonant ultrasound spectroscopy. RUS measurements were conducted as in
ment Dh. Initial yield pressure Py is calculated from: earlier work40,41. Measurements were at room temperature on 1.5 3 1.5 3 3.0 mm3
Fy cuboids cut from the as-cast 3-mm-diameter rod of Cu46Zr46Al7Gd1 BMG. For
Py ~ 2 each sample, a resonance spectrum was collected in the range 0.052.0 MHz with
pa2y
65,000 data points. To detect changes in shear and bulk modulus, 10,000 data
where ay is the contact radius at initial yield. The reduced or indentation modulus points were collected in each case in the ranges 0.9000.910 MHz and 0.9780.982
Er is determined by the standard method36,37, on unloading from Fmax. Er is also MHz, respectively. Within experimental error (less than 60.1% for the shear
determined by using the Hertzian elastic solution to fit the shape of the load- modulus), room temperature77 K thermal cycling has no effect on the elastic
displacement curve up to the first pop-in. Er is given by38: moduli of the glass (Extended Data Fig. 5). This suggests that the reductions
1 (1{vi2 ) (1{v2 ) in Youngs modulus (Extended Data Fig. 4b, c) and correspondingly hardness
~ z 3 (Fig. 3b) seen under quasi-static conditions are largely attributable to anelastic
Er Ei E
(that is, time-dependent elastic) strains.
where E and n are, respectively, the Young modulus and Poisson ratio of the glass, Dynamic mechanical analysis. This technique (not discussed in the paper), per-
while the corresponding values36 for the diamond indenter tip are taken to be mitting measurement of the loss and storage moduli of a material, was applied
Ei 5 1,141 GPa and ni 5 0.07. Using equation (3), values of E were obtained. (using methods reported previously42) to ribbons of La55Ni20Al25 glass (Extended
Tests were conducted on ribbon samples in three states: (1) as-cast; (2) after one Data Fig. 8) to explore the nature of the rejuvenation effect. The loss modulus
cycle from room temperature to a 10-min hold 77 K and back to room temper- shows a sharp maximum near the glass transition, associated with a relaxation.
ature; and (3) after a further ten cycles to 77 K each with 1 min hold. The effect of a After ten room temperature77 K cycles, the maximum is lowered by 2.3 K. This
longer hold at 77 K can thus be compared with cycling. All metallic-glass samples, earlier onset of softening on heating is consistent with a rejuvenated state.
as-cast or treated, are found to have some variation in properties over length scales The broad maximum at lower temperature (centred around 115 uC, Extended
of millimetres. For this reason it is important to compare data taken from a given Data Fig. 8) shows b relaxation. The b relaxation in metallic glasses has been
region of the sample surface. In the present work, indentations were made in the linked to mechanical properties42,43. In view of the improved plasticity seen after
same region (within 200 mm) of the given sample surface in the three successive thermal cycling (Fig. 4 and Extended Data Fig. 7), it is surprising that the effect of
states. Extended Data Table 2 collects data on the instrumented-indentation tests thermal cycling on the b relaxation is so slight (inset in Extended Data Fig. 8),
in the present work. Cumulative distributions, such as those shown in Fig. 3, are especially compared to the clear effect on the a relaxation.
characterized by the median value and by the 1st and 9th deciles. X-ray diffraction. There are reports that under cyclic mechanical loading at room
Extended Data Figure 4a shows the distribution of values of initial yield pressure temperature, in the elastic regime, a Zr-based BMG can undergo some crystalliza-
Py and initial yield displacement Dh for the three states of the La55Ni20Al25 glass tion44. Deep cryogenic treatment of steels certainly induces phase changes22,25. In
ribbons. On cycling, both Py and Dh are reduced. There is a correlation, as the present work, X-ray diffraction (BraggBrentano geometry, Bruker D8 instru-
expected, between higher values of Py and higher values of Dh. The spread in ment, Cu Ka radiation) was used to check that the samples remained fully glassy
values of Dh is particularly wide. After ten room temperature77 K cycles, there throughout, and especially after thermal cycling. Representative diffractograms for
is a marked reduction in the incidence of larger values: the initial yield events are ribbon and bulk samples are shown in Extended Data Fig. 6. We found no evidence
smaller, reflecting less catastrophic shear banding. for any crystallization induced by the thermal-cycling treatments. The property
The Youngs modulus of the glass E determined by unloading from full load changes suggest that the cycling does induce changes in the glassy structure but,
(Extended Data Fig. 4b) shows similar effects to those found for hardness H typical of such structural relaxation, these changes are too subtle to be detectable
(Fig. 3b). The median value of Eunload in the as-cast sample decreases by 0.6% with the simple X-ray methods used in the present work.
after 10 min hold at 77 K (compared to a 0.9% decrease in H), and by a further 7%
Other methods. Microhardness was measured on an Akashi MVK-HVL hardness
after ten room temperature77 K (1 min) cycles (compared to 4% for H). After
testing machine, with 20 measurements for each data point. Scanning electron
the cycling treatments the relative width (1st to 9th decile) of the E distribution
microscopy (secondary-electron imaging) was performed using a JEOL 5200
(62.4%, compared to 63.1% for H) is greater than in the as-cast sample (61.4%,
instrument.
compared to 61.3% for H).
The Youngs modulus determined from the Hertzian fit of the Fh curve up to 30. Wang, W. H. Roles of minor additions in formation and properties of bulk metallic
the first pop-in (Extended Data Fig. 4c) has a median value ,5% lower than that glasses. Prog. Mater. Sci. 52, 540596 (2007).
determined by unloading. The median value of EHertz in the as-cast sample 31. Wang, X., Shao, Y., Gong, P. & Yao, K. F. The effect of simulated thermal cycling on
decreases by 2.4% after 10-min hold at 77 K, and by a further 8.8% after ten room thermal and mechanical stability of a Ti-based bulk metallic glass. J. Alloy. Comp.
575, 449454 (2013).
temperature77 K (1 min) cycles. The relative width of the EHertz distribution in
32. Victoria Santos, M., Sansinena, M., Chirife, J. & Zaritzky, N. Determination of heat
the as-cast sample (61.6%) is similar to that for Eunload, but this width increases transfer coefficients in plastic French straws plunged in liquid nitrogen.
more (to 67.3%) after the cycling treatments. This suggests that the elastic prop- Cryobiology 69, 488495 (2014).
erties up to the first pop-in are more sensitive to local variations in the sample. 33. Yamasaki, M., Kagao, S., Kawamura, Y. & Yoshimura, K. Thermal diffusivity and
Compression tests. These were made on Instron 5581 and Hounsfield 25 kN conductivity of supercooled liquid in Zr41Ti14Cu12Ni10Be23 metallic glass. Appl.
machines, loading at a strain rate of 5 3 1024 s1. The Zr62Cu24Fe5Al9 samples Phys. Lett. 84, 46534655 (2004).
34. Conner, R. D., Choi-Yim, H. & Johnson, W. L. Mechanical properties of
were cylindrical rods of 2:1 height:diameter, with diameter of 1.5, 2.0 or 2.5 mm.
Zr57Nb5Al10Cu15.4Ni12.6 metallic glass matrix particulate composites. J. Mater.
The Cu46Zr46Al7Gd1 samples were 1.5 3 1.5 3 3.0 mm3 cuboids cut from the Res. 14, 32923297 (1999).
original cast rod of 3 mm diameter. All samples were polished to a mirror finish. 35. Mear, F. O., Lenk, B., Zhang, Y. & Greer, A. L. Structural relaxation in a heavily cold-
The gradients of the elastic portions of the stressstrain curves in Fig. 4a are worked metallic glass. Scr. Mater. 59, 12431246 (2008).
affected by the machine compliance and underestimate the Youngs modulus. 36. Oliver, W. C. & Pharr, G. M. An improved technique for determining hardness and
The curves are used only to extract values for the plastic strain as shown. The elastic modulus using load and displacement sensing indentation experiments.
curves in Fig. 4c are almost fully corrected for machine compliance effects. J. Mater. Res. 7, 15641583 (1992).
37. Field, J. S. & Swain, M. V. A simple predictive model for spherical indentation.
The curves in Fig. 4a and Fig. 4c show engineering stress and engineering strain. J. Mater. Res. 8, 297306 (1993).
The strain values are calculated from the displacement of the testing machine and 38. Johnson, K. L. Contact Mechanics (Cambridge Univ. Press, 1985).
not from strain gauges on the samples. The gradients of the elastic portions of 39. Cheng, Y. Q., Han, Z., Li, Y. & Ma, E. Cold versus hot shear banding in bulk metallic
the curves are therefore affected by the machine compliance and significantly glass. Phys. Rev. B 80, 134115 (2009).

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

40. McKnight, R. E. A. et al. Grain size dependence of elastic anomalies accompanying 42. Yu, H. B. et al. Tensile plasticity in metallic glasses with pronounced b relaxations.
the alpha-beta phase transition in polycrystalline quartz. J. Phys. Condens. Matter Phys. Rev. Lett. 108, 015504 (2012).
20, 075229 (2008). 43. Yu, H. B., Samwer, K., Wang, W. H. & Bai, H. Y. Chemical influence on b-relaxations and
41. Concustell, A., Godard-Desmarest, S., Carpenter, M. A., Nishiyama, N. & Greer, A. L. the formation of molecule-like metallic glasses. Nature Commun. 4, 2204 (2013).
Induced elastic anisotropy in a bulk metallic glass. Scr. Mater. 64, 10911094 44. Churyumov, A. & Yu. . et al. Phase transformations in Zr-based bulk metallic glass
(2011). cyclically loaded before plastic yielding. Mater. Sci. Eng. A 550, 358362 (2012).

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Figure 1 | DSC traces for ribbons of La55Ni20Al25 glass

(heated at 20 K min1). a, These example traces for as-cast samples show the
exotherm, just below the glass-transition temperature, from which the heat
of relaxation DHrel is determined. b, The effect of holding at 77 K: the traces
show that the heat of relaxation DHrel, given by the exotherm just below
the glass-transition temperature, is, within the experimental error of 4%, the
same for an as-cast ribbon, and for a sample held for 4 h at 77 K.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Figure 2 | Differential scanning calorimetry (DSC) of

La55Ni10Al35 bulk metallic glass. The heat of relaxation DHrel is compared for
discs (250500 mm thick) pre-cut from the bulk rod and then treated with room
temperature77 K cycles, and for discs post-cut from treated rod samples.
There is no difference between these cases within the error of 630 J mol1.

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Figure 3 | Load-displacement curves for ribbons of

La55Ni20Al25 metallic glass tested up to a maximum load Fmax of 40 mN. The
initial yield load Fy is indicated on the curves for the glass in three states: as-cast,
after a 10-min hold at 77 K, and after a further ten room temperature77 K
cycles each with 1 min hold.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Figure 4 | Instrumented indentation of La55Ni20Al25

metallic glass ribbon. a, Initial yielding, characterized by initial yield pressure
Py and initial yield displacement Dh. Indentations are made for three states of
the sample: as-cast, after a 10-min hold at 77 K, and after a further ten room
temperature77 K cycles each with 1-min hold. b, c, Distributions of the
Youngs modulus, using data from the same indentations of the sample in three
states as in Fig. 3 and in a. Values of the Youngs modulus in the glass E are
determined (b) by the standard Oliver and Pharr36 method on unloading from
Fmax and (c) from Hertzian fitting to the Fh curve up to the first pop-in.

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Figure 5 | Elastic moduli derived from resonant ultrasound

spectroscopy for Cu46Zr46Al7Gd1 BMG treated with room temperature77
K thermal cycles. Shear modulus (left axis); bulk modulus (right axis).
Error bars, root mean square errors in the fitted frequencies.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Figure 6 | X-ray diffraction traces for Zr-based metallic

glasses subjected to 33877 K thermal cycles. a, For melt-spun ribbons of
Zr61.1Cu26.3Fe2.1Al10.5; b, for rods of Zr62.2Cu23.9Fe4.8Al9.1. No clear changes are
induced by the cycling treatments.

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Figure 7 | Compressive stressstrain curves for rod samples

of Zr62Cu24Fe5Al9 BMG. a, For rods of 2 mm diameter; b, for rods of 2.5 mm
diameter. In each case, increasing numbers of 338 K to 77 K thermal cycles
cause the plastic strain to increase.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Figure 8 | Dynamic mechanical analysis of ribbons of

La55Ni20Al25 metallic glass. The heating rate is 3 K min1. The general form of
the curve matches that shown in Fig. 1a, where an example was chosen of a
metallic glass showing b relaxation at a particularly low value of T/Tg. For
La55Ni20Al25 the b relaxation is centred at T/Tg < 0.8.

G2015 Macmillan Publishers Limited. All rights reserved

 LETTER RESEARCH

Extended Data Table 1 | Values of the Biot number Bi for immersion in liquid nitrogen of the various sample geometries in the present work

The film-boiling regime applies until the sample surface temperature decreases to between 150 K and 100 K, and the nucleate-boiling regime applies thereafter.

G2015 Macmillan Publishers Limited. All rights reserved

 RESEARCH LETTER

Extended Data Table 2 | Summary of instrumented-indentation tests

Cumulative distributions, such as those shown in Fig. 3, are characterized by the median value and by the first and ninth deciles.

G2015 Macmillan Publishers Limited. All rights reserved