\\ Compressed correlation functions and fast aging dynamics in metallic glasses B. Ruta,1,a) G. Baldi,2 G. Monaco,1 and Y. Chushkin1 1)European Synchrotron Radiation Facility, BP220, F-38043 Grenoble, France. 2)IMEM-CNR Institute, Parco Area delle Scienze, I-43124 Parma, Italy (Dated: 29 January 2013) Wepresentx-rayphotoncorrelationspectroscopymeasurementsoftheatomicdynamicsinaZr Ni metallic 67 33 glass, well below its glass transition temperature. We find that the decay of the density fluctuations can be welldescribedbycompressed,thusfasterthanexponential,correlationfunctionswhichcanbemodeledbythe 3 well-known Kohlrausch-Williams-Watts function with a shape exponent β larger than one. This parameter 1 is furthermore found to be independent of both waiting time and wave-vector, leading to the possibility to 0 rescaleallthe correlationfunctions to a single master curve. The dynamics in the glassystate is additionally 2 characterized by different aging regimes which persist in the deep glassy state. These features seem to be n universalinmetallicglassesandsuggestanondiffusivenatureofthedynamics. Thisuniversalityissupported a J by the possibility of describing the fast increase of the structural relaxation time with waiting time using a unique model function, independently of the microscopic details of the system. 8 2 PACS numbers: 64.70.pe,65.60.+a,64.70.pv ] t f o I. INTRODUCTION have been reported also in studies of aging effect on s the intensity of the high frequency tail of the dielectric . at Understanding physical aging in glasses is essential spectra11–16. All these worksprovide information on the m at a technological level, for controlling the material equilibration toward the supercooled liquid phase, but properties and their temporal evolution, as well as at theydonotgiveanymicroscopicdetailsontheagingdy- - d a fundamental level, since glasses are often considered namics in the glassy state. This information can instead n as archetypes for systems far from thermodynamical be obtained by the investigation,on a microscopic scale, o equilibrium1,2. However, and despite the strong efforts of the temporal evolution of the structural relaxation c done in the field, a complete picture of the dynamics in time, τ, which represents the time necessary for the [ theglassystateisstillmissing. Themaindifficultyarises system to rearrange its structure toward a more stable 1 from the complexity of the glassy properties. When a configuration. Its value can be measured, for instance, v supercooled liquid is cooled below its glass transition by monitoring the long time decay of the intermediate 0 temperature, T , the system falls out of equilibrium in a scattering function, f(q,t), on a length scale 2π/q, 1 g 6 metastable state which depends on the previous thermal with the wave-vector q corresponding to the mean inter 6 history, and which evolves, with waiting time, through particle distance. The intermediate scattering function . different states toward the corresponding supercooled is related to the density-density correlation function 1 equilibriumliquidphase. Theexistenceofamultitudeof and contains all the information on the relaxation 0 3 different metastable glassy states which can be explored dynamics in the system on a length scale determined by 1 through different annealing procedures or cooling rates, q. While the dynamics in the supercooled liquid phase : makes difficult the development of a universal theory hasbeen widely investigatedin the past,the difficulty of v for glasses or even a plain comparison between different probingthe particle-leveldynamicsinrealisticmolecular i X experimental results. glasses both with experiments and simulations has r Physical aging in polymeric, metallic or more conven- instead strongly delayed progress in the field. In fact a tional inorganic glasses, is commonly studied by looking only very few studies exist of the behavior of τ below at the temporal evolution of a given observable, keeping Tg3,12,17–24. Numerical simulations on a Lennard Jones fixedotherexternalparameters,suchasthetemperature glass show that for small quenching from an equilibrium and the pressure of the system. Macroscopic quantities configuration, τ slows down linearly or sub linearly with like viscosity, refractive index, volume or elastic con- waiting time18 in agreement with previous results on stants,areusually the subjectofthis kindofstudies3–10. polymericglasses3,12,17. Inaddition,the longtime decay In these works, the investigated quantities evolve with of the correlation functions can be rescaled into a single waitingtime,t ,inawaywhichdependsontheprevious master curve, leading to the validity of a time-waiting w thermal history and which can be accurately described time superposition principle in the glassy state18,25. by stretched exponential functions6. Similar results Recently, the aging of a Mg65Cu25Y10 metallic glass has been studied in both the supercooled liquid phase and the glassy state by looking directly at the evolution of the dynamics at the interatomic length scale23. The a)Electronicmail: [email protected] glass transition has been found to be accompanied by 2 i) a crossover of the intermediate scattering function pre-alloyed by arc-melting under Ti-gettered Ar atmo- from the well characterized stretched exponential shape sphere. The melt was then fast-quenched with a cooling - thus described by a stretching exponent β < 1 - in rate of 106 K/s by injecting it on a Cu wheel spinning the supercooled liquid phase, to an unusual compressed with a 40m/s perimeter velocity. The resulting metallic exponentialform- defined by a parameterβ >1 - in the ribbonshaveathicknessof22±4µm,closetotheoptimal glassy state and ii) to the presence of two very distinct value requiredtomaximize the scatteredintensity inthe agingregimesbelowT . Inparticular,afast,exponential experiments. The ribbons were then held in a resistively g growth of τ has been reported for short waiting times, heatedfurnacecoveringatemperaturerangebetween270 followed by an extremely slow aging regime at larger t . K and 1000 K with a thermal stability of less than 1 K. w While the latter slow regime has been already reported The temperature of the sample was held at T = 373 K, both in numerical simulations18 and in experiments on reached by heating from room temperature with a fixed polymeric glasses3,12,17, the compressed shape of the rate of 3 K/min. correlation functions and the fast aging regime had not X-rayphotoncorrelationspectroscopy(XPCS)measure- been reported before in structural glasses. Two similar ments were carried out at the beamline ID10 at the agingregimeshavebeenrecentlyobservedinanumerical European Synchrotron Radiation Facility in Grenoble, study of a concentratedsystemof nearly hardspheres20. France. An incident x-ray beam of 8.0 keV (λ = 1.55 In this case, however,the correlationfunctions display a ˚A) was selected by using a single bounce Si(1,1,1) crys- more stretched exponential behavior, and no signature tal monochromator with a bandwidth of ∆λ/λ ≈ 10−4. of a compressed shape was reported20. Hard X-rays originating from higher order monochro- Compressed correlation functions and different aging mator reflections were suppressed by a Si mirror placed regimeshaveinsteadbeenreportedforoutofequilibrium downstream of the monochromator. A partially coher- soft materials, where the dynamics is controlled more ent beam with ∼ 1010 photons/s was then obtained by by ballistic-like rather than by diffusive motions26,27. rollerbladeslitsopenedto10×10µm,placed∼0.18mup- The previous work on the metallic glass thus suggests a stream of the sample. In order to enhance the scattered common nature of the dynamics in metallic glasses and intensity, speckle patterns were collected in transmission soft materials. A very similar faster than exponential geometrybytwoIkonMcharge-coupleddevicesfromAn- decay and strong aging behavior for the intermediate dor Technology (1024× 1024 pixels, 13× 13 µm pixel scattering function has been observed also in a Zr-based size each one) installed in the vertical scattering plane metallic glass by Leitner and coworkers24. In this case, ∼70 cm downstreamof the sample. All pixels of the two however, this unusual dynamics has been associated CCD detectors were considered to belong to the same q with the onset of a crystallization process. There are with a resolution of ∆q =0.04 ˚A−1. The detectors were consequently several questions to clarify: i) are the mountedonadiffractometerandcouldrotatearoundthe results found in Mg Cu Y really a universal feature center of the sample position, covering a q range up to 65 25 10 of metallic glasses or do they depend on the microscopic several ˚A−1. XPCS measurements were performed for details of the system? ii) is it possible to find a common different wave-vectorsat and around the first diffraction description of physical aging for different glasses? iii) peak of the static structure factor which corresponds to what is the effect played by the wave-vector, q, on the q =2.55˚A−1, therebyprobingdirectlythe temporalre- p dynamics? laxation of density fluctuations at the inter-particle dis- Here, we answer to these questions by presenting a tance 2π/q ∼ 2 ˚A. Time series of up to ∼5000 images p detailed investigation on physical aging in a Zr Ni were taken with 7 s exposure time per frame and were 67 33 metallic glass at the atomic length scale. We find analyzedfollowingtheproceduredescribedinRef.29. No that in this case the dynamics in the glassy state is beamdamagewasobservedandtheamorphousstructure also characterized by compressed correlation functions of the sample was checked during the whole experiment and different aging regimes, supporting the idea of bymeasuringthecorrespondingstaticstructurefactorof a universal mechanism for physical aging in metallic the system. glasses. This hypothesis is moreover strengthened by the possibility of describing the aging dynamics using the same phenomenological exponential growth and rescaling the results obtained for different systems into a single master curve. III. RESULTS AND DISCUSSION As previously discussed, physical aging implies a re- laxation dynamics which is not the same at all times, II. EXPERIMENTAL DETAILS but strongly depends on the sample age or waiting time. Thetemporaldependenceofthedynamicscanbeentirely The Zr Ni samples (T = 647 K28) were produced capturedbyXPCSthroughthedeterminationofthetwo- 67 33 g asthin ribbons by melt-spinning atthe PolytechnicUni- time correlation function G(t ,t ) (TTCF) which repre- 1 2 versity of Catalonia. The Zr and Ni pure metals were sents the instantaneous correlation of the intensity I at 3 indicates the sample age. The width of this diagonal contour is instead proportional to the relaxation time τ. The steady slowing down of the dynamics, and thus the increase in the relaxation time due to physical aging, is illustrated by the broadening of the diagonalcontour on increasing the sample age. This effect is clearly visible in the top panel of Figure 1 where we report data taken between ∼ 1000 and ∼ 5000 s after temperature equili- bration. After almost13 hours ofannealing (lowerpanel of Figure 1), the dynamics was so slow that the inten- sity along the diagonal covered almost the whole image, despite the time interval was almost the double with re- spect to the one reported in the top panel. After this very long annealing the width of the diagonal contour seems to be constant with time, as it would be in the case of stationary equilibrium dynamics. As discussed later,heretheeffectofphysicalagingisprobablysoslow that the dynamics appears as stationary at least on our experimental time scale. Althoughthedynamicsisnotstationary,theagingissuf- ficientlyslowthatitispossibletofindtimeintervalsover which the G(t ,t ) can be averaged and thus to get the 1 2 standard intensity autocorrelationfunction hI(q,t )I(q,t +t)i D 1 1 pE g (q,t)= (2) 2 hI(q,t )i hI(q,t )i D 1 p 1 pE where h...i is the temporal average over all the possible starting times t . The intensity autocorrelationfunction 1 isrelatedtotheintermediatescatteringfunctionthrough the Siegert relation g (q,t) = 1 + B(q)|f(q,t)|2, with 2 B(q) a setup-dependent parameter31, and its determina- tion allows capturing snapshots of the dynamics at dif- ferent waiting times. The Siegert relation requires the temporal average to be equal to the ensemble average. FIG.1. Two-timecorrelationfunctionsmeasuredwithXPCS This condition is fulfilled for systems at thermodynamic inZr Ni atT=373K.Dataacquisitionstartedatt =1400s equilibrium. The extensionto non-ergodicsystems, such 67 33 0 (toppanel)andt =49000s(bottompanel)fromtemperature as glasses,is possible thanks to the use of an area detec- 0 equilibration. TheageofthesampleatanypointintheFigure tor, like the CDD used during our experiments. In this canbetakenastage=t0+(t1+t2)/2s. Thehugeslowdown case in fact, the intensity distribution is sampled cor- ofthedynamicsatthedifferentsampleagesiswellillustrated rectly by the average over all the pixels, independently by the clear broadening of the intensity around the diagonal of the nature of the dynamics32,33. from the left bottom corner to the top right corner in the The time interval for temporal averaging was chosen bottom panel with respect to the case corresponding to a smallenoughthattheshapeofg (q,t)wasnotnoticeably much shorter tw (top panel). 2 affected by the evolution of the dynamics, while keeping a good signal to noise ratio. For each set of images, the two times t1 and t2, being: sampleagewasdefinedastw =t0+(tf−ti)/2,wheret0is the delayofthe startingpointfor the measurementfrom hI(t1)I(t2)ip temperatureequilibration,andtf andti arethefinaland G(t ,t )= . (1) 1 2 hI(t )i hI(t )i starting times of the interval used for the average. 1 p 2 p Theαrelaxationprocessinglass-formerscanbe wellde- Here, the average is performed on all the pixels of the scribedbyusingtheempiricalKohlrauschWilliamsWatt CCD, that correspond to the same wave-vector q29–31. (KWW) model function1 Figure 1 shows the TTCF measured in Zr Ni in the 67 33 f(q,t)=f (T)·exp[−(t/τ)β] (3) glassy state at two different sample age times from tem- q perature equilibration. The colors represent the values Inthisexpression,τ isthecharacteristicrelaxationtime, of G. The diagonal from bottom left to top right cor- β is the stretching parameter which quantifies the de- responds to the time line of the experiment and it thus viation from a simple exponential behavior, and f (T) q 4 the liquid above the glass transition temperature, these theories predict a stretched, thus with β ≤ 1, decay of thedensityfluctuationsevenintheglassystate. Theidea is that the dynamic heterogeneities present in the liquid 1.0 phase partially freeze below T , leading to a narrower g distribution of relaxation times, and thus to an increase c 0.8 of the shape exponent, up to the limiting case of β = 1. / 1) This explanationwell describes the correlationfunctions )- calculated in numerical simulation18 and the increase of q,tp 0.6 β measured in some polymeric glasses12,17, but cannot ( 2 take into account the behavior reported in Figure 2. g ( 0.4 tw= 2300 From the comparison with other systems, it seems then tw= 3700 clear that metallic glassesbehave in a different universal 0.2 tw= 5000 way,independently oftheiratomicstructure. Inthecase tw= 54800 ofMg65Cu25Y10,theunusualfasterthanexponentialbe- 0.0 haviorhasbeenattributedtothestrainfieldgeneratedby 10 100 1000 10000 arandomdistributionofslowly-evolvingsourcesofinter- t (s) nal stresses stored in the system during the quench, and FIG. 2. Selection of the intensity correlation functions mea- which can then be released upon annealing the system sured with XPCS at T = 373 K for different waiting times close to T 23,26. This interpretation suggests the exis- g from temperature equilibration. On increasing the waiting tence of a different kind of dynamics, rather than pure time, the decay clearly shifts toward longer time scales due diffusion,inmetallicglasses,whichdisplaymanysimilar- to physical aging. The data are reported together with the ities with that of out-of-equilibrium soft materials, such resulting fit obtained using equation (4). as concentrated colloidal suspensions and nanoparticle is the nonergodicity factor. It represents the height of probes in a glass former matrix23,39–42. the intermediate-time plateau before the final decay as- InthecaseofZr67Ni33 theshapeparameterisevenlarger sociated to the α-relaxation and it slightly depends on than the one reported for Mg65Cu25Y10 (β ∼ 1.5 for temperature in the glassy state. a similar thermal path). This may suggest the pres- Following the Siegert relation, the intensity correlation ence of larger atomic mobility or more internal stresses functionsmeasuredwithXPCScanthenbedescribedby in Zr67Ni33, a conclusionwhich however requires further investigations. While the value β ∼ 1.5 has been often g2(q,t)=1+c(q,T)·exp[−2(t/τ)β] (4) observedin numerous experimentalstudies of soft mate- rials and corresponds to the early-time regime proposed where c = B(q)f2, being B(q) the parameter entering q in the model of Bouchaud and Pitard26, values of β as the Siegert relation. large as those here reported for Zr Ni have been ob- A selection of normalized intensity autocorrelationfunc- 67 33 served only in another Zr-based metallic glass24 and in tionsobtainedatT =373Kforafixedq =q anddiffer- p fewothersoftsystems,andtheirinterpretationisstillfar ent waiting times is reported in Figure 2. The data are from being reached41,43. shown together with the best-fit line shapes obtained by The waiting time dependence of the structural relax- using equation (4). The intensity correlation functions ation times obtained from the analysis of the g (q ,t) is decorrelate completely on the experimental time scale, 2 p shown in Figure 3 a) (squares). Data corresponding to implying that the system is able to rearrange its struc- differentq valuesaroundthemainpeakofthediffraction ture on the probed spatial length scale of few ˚A. The patternarealsoreported. Asexplainedintheexperimen- effect of aging is illustrated by the shift of the decay to- taldetails,thelowscatteredsignalallowedustomeasure ward longer time scales upon increasing the sample age, justone q vectorfor eachdetectorsetting. Consequently in agreement with the analysis of the two times correla- thedifferentqsaretakenatdifferentt fromtemperature w tion functions reported in Figure 1. As shown in Figure equilibration. At all the investigated qs, τ grows rapidly 2, aging is so important that the decay time increases of withthesampleage,inagreementwiththeresultsfound almost a factor 10 in less than 13 hours. This fast aging in the Mg Cu Y glass23. Interestingly, the dynam- is likelyto be the reasonofthe wellknownrapidembrit- 65 25 10 ics is characterized by structural rearrangements which tlement of metallic glasses34,35. require ∼ 103 −104 s thus, relatively short times with Interestingly, the correlation functions display a com- respect to the common idea of an almost frozen dynam- pressed,thusfasterthanexponentialbehavior,whichcan ics in the deep glassy state. These values are however in be well described by equation (4) with a shape param- perfect agreement with those reported for both the Mg- eter β ∼ 1.8, independent of the sample age. Such a based and Zr-based metallic glasses23,24 and suggest the high value is in agreement with the results reported for presence of a peculiar dynamical behavior at the atomic other metallic glasses23,24, while it cannot be explained lengthscale. Hence,itwouldbeinterestingtoinvestigate withinthecurrenttheoriesforthedynamicsintheglassy the nature of the atomic dynamics in the glassy state state36–38. Basedonthe observationsof the dynamics in 5 single master curve and define a time-waiting time-wave vector superposition principle in the glass. This scaling is reported in 3 d) where a selection of intensity auto 10000 correlationfunctions corresponding to the data reported 8000 a) s) b) in panel a) are plotted as a function of t/τ. In the case (6000 of Mg65Cu25Y10, β was found to be temperature and tw 4000 independentin the fastagingregime23. Ourresultsindi- ) s catethatthisparameterisalsonotaffectedbythechoice ( 30000 40000 50000 ofthewave-vector,atleastforvaluesofq closetothatof tw (s) the maximum ofthe static structure factor (see Figure 3 1000 e)). As shown in Figure 3 a) and b), for q = q and large p t , the fast aging regime seems to change to a more w c) stationary or very slow one, as also suggested by the 1.8 analysis of the two times correlation function reported in the bottom panel of Figure 1. The presence of a com- -1 plexhierarchyofagingregimeshasbeenobservedalsoin 1.4 Q=2.38 ¯ Q=2.47 ¯-1 Mg65Cu25Y10butfortemperaturesveryclosetotheglass -1 transitionone(T/Tg =0.995). Theresultsherereported 1.0 Q=2.75 ¯-1 forT/Tg =0.577showthatthis secondagingregimecan Q=2.55 ¯ be foundalsointhe deepglassystate,after asufficiently -1 0.6 Q=2.65 ¯ long annealing time. In the case of Mg65Cu25Y10, the 1000 10000 second aging regime was accompanied by a decrease in t (s) the shape parameter. As shown in Figure 3 c), in the w presentcase the lack of a full complete decay ofg (t) for 2 1.0 d) thehighestt leadstoalargererrorinthedetermination w c tw=1 h ofβ, whichcanbe determined only fortw ≥50000s and 1)/ 0.8 tw=3 h remains however similar to the value obtained at lower g(t)-20.6 Q)/I(Q)0 e) ttww==58 hh Fttworo.manthaelmeonsetrgsettaitciopnoairnyt odfyvniaemwi,csthesuagbgreustpst tchroesseoxviesr- ( 0.4 I( tence of a deep energy minimum in the potential en- t =10 h w ergy landscape, where the system remains trapped for 0.2 1.5 2.0 2-.15 3.0 tw=15 h a very long time before eventually jumping into a dif- Q (¯ ) ferent configuration44,45. A similar plateau has been re- 0.0 0.01 0.1 1 cently observed in the enthalpy recovery of glassy poly- t/ (Q,t ) mers and has been interpreted as a thermodynamically w stable state10. This explanation cannot be extended to FIG. 3. a) Waiting time dependence of the structural relax- the case of metallic glasses. In fact, the comparison be- ation time measured at T=373 K for different wave-vectors around the first sharp diffraction peak (qp =2.55 ˚A−1). The tdwateaenfoXrPMCSg aCnduloYw temsupgegreasttusreinesqteuaidlibariugmradvuisaclosaitpy- inset b) reports a zoom of the data taken at large waiting 65 25 10 proach toward equilibrium23, in agreement with the sub times. c) Corresponding shape parameter. d) Selection of linear aging regime found in other polymeric glassesand the correlation curves corresponding to the data reported in a) as a function of the time rescaled by the structural relax- colloids3,12,46. ation time obtained from the analysis of the curves with the Following the procedure reported in Ref.23, we can de- KWW model function of Eq. (4). The inset e) indicates the scribe the short t behavior of τ (t ) using the phe- w qp w position of thedifferent investigated qs. nomenologicalexpression through the q dependence of the structural relaxation τ(T,t )=τ (T)exp(t /τ∗) (5) w 0 w time. Unfortunately, this information would require at each investigated q a good knowledge of the aging in a where τ (T) is the value of τ for t = 0 at the temper- 0 w ∗ wide t range, which can be gained only by measuring ature equilibration time, and τ is a fitting parameter w the dynamics simultaneously for different wave-vectors. which describes the rate growth and is independent of A higher incident flux, as the one available at free elec- temperature23. A fast exponential growth with t is ev- w tronlasersources,willmakethiskindofstudiespossible. ident also at the other investigated qs. However, in that Interestingly, the shape parameter does not display any cases the lack of information on the behavior at shorter dependence on q or t , and remains constant with β ∼ t does not allow for a more quantitative description of w w 1.8±0.08 for all the investigated qs (Figure 3 c)). This the aging dependence for τ. ∗ means that it is possible to rescale all the data sets to a For q = q we find τ = 407±6 s and τ = 6490±170 p 0 6 IV. CONCLUSIONS In conclusion, we have studied the physical aging in a Zr Ni metallic glass, by following the temporal 67 33 evolution of the structural relaxation process at the interatomic length scale. We find that the decay 10 ) of the density fluctuations can be well described by T ( compressed, thus faster than exponential correlation 0 / functions, suggesting a peculiar origin for the atomic dynamics27,31. This behavior seems to be universal Mg65Cu25Y10 in metallic glasses23,24, and cannot be explained by Mg65Cu25Y10 the well-known theories for glasses dynamics, which Zr67Ni33 predict a stretched exponential shape of the correlation functions, even in the glassy state36–38. Model systems 1 such as hard spheres and Lennard-Jones are unable to 1 2 3 4 5 6 7 8 9 10 * reproduce these intriguing dynamic effects18,20. In these t / w model systems, indeed, the dynamics appears to be alwayscharacterizedbyabroadspectrumofrelaxations. FIG. 4. Structural relaxation times measured in two as Conversely, very similar results have been previously quenched glasses of Mg65Cu25Y10 at T/Tg = 0.995 with dif- ferentthermalhistoriesandinZr67Ni33 atT/Tg =0.577. All reportedforoutofequilibriumsoftmaterials,suggesting the data are normalized for the τ parameter obtained from a universalmicroscopicdynamics forthese verydifferent 0 the analysis of the fast aging with the empirical model func- systems23. In these systems, the compressed behavior tion (5) and are reported as a function of the time rescaled arises from ballistic-like motion due to the presence ∗ by the τ parameter. The data for Mg65Cu25Y10 are taken of internal stresses in the out of equilibrium state27. from Ref.23. It would be interesting to check whether this is also the case for metallic glasses. However, a q dependence study wouldrequire a muchstrongersignalthan the one achievable evenat a third generationsynchrotronsource as the one used for the present study. Notwithstanding, indication of the presence of distinct processes of atomic s. The latter value is quite close to the one reported motion in metallic glasses have been reported for differ- in ref.23 for Mg Cu Y at T/T = 0.995 and it is ent Zr-based metallic glass-formers studied with nuclear 65 25 10 g of the same order of magnitude of that previously re- magnetic resonance48. In these systems the diffusive, portedinthecaseofcolloidalgelsandLaponite39,47,thus collectivemotionisfoundtobeaccompaniedbyhopping strengthening the idea of a universal out-of-equilibrium which becomes dominant in the glassy state, and could behavior. The comparison with the Mg-based metallic then be related to the observed compressed relaxation glass suggests that the physical aging is very similar for process in our glass. both systems and that it should be possible to rescale Albeitwecouldnotgetinformationontheq dependence the rapid growth of τ(t ) to a unique curve, indepen- of the relaxation time due to the impossibility of mea- w dently on the microscopic details of the system. This suring simultaneously different qs, we have investigated scaling is shown in Figure 4, where the structural relax- the nature ofthe dynamics at differentwave-vectorsand ation times of Zr Ni at T/T = 0.577 are compared sample ages. We find that the shape of the correlation 67 33 g to those measuredin Mg Cu Y atT/T =0.995and functions does not show any dependence neither on 65 25 10 g fortwodifferentthermalpaths. Whilesquaresandcircles the sample age nor on the wave-vector, at least for qs correspondto as-quenched samples heated up to a given aroundthemaximumofthestaticstructurefactorofthe temperature, stars refer to a state reached after having system, which is at q = 2.55 ˚A−1. These findings allow p partiallyannealedtheglassathighertemperatureswith- then to rescale all the correlation curves on the top of outequilibration23. Allthe dataarerescaledforthe cor- each other and define a time-waiting time-wave-vector responding τ parameter obtained from the analysis of superposition principle in the fast aging regime. 0 the fast aging using equation (5) and are reported as a The value of β seems to depend on the details of the ∗ function of the time rescaled by the τ parameter. For system, being larger in Zr-based systems than in the short t , all the data fall into a unique master curve, previously reported Mg Cu Y 23. As both these w 65 25 10 suggesting a common origin of the fast aging regime in glasses were obtained with the same thermal protocol, metallic glasses. Differently, the crossover to the subse- we speculate that the observed difference could be re- quent almost stationary dynamics strongly depends on lated to a larger atomic mobility in the first system and the temperature and thermal path, and is reached only to the consequent presence of larger internal stresses. for very long annealing times (circles) or after particular Despite this small difference, the comparison between thermal treatments (stars). the two metallic glasses suggests the existence of a 7 universal dynamical behavior, also indicated by the 23B. Ruta, Y. Chushkin, G. Monaco, L. Cipelletti, E. Pineda, scalingofthefastagingregimefortherelaxationtime to P.Bruna,V.M.Giordano,andM.Gonzalez-Silveira,Phys.Rev. a single curve, independently of the microscopic details Lett.109,165701 (2012). 24M.Leitner,B.Sepiol,L.M.Stadler,andB.Pfau,Phys.Rev.B of the systems. Additional theoretical and experimental 86,064202 (2012). works is clearly required to elucidate this remarkable 25T. Hecksher, N. B. Olsen, K. Niss, and J. C. Dyre, J. Chem. dynamical behavior of metallic glasses. Phys.133,174514 (2010). 26J.P.Bouchaud andE.Pitard,Eur.Phys.J.E6,231(2001). 27L. Cipelletti, L. Ramos, S. Manley, E. Pitard, D. A. Weitz, E. E. Pashkovskid, and M. Johansson, Faraday Discuss. 123, We acknowledge F. Zontone, H. Vitoux, K. L’Hoste, 237(2003). L. Claustre for technical support during the XPCS ex- 28K.Georgarakis,A.R.Yavari,M.Aljerf,D.V.Louzguine-Luzgin, periments and E. Pineda an P. Bruna for providing the M.Stoica,G.Vaughan,andA.Inoue,J.Appl.Phys.108,023514 samples. (2010). 29Y. Chushkin, C. Caronna, and A. Madsen, J. Appl. Cryst. 45, 1C.A.Angell,K.Ngai,G.McKenna,P.McMillan,andS.Martin, 807(2012). J.Appl.Phys.88,3113(2000). 30M. Sutton, K. Laaziri, F. Livet, and F. Bley, Opt. Express 11, 2L.BerthierandG.Biroli,Rev.Mod.Phys.83,587(2011). 2268(2003). 3L.C.E.Struik,Physicalaginginamorphouspolymersandother 31A. Madsen, R. L. Leheny, S.-M.Guo, H., and O. Czakkel, New materials (Elsevier,Amsterdam,1978). J.Phys.12,055001(2010). 4R. Busch, R. Bakke, and W. L. Johnson, Acta Mater. 46, 4725 32L.CipellettiandD.A.Weitz,Rev.Scient.Instr.70,3214(1999). (1998). 33E.Bartsch,V.Frenz,J.Baschnagel,W.Sch¨artl,andH.Sillescu, 5A.NapolitanoandP.B.Macedo,J.Res.Nat.Bur.Stand.72A, J.Chem.Phys.106,3743(1997). 425(1968). 34T.W.WuandF.Spaepen, Philos.Mag.B61,739(1990). 6A.J.Kovacs,J.Polym.Sci.30,131(1958). 35P.MuraliandU.Ramamurty,ActaMater.53,1467(2005). 7P.Wen,G.P.Johari,R.J.Wang,andW.H.Wang,Phys.Rev. 36O.S.Narayanaswamy, J.Am.Ceram.Soc.54,491(1971). B73,224203(2006). 37C. T. Moynihan and J. Schroeder, J. Non-Cryst. Sol. 160, 52 8R. S. Miller and R. A. MacPhail, J. Chem. Phys. 106, 3393 (1993). (1997). 38V. Lubchenko and P. G. Wolynes, J. Chem. Phys. 121, 2852 9T. S. Grigera and N. E. Israeloff, Phys. Rev. Lett. 83, 5038 (2004). (1999). 39L.Cipelletti,S.Manley,R.C.Ball,andD.A.Weitz,Phys.Rev. 10V. M. Boucher, D. Cangialosi, A. Alegria, and J. Colmenero, Lett.84,2275(2000). Macromolecules44,8333(2011). 40R.Bandyopadhyay,D.Liang,H.Yardimci,D.A.Sessoms,M.A. 11E.SchlosserandA.Sch¨onhals,Polymer32,2135(1991). Borthwick, S. G. J. Mochrie, J. L. Harden, and R. L. Leheny, 12A.Alegria, L.Goitiandia, I.Telleria, andJ. Colmenero, Macro- Phys.Rev.Lett.93,228302(2004). molecules30,3381(1997). 41C. Caronna, Y. Chushkin, A. Madsen, and A. Cupane, Phys. 13R.L.LehenyandS.R.Nagel,Phys.Rev.B57,5154(1998). Rev.Lett. 100,055702(2008). 14P. Lunkenheimer, R. Wehn, U. Schneider, and A. Loidl, Phys. 42H. Guo, G. Bourret, M. K. Corbierre, S. Rucareanu, R. B. Rev.Lett.95,055702(2005). Lennox,K.Laaziri,L.Piche,M.Sutton,J.L.Harden,andR.L. 15P.Lunkenheimer,R.Wehn,andA.Loidl,J.Non-Cryst.Sol.352, Leheny, Phys.Rev.Lett.102,075702(2009). 4941(2006). 43O.CzakkelandA.Madsen,Europhys.Lett.95,28001(2011). 16R.Richert,Phys.Rev.Lett.104,085702 (2010). 44M.Goldstein,J.Chem.Phys.51,3728(1969). 17A.Alegria,E.Guerrica-Echevarria,L.Goitiandia,I.Telleria,and 45F.H.Stillinger,Science267,1935(1995). J.Colmenero,Macromolecules28,1516(1995). 46L. Cipelletti and L. Ramos, J. Phys.: Cond. Mat. 17, R253 18W.KobandJ.L.Barrat,Phys.Rev.Lett. 78,4581(1997). (2005). 19W.KobandJ.Barrat,Eur.Phys.J.B13,319(2000). 47F.Schosseler,S.Kaloun,M.Skouri,andJ.P.Munch,Phys.Rev. 20D. El Masri, L. Berthier, and L. Cipelletti, Phys. Rev. E 82, E73,021401(2006). 031503(2010). 48X.-P. Tang, U. Geyer, R. Busch, W. L. Johnson, and Y. Wu, 21D. Prevosto S. Capaccioli, M. Lucchesi, P. A. Rolla, and K. Nature402,160(1999). L.Ngai,J.Chem.Phys.120,4808(2004). 22R. Casalini and C. M. Roland, Phys. Rev. Lett. 102, 035701 (2009).