betaFeedback

WorksPhysics and Astronomy

At low temperatures, glass-forming liquids relax in a simple way

Abstract

Glass-forming liquids have only a modest tendency to crystallize and hence their dynamics can be studied even below the melting temperature. The relaxation dynamics of most of these liquids shows at a temperature TcT_c, somewhat above the glass-transition temperature TgT_g, a crossover, which indicates the conjunction of two different dynamical regimes. For temperatures slightly above TcT_c, experiments and computer simulations have extensively probed this dynamics on the particle level and identified several universal scaling laws that are often compatible with theoretical predictions. Using large scale computer simulations we extend these studies to temperatures below TcT_c and find that the relaxation mechanism is qualitatively different from the one found at higher temperatures. We identify new scaling laws that allow to give a simple description of the relaxation dynamics at very low TTs. Specifically we reveal that the cage-escape process is related to rare but large particle displacements that give rise to a distinctive sub-diffusive power-law in the time correlation functions. This insight helps to advance our understanding on the relaxation dynamics of glass-forming systems at temperatures that are close to the experimental glass transition.

In the last decade impressive advancement in our understanding of glassy systems has been made, progress that was triggered by novel theoretical tools like machine learning, advanced simulation techniques, as well as experimental methods that have allowed to probe the static and dynamic properties of supercooled liquids and glasses in unprecedented details [1, 2, 3, 4]. These studies have permitted to reach, on the particle level, a satisfying understanding of the relaxation dynamics of glassy systems for temperatures that are around or above the critical temperature TcT_{\mathrm{c}} of Mode-Coupling Theory (MCT) [5]. Much less is known, however, about the dynamics below TcT_{\mathrm{c}}, since at these temperatures the relaxation times are usually by many orders of magnitude larger than the vibrational ones, making it very challenging to examine the details of the α\alpha-relaxation. (Although network glass-formers can be simulated also below TcT_{\mathrm{c}} [6], their relaxation dynamics is very different from the one of fragile glass-formers [7, 8, 9].) Also from the theoretical side the sub-TcT_{\mathrm{c}} regime is not studied well, since it is difficult to make analytical predictions that go beyond the ones made by the simplest version of MCT [10, 11, 12]. Although there is consensus that close to and below TcT_{\mathrm{c}} the divergence of the relaxation time as predicted by MCT must be cut off, the details of the process governing the α\alpha-relaxation, often called “hopping process” [13, 14, 15], have not been elucidated so far. The goal of the present work is thus to investigate the relaxation dynamics in this temperature regime on the level of the particles. We find that around and below TcT_{\mathrm{c}} this dynamics displays universal aspects that make it surprisingly simple, a result that will allow to advance our theoretical understanding of deeply supercooled liquids.

System

Our system is a binary Lennard-Jones mixture that has been often used in the past to study the relaxation dynamics of simple glass-formers [16]. We use a slight modification of the interaction parameters which suppresses crystallization [17], but does not affect the (above TcT_{\mathrm{c}}) relaxation dynamics, which the critical temperature TcT_{\mathrm{c}} remains around 0.435 [16]. The number of particles is 32,400 and the simulations were done at constant volume, with T=0.375≈0.86TcT = 0.375 \approx0.86T_{\mathrm{c}} being the lowest investigated temperature. At the lowest TTs, the length of the runs was 2⋅1092 \cdot10^{9} time-steps, and we averaged over 4 independent samples. More details on the simulations are given in the SI. In the following we discuss the dynamics of the larger particles since they are in the majority (80%).

Results

fig:1a presents ⟨r2(t)⟩\langle r^{2}(t) \rangle, the Mean Squared Displacement (MSD) of a particle, as a function of the rescaled time t/τFt/\tau_{\mathrm{F}}, where the Fickian time scale τF\tau_{\mathrm{F}} [20] is defined as τF=σ2/6D\tau_{\mathrm{F}} = \sigma^{2}/6D, with DD the diffusion constant obtained from the MSD at long times. By construction, this rescaling makes that in the diffusive regime all MSD’s fall on a master curve. For TTs slightly above TcT_{\mathrm{c}}, data follow a master curve also in the time range in which the particles start to leave the plateau, i.e., at the end of the caging regime. This behavior is compatible with MCT which predicts a tt-dependence of the form rc2+a(t/τF)b+c(t/τF)r_{\mathrm{c}}^{2} + a(t/\tau_{\mathrm{F}})^{b} + c(t/\tau_{\mathrm{F}}), with cage size rcr_{\mathrm{c}} and exponent bb, both independent from TT. Previous fitting results [21] indicated rc≈0.2r_{\mathrm{c}} \approx0.2 and b≈0.5b \approx0.5, fully compatible with the present results. Interestingly, one finds that for temperatures below TcT_{\mathrm{c}} the MSD’s no longer fall onto a master curve, indicating that the escape of the particles from the cage is related to a microscopic mechanism that differs from the one at higher TTs. The height of the plateau, i.e., the size of the cage, now depends on TT, see Inset of panel (a). This dependence is linear in TT, at variance with the T1/2T^{1/2}dependence expected for a harmonic system. Hence one concludes that, despite the fact that the particles are trapped for a significant amount of time, the cage has a significant anharmonic shape [15], in agreement with previous studies on the configurational entropy [22] and with the result that TcT_c corresponds to a transition in the dynamic exploration of the potential energy landscape [23, 24]. Panel (b) shows that also the Self-Intermediate Scattering Function Fs(q,t)F_s(q,t) (SISF) plotted versus the α\alpha-relaxation time τα\tau_\alpha, displays the same qualitative scenario emerging from panel (a). In agreement with MCT, the curves at intermediate TTs fall onto a master curve with a decay from a plateau that is well described by a power-law fc−B(T)tbf_c - B(T)t^b, with the non-ergodicity parameter fcf_c and the von Schweidler exponent bb, both independent of TT. For low TTs, the curves no longer fall on a master curve with fcf_c increasing with decreasing TT, see Inset of panel (b). Thus we find again qualitative differences in the dynamics at intermediate and deeply supercooled conditions (approximately above and below TcT_c). Panel (c) shows the relation between the two timescales τα\tau_\alpha and τF\tau_F. For high and intermediate temperatures one finds the expected Stokes-Einstein relation, i.e., τF∝τα\tau_F \propto\tau_\alpha. About 10% above TcT_c (T≈0.475T \approx0.475), this dependence is replaced by a fractional Stokes-Einstein relation τF∝ταc\tau_F \propto\tau_\alpha^c [25, 26], with an exponent cc around 0.78, indication of strong dynamical heterogeneities (DH) in the sample [27, 28]. One indicator for the strength of these DH is the non-Gaussian parameter α2(t)=5⟨r4(t)⟩/3⟨r2(t)⟩2−1\alpha_2(t) = 5\langle r^4(t)\rangle/3\langle r^2(t)\rangle^2 - 1, the tt-dependence of which is presented in panel (d). One recognizes that at high and intermediate temperatures the rise of α2(t)\alpha_2(t) is essentially

independent of TT and is well described by a power-law, in agreement with earlier studies [19]. For temperatures below TcT_c, this is no longer the case, showing that there is a change in the nature of the TT-dependence of the DH. The TT-dependence of τ∗\tau^*, the time at which α2\alpha_2 reaches its maximum, is presented in Figure S2.

Time of the maximum of $\alpha_2(t)$, $\tau^*$, as a function of $1/T$

Figure S2. Time of the maximum of α2(t)\alpha_2(t), τ∗\tau^*, as a function of 1/T1/T. The vertical line indicates the position of the critical temperature Tc=0.435T_c = 0.435. Inset: Fickian time-scale τF\tau_F vs τ∗\tau^*. Dashed line is a power law τF∝(τ∗)1.1\tau_F \propto(\tau^*)^{1.1}.

Overall, Figure 1 demonstrates that, upon changing TT, there are two different dynamic regimes. We do not claim, however, that TcT_c is the exact boundary between these two different dynamical behaviors. Rather, this temperature should be regarded as indicative of a point at which where the nature of the dynamics changes, and we emphasize that we reach this conclusion from the analysis of the raw data only, without making any fit inspired by MCT.

Four panels showing mean-squared displacement, self-intermediate scattering function, relaxation timescales, and non-Gaussian parameter

Figure 1. (a) MSD as a function of t/τFt/\tau_F, for the temperatures T=1.0T = 1.0, 0.60.6, 0.550.55, 0.50.5, 0.4750.475, 0.40.4, 0.3850.385, 0.380.38, and 0.3750.375, from right to left. Inset: MSD plateau height rcr_c as a function of temperature for T<TcT < T_c. The dashed line is a linear fit. (b) The self intermediate scattering function (SISF) computed at wave-vector q∗=7.25q^* = 7.25, the location of the main peak in the static structure factor, as a function of t/ταt/\tau_\alpha, for the same temperatures as in (a). Inset: Plateau height of the SISF as a function of temperature for T<TcT < T_c. The dashed line is Gaussian fit fc∝e−aT2f_c \propto e^{-aT^2}. (c) τF\tau_F versus τα\tau_\alpha. The orange line is a linear fit to the data for T≥ToT \ge T_o, ToT_o being the estimated onset temperature of glassy dynamics for this system, To≈0.8T_o \approx0.8 [18, 19]; the black line is a power-law fit τF∝τα0.78\tau_F \propto\tau_\alpha^{0.78} to the data for T<TcT < T_c (filled symbols). (d) Non-Gaussian parameter as a function of time for the same temperatures as in panel (a).

In the following we focus on the dynamics at low TTs and demonstrate that it shows system-universal features that permit to describe it in a simple manner. Figure 2(a) presents ⟨r2(t;T)⟩−rc2(T)\langle r^2(t;T)\rangle-r_c^2(T) as a function of t/τFt/\tau_F. One recognizes that the time-dependence of this difference is independent of TT over several decades in time, i.e., the escape from the cage is independent of TT. For times shorter than τ∗\tau^* (vertical bar) the increase of ⟨r2⟩−rc2\langle r^2\rangle-r_c^2 is a power-law with exponent b′=0.67±0.02≈2/3b'=0.67\pm0.02\approx2/3; for times larger than τ∗\tau^*, the curve starts to bend upwards to enter the Fickian regime around τF\tau_F [20, 29]. To the best of our knowledge, the presence of such a unique sub-diffusive behavior after the caging regime, at temperatures significantly below TcT_c, has never been observed in earlier studies of supercooled liquids. The emergence of a low-TT dynamics that, up to a scaling factor, is independent of TT can be observed also in Fs(q,t)F_s(q,t). In Figure 2(b) we present the time dependence of Fs(q,t;T)/fc(T)F_s(q,t;T)/f_c(T) and one recognizes that all curves below TcT_c fall indeed on a master curve when plotted as a function of t/ταt/\tau_\alpha. Note that this master curve is observed in the whole α\alpha-regime as well as in the late β\beta-regime (see Inset), i.e., throughout a time-range that is much longer than the one found for temperatures above TcT_c [30]. The master curve is very well described by a stretched exponential, exp⁡{−(t/τα)β}\exp\{-(t/\tau_\alpha)^\beta\}, with β=0.67±0.02≈2/3\beta=0.67\pm0.02\approx2/3, i.e., the exponent is the same as the one we found for the MSD in panel (a). To the best of our knowledge, the identification of such well defined sub-diffusive post-caging scalings, with b′=βb'=\beta, has never been reported for glass-forming liquids. This demonstrates that at low temperatures the β\beta- and α\alpha-processes for the MSD and Fs(q,t)F_s(q,t) are strongly related to each other, in contrast to the case at higher temperatures for which the von Schweidler exponent (independent of TT) and the stretching exponent (often varying with TT [15, 26, 31]) are different [16]. For the sake of completeness we show in Figure S1 the temperature-dependence of β\beta as obtained from our data, throughout a range of temperatures, 5.0≥T≥0.3755.0\geq T\geq0.375. It is indeed evident that at high and intermediate TT a significant TT-dependence of β\beta exists, but it saturates at very low TT, indicating that the asymptotic regime has been reached.

Mean-squared displacement and self-intermediate scattering function

Figure 2. (a) MSD - rc2(T)r_c^2(T) as a function of t/τFt/\tau_F, for different temperatures in the sub-TcT_c regime. Black and red lines are tt and t2/3t^{2/3} power-laws, respectively. Contribution MSDtail\mathrm{MSD}_{\mathrm{tail}} (see text) for T=0.375T=0.375 is plotted as gray points. (b) SISFs rescaled by the non-ergodicity factor fc(q∗)f_c(q^*) as a function of t/ταt/\tau_\alpha, for the same temperatures as in (a). Red dashed line is a fit with a stretched exponential with exponent β=2/3\beta=2/3. Inset: Red dashed line indicates the short time expansion ∝tβ\propto t^\beta of the stretched exponential fit. The red shadowed regions in the panels indicate the times τ∗\tau^* (see Fig. S2) at the different displayed temperatures.

Two-panel plot of stretching exponent and deviation from stretched exponential fit

Figure S1. (a) Stretching exponent β\beta, as obtained from the intermediate scattering function, as a function of inverse temperature, at various investigated temperatures. The red vertical line indicates the critical temperature Tc=0.435T_{\mathrm{c}} = 0.435; black horizontal line indicate the value β=2/3\beta= 2/3. (b) ϵ(t;T)\epsilon(t;T) as a function of rescaled time, highlighting the deviation of data at various TTs (indicated in the legend) from the stretched exponential fit with β=2/3\beta= 2/3.

Deeper insight into the nature of the observed low-TT dynamics after the caging regime is obtained by probing the self

Van Hove function Gs(r,t)G_s(r,t), i.e., the probability that a particle has moved in time tt by a distance rr. Figure 3a shows that 4πr2Gs(r;t)4\pi r^2G_s(r;t) peaks at a distance r1r_1 at short times, i.e., it shows a Gaussian core due to the particles that rattle inside the cage they were in at t=0t=0; the peak position remains almost fixed at r1≈rcr_1\approx r_c for a long time (up to around τF\tau_F, see below). For times around τ∗\tau^* the distribution shows a shoulder which evolves into a second peak at a distance r2≈0.9r_2\approx0.9, i.e., close to the position of the main peak in the radial distribution function. This second peak is due to the particles that have left their original cage and have effectively hopped to the location that corresponds to the first neighbor distance [15]. This second maximum becomes predominant at long times (t>τFt>\tau_F) and subsequently starts to move to larger distances, while the first peak decreases and eventually fades away, indicating that the dynamics has become diffusive. Remarkably, for all times the distribution displays an exponential-like tail, present even at very short times (t=0.01τ∗t=0.01\tau^*), i.e., the original cages start to “leak” very early, and this leads to the MSD detaching from the plateau. This exponential-like tail persists even at very long times (t>τFt>\tau_F), where Fickian non-Gaussian Diffusion [20, 32, 33] is at play (α2\alpha_2 has not yet decayed to zero). For temperatures above TcT_c, Chauduri et al. have found that exponential tails are present in a variety of glass-formers [34], and thus our finding generalizes these results to the TT-range below TcT_c. The observation that around τ∗\tau^* the exponential tail starts at a distance r≈1r\approx1, indicates that at low temperatures the effective hopping distance is given by the nearest neighbor distance, in contrast to the finding of Ref. [34], for which a roughly three times smaller distance was found for T>TcT>T_c, a value that is compatible with the conclusions reached by the direct analysis of the cage-jumps within single particle trajectories [35]. Thus, this is further evidence that the dynamics above TcT_c differs qualitatively from the one below TcT_c. Figure 3b illustrates the qualitative differences in the shape of Gs(r,t)G_s(r,t) for temperatures above/below TcT_c at fixed t=τ∗(T)t=\tau^*(T). At intermediate temperatures, T=0.55T=0.55, the distribution shows a single maximum and no sign of a shoulder is visible, i.e., the caging core is broad and connects smoothly to the exponential tail. As temperature is lowered significantly below TcT_c, the width of this core shrinks, and a pronounced shoulder is present before the exponential tails. We note that at low TT Gs(r,t)G_s(r,t) at t=τ∗t=\tau^* seems to converge to a TT-independent master-curve, hinting that the beginning of the α\alpha-relaxation process tends to become independent from TT (if measured on a rescaled time scale) and hence “simple”, coherent with the findings of Figure 1 and Figure 2.

Plots of the self part of the van Hove function

Figure 3. (a) rr-dependence of 4πr2Gs(r,t)4\pi r^2G_s(r,t) at T=0.38T=0.38 for different times, as indicated in legend. (b) 4πr2Gs(r,t)4\pi r^2G_s(r,t) as a function of rr, for different temperatures TT, indicated in the legend, at t=τ∗(T)t=\tau^*(T). The vertical dashed lines indicate the length scales r1r_1 and r2r_2, the size of the cage and the hopping distance, respectively. (c) Value of 4πr2Gs(r,t)4\pi r^2G_s(r,t) in r=r1(t)r=r_1(t) and r≃0.9r\simeq0.9, as a function of rescaled time. ⟨r2(t)⟩−rc2\langle r^2(t)\rangle-r_c^2 is plotted as red dashed line, ⟨r2(t)⟩\langle r^2(t)\rangle is plotted as black solid line, for T=0.375T=0.375. Red shaded regions in figure indicate the range where the τ∗\tau^* (see Figure S2) are located at the different displayed temperatures.

Figure 3a and Figure 3b also demonstrate that below TcT_c the dynamics is much more heterogeneous than above TcT_c in that, at the lowest temperatures and up to a time of order τF\tau_F, we have markedly separate populations of particles: Those that are still caged and those that have escaped far from their original cages. This is in contrast to the case of higher temperatures for which this difference is significantly less pronounced. For intermediate TTs, the escape far from the cage is due to a sequence of rare (short) jumps of length around 0.30.3 [34]. Instead, at very low temperatures, we observe a few rare events in which particles hop by a distance of O(1)O(1), before vibrating again in a new cage at r2r_2. (For the lowest TT one indeed sees in Figure 3b a further weak peak at r≈2r\approx2.) In both cases, exponential tails are the result of successions of rare events. The relative importance of the two peaks is quantified in Figure 3c where we plot the tt-dependence of their height for the lowest temperatures. The peak at r=r1r=r_1 is the only one present at really short times, and its magnitude remains basically constant for t≲τ∗t\lesssim\tau^*, i.e. in the time-window in which one observes the 2/32/3 power-law growth in the MSD discussed above. In this time-range, no secondary maximum exists, but we still can read off the value of Gs(r,t)G_s(r,t) at r2=0.9r_2 = 0.9, and one finds that it grows with a 2/3 power-law (see red dashed line). For larger times, τ∗≤t≤τF\tau^* \leq t \leq\tau_F, the shoulder at r2r_2 becomes a real peak and reaches its maximum height, before decreasing at very long times. Only for times well beyond the Fickian onset this secondary peak becomes the highest one, which demonstrates that Fickian diffusion does not imply that all particles have left their original cage.

In conclusion, the sub-diffusive growth of the MSD is due to particles that have hopped once (secondary maximum) and those anomalously far from their original cages (exponential tails). The fact that in this low-TT regime this simple short-time dynamics depends only on t/τFt/\tau_F is highly non-trivial, indicating a distinctive uncaging process of the particles.

To clarify further the link between the 2/3 power-law dependence documented in Figure 2 and the exponential tails in Gs(r,t)G_s(r,t), we present in Figure 4a the one-dimensional Van Hove function at the lowest TTs. To compare the distribution at different times, we plot this probability P(X,t)P(X,t) as a function of the reduced variable X=x/⟨x2(t)⟩X = x/\sqrt{\langle x^2(t)\rangle}. For a Gaussian process, this rescaling results thus in a universal time-independent master-curve: G(X)=2/πexp⁡[−X2/2]G(X) = \sqrt{2/\pi}\exp[-X^2/2], included in the figure as well. For large XX all distributions shown in panel (a) display an excess probability with respect to G(X)G(X). We note that in this representation, the double-peak seen in Figure 3(a) is not visible anymore, but is replaced by a marked shoulder for times around τ∗\tau^*, where deviations from G(X)G(X) are most pronounced. For large displacements, i.e., X≥6=X0X \geq6 = X_0, which is beyond the shoulder at r2r_2, tails at all times are well described by an exponential A(t)exp⁡[−(X−X0)/L(t)]A(t)\exp[-(X-X_0)/L(t)]. [Note that, at t=τ∗t = \tau^*, the condition X>X0X > X_0, corresponds in the 3D real-space to the condition r>r2r > r_2, since 6⟨r2(τ∗)⟩≈0.96\sqrt{\langle r^2(\tau^*)\rangle} \approx0.9.] One sees immediately that the decay length of the tail, L(t)L(t), does not change much as a function of time, while the amplitude A(t)A(t) does. Panels (b) and (c) present, respectively, the time evolution of the amplitude A(t)A(t) and of the length scale L(t)L(t). A(t)A(t) increases significantly (by more than one order of magnitude) for t<τ∗t < \tau^* and peaks at t≃τ∗t \simeq\tau^*. Remarkably, the curves for the different TTs fall onto the same master curve before τ∗\tau^*, showing a power-law with an exponent ≃0.55\simeq0.55. We note that for temperatures above TcT_c no such master curve is observed (see Figure S3). Also two other features of A(t)A(t) for t>τ∗t > \tau^*, namely their similar t-decay and the presence of a shoulder (at around t=8τ∗t = 8\tau^*) are not found at higher temperatures (see Figure S3). These observations show that we are here probing a TT-regime in which the dynamics is qualitatively different from the one at/above TcT_c, and becomes independent of TT, if time is appropriately rescaled. Also L(t)L(t) shows non monotonic behavior, panel (c), with a tt-dependence that is similar for all temperatures, with a maximum approximately at t=0.1τ∗t = 0.1\tau^{*}. Throughout the examined tt-range, however, L(t)L(t) at fixed TT varies at most by 35% of its maximum value. Thus one concludes that it is the amplitude of the tails, rather than their extent, which is the key ingredient for the DH. (In Figure S4 we report the time-evolution of the dimensional decay length l(t)=L(t)/⟨x2(t)⟩l(t) = L(t)/\sqrt{\langle x^2(t)\rangle} and one sees that the variation with time is not very pronounced.) In Figure 4d we present νtail(t)\nu_{\mathrm{tail}}(t), the fraction of particles that have made an XX-displacements larger than X0X_0 (computed as the total area A(t)L(t)A(t)L(t) under the tail of the distribution function). We recognize that up to times τ∗(T)\tau^{*}(T) this fraction shows a nice power-law behavior, with an exponent and prefactor that is independent of TT. Hence, this is further evidence that at these low temperatures the relaxation mechanism is simple, i.e., independent of temperature.

Normalized 1D displacement distribution and exponential-tail parameters

Figure 4. (a) Normalized 1D displacement distribution function as a function of scaled distance XX for different times at T=0.38T = 0.38. The vertical dashed line is located at X=X0X = X_0. The red dashed line is a normal Gaussian. (b) and (c) show the parameters of the exponential fit to the tails, as a function of t/τ∗t/\tau^* for the lowest temperatures. (d) νtail(t)\nu_{\mathrm{tail}}(t) as a function of t/τ∗t/\tau^*. Colors as in Figure 2.

A-dimensional pre-factor $A(t)$ of the exponential fit to the Van Hove tails as a function of rescaled time $t/\tau^{*}$

Figure S3. A-dimensional pre-factor A(t)A(t) of the exponential fit to the Van Hove tails as a function of rescaled time t/τ∗t/\tau^{*}, for temperatures above and below the critical temperature T=0.435T = 0.435, as displayed in the legend.

Length scale $l(t)$ as a function of rescaled time $t/\tau^*$

Figure S4. Length scale l(t)=L(t)⟨x2⟩0.5l(t) = L(t)\langle x^2\rangle^{0.5} of the exponential fit to the Van Hove tails as a function of rescaled time t/τ∗t/\tau^*, for temperatures below the critical temperature T=0.435T = 0.435. Same colors as in the main text.

Finally we make a quantitative comparison between the MSD given by the particles in the tail of the distribution and the total one. In the Supporting Information we give an analytic formula for the ratio ⟨r2(t)⟩tail/⟨r2(t)⟩\langle r^2(t)\rangle_{\mathrm{tail}}/\langle r^2(t)\rangle, where ⟨r2(t)⟩tail\langle r^2(t)\rangle_{\mathrm{tail}} is the just mentioned contribution to the MSD. It is readily shown that, in leading order, this ratio is ∝A(t)L(t)\propto A(t)L(t) (=νtail(t)=\nu_{\mathrm{tail}}(t)). The complete time-dependence of ⟨r2(t)⟩tail\langle r^2(t)\rangle_{\mathrm{tail}} is included in Figure 2a as well (for the sake of clarity only for T=0.375T = 0.375). The figure shows that ⟨r2(t)⟩−rc2\langle r^2(t)\rangle-r_c^2 and ⟨r2(t)⟩tail\langle r^2(t)\rangle_{\mathrm{tail}} have the same t2/3t^{2/3} time-dependence at short times (below τ∗\tau^{*}), and are in fact close to each other. The same holds for the other temperatures below TcT_c. Hence we conclude that the scaling at the beginning of the α\alpha-relaxation can be understood quantitatively from the exponential tail in the Van Hove function, i.e., rare escaping events account for the sub-diffusive t2/3t^{2/3}-departure from the plateau in the post-caging regime.

Discussion and Outlook

Our exploration of the relaxation dynamics at temperatures well below TcT_c reveals that this dynamics shows unexpected features, some of which render the description of the dynamics more complex, while others simplify it.

The fact that the size of the cage decreases linearly with TT, i.e., that the cage is strongly anharmonic [15], is surprising since the MSD shows that particles are trapped in their cage for times spanning several decades of their oscillation period, indicating that the cage is very stable. The linear TT-dependence of rcr_c indicates that, at TTs somewhat below TcT_c, the shape of the cage is not harmonic but has significant higher order components, a result that is important for quantities like the specific heat [36, 37]. The marked presence of these anharmonic effects also implies that the prediction of MCT that at TcT_c the size of the plateau shows a square-root singularity [5] is not correct for the present system, and we deem that this discrepancy may not be a particularity of the glass-former we studied, but is instead a quite general result.

The main result of this work is the finding that around TcT_c the relaxation dynamics changes qualitatively, an insight that has been expected on general arguments [13, 14, 15, 38, 39], but never been probed and documented in detail. While previous studies have shown that the dynamics above TcT_c is described well by MCT [16, 21, 30], we show here that the one below this temperature shows features that are absent at intermediate and high TTs. In particular, we find that at low TT the process that allows the particles to leave their cages is independent of temperature, if time is measured in terms of the Fickian time τF\tau_F. The Van Hove function shows that this process is closely related to rare events in which the particles hop by a distance that is close to the nearest neighbor distance, in contrast with the case of higher TTs, for which the jump length is significantly smaller. In conclusion, rare cage-escaping events give rise to a power-law dependence of the MSD below τ∗\tau^{*}, i.e., short (rescaled) times, with an exponent that is identical to the stretching exponent found in the α\alpha-relaxation process, thus at long times. This equality shows that at low TTs the β\beta-process is linked to the α\alpha-process in a way that is much more direct than at TTs above TcT_c. This implies that the rich and complex relaxation dynamics predicted by MCT above TcT_c is replaced by a significantly simpler escape problem.

The fact that, at low TT, one can clearly identify a hopping length that is given by the nearest neighbor distance, indicates that the relaxation process is directly related to simple structural length scale of the liquid. This makes one to expect that the results presented here are generic, i.e., do not depend on the details of the glass-former considered. Indeed, one can expect that, once the temperature is significantly below TcT_c, the relaxation dynamics will distinctly involve a simple activated process [40] with an activation energy that depends on the strength of the local bonding, i.e., a short-range property of the system. This conclusion is supported by the experimental observation that the temperature-dependence of the α\alpha-relaxation time is independent of the glass-former for temperatures below TcT_c, once the fragility of the glass-former has been scaled out [41, 42, 43]. On the other hand, close to TcT_c, the complex non-linear dynamics as described by MCT makes the TT-dependence of the relaxation time to depend on many details of the interactions, thus determining the fragility of the glass-former.

The above conclusions are based on the analysis of typical liquid state observables, like Gs(r,t)G_s(r,t). It is, however, also interesting to characterize the relaxation dynamics below TcT_c by probing directly the properties of the particle trajectories and identifying the jump processes via a temporal filtering of these trajectories [35, 44, 45, 46]. Such an analysis will allow to study the presence of memory effects in the dynamics and hence to establish a connection with the visco-elastic behavior of the system at low TT.

Furthermore, we note that the quantities presented here are all one-particle observables. As a complementary perspective, it is therefore of interest and important to probe how collective quantities, such as the intermediate scattering function F(q,t)F(q,t) and correlation functions allowing to monitor dynamic heterogeneity, depend on temperature, i.e., whether also they show a simple TT-dependence below TcT_c. From the experimental results mentioned above, Refs [41, 42], one expects that this is indeed true, but a broader verification on the level of the particles is certainly necessary. Also, it is important to study to what extent the results presented here are generic, i.e., whether the here presented simple relaxation dynamics at temperatures below TcT_c is a common feature of glass-forming liquids. Present day computer simulations do allow to investigate this question, and hence such studies should be done since they will allow to verify whether the relaxation dynamics of deeply supercooled glass-formers is actually simpler than the one close to TcT_c. If this is indeed the case, this will be a major step forward in our understanding of glassy dynamics.

Acknowledgments: We acknowledge financial support from the projects MUR-PRIN 2022ETXBEY, “Fickian non-Gaussian diffusion in static and dynamic environments” and MUR-PRIN 2022 PNRR P2022KA5ZZ, funded by the European Union – Next Generation EU. W.K. is a senior member of the Institut Universitaire de France.

Methods and Materials

Details on the simulations: We consider a binary Lennard-Jones mixture with particles that we denote by AA (80%) and BB (20%). The interaction between particles of type α\alpha and β∈{A,B}\beta\in\{A,B\} is given by

Vαβ(r)=4εαβ[(σαβr)12−(σαβr)6],(1)V_{\alpha\beta}(r)=4\varepsilon_{\alpha\beta}\left[\left(\frac{\sigma_{\alpha\beta}}{r}\right)^{12}-\left(\frac{\sigma_{\alpha\beta}}{r}\right)^6\right], \tag*{(1)}

where the interaction parameters are σαβ\sigma_{\alpha\beta} and εαβ\varepsilon_{\alpha\beta} are given in Ref. [16]. Schrøder and Dyre showed that if the AAAA and BBBB forces from this potential are shifted and cut-off at 1.5σAA1.5\sigma_{AA}, while the one for the ABAB interaction is at 2.5σAA2.5\sigma_{AA}, the system is not prone to crystallization [17]. In the present work we use σAA\sigma_{AA} and εAA\varepsilon_{AA}, as units of length and energy, setting the Boltzmann constant kBk_B equal to 1.0.

We have simulated this system using N=32.400N=32.400 particles at a constant density of 1.200, corresponding to a box size of 30. The time step was 0.005 and the simulations were carried out at constant energy. In order to improve the statistics of the results we have done simulations that were significantly longer than the α\alpha-relaxation time (for T≤0.39T\leq0.39 the runs extended over 2⋅1092\cdot10^9 time steps). In addition, we have carried out 4 independent simulations of the system at the various TTs.

A comparison of the static and dynamical properties of the original model with those of the modified one shows, as long as crystallization is avoided, no significant difference of the two systems.

Analysis of single-particle trajectories: The MSD for particles AA, referred in the text as ⟨r2(t)⟩\langle r^2(t)\rangle, is computed as follows:

⟨r2(t)⟩=1NA∑i=1NA⟨(ri(t0+t)−ri(t0))2⟩t0(2)\langle r^2(t)\rangle=\frac{1}{N_A}\sum_{i=1}^{N_A}\left\langle\left(\mathbf{r}_i(t_0+t)-\mathbf{r}_i(t_0)\right)^2\right\rangle_{t_0} \tag*{(2)}

where ri(t)\mathbf{r}_i(t) is the position of particle ii at time tt and ⟨⋅⟩t0\langle\cdot\rangle_{t_0} indicates the average over all time origins t0t_0. In addition, results were averaged over the four independent simulated samples.

The plateau height of the MSD, rc(T)r_c(T), is defined as the square-root of the MSD evaluated at the time at which the logarithmic derivative of the MSD has increased by 5% after reaching its minimum value in the caging regime.

The Self-Intermediate Scattering Function (SISF) for particles AA, referred in the text as Fs(q,t)F_s(\mathbf{q},t), is computed as follows:

Fs(q,t)=1NA∑j=1NA⟨e−iq⋅[rj(t0+t)−rj(t0)]⟩t0(3)F_s(\mathbf{q},t)=\frac{1}{N_A}\sum_{j=1}^{N_A}\left\langle e^{-i\mathbf{q}\cdot[\mathbf{r}_j(t_0+t)-\mathbf{r}_j(t_0)]}\right\rangle_{t_0} \tag*{(3)}

where q\mathbf{q} is the wavevector corresponding to the location of the main peak in the static structure factor S(q)S(q), with q=∣q∣=7.25q=|\mathbf{q}|=7.25. The structural relaxation time τα(T)\tau_\alpha(T) is defined as the time at which the SISF reaches the value 1/e1/e. The height of the plateau of the SISF, fc(T)f_c(T), is computed as the value of the SISF at which the MSD reaches the value of rc2(T)r_c^2(T).

To estimate the TT-dependence of the stretching exponent β\beta of the SISF we measured the quantity m(t;T)=−dlog⁡Fs(q∗,t;T)/dtm(t;T)=-d\log F_s(\mathbf{q}^*,t;T)/dt, which is independent on the height of the plateau fc(T)f_c(T). Since at long times we expect the SISF to be compatible with a stretched-exponential behavior, we fitted log⁡m(t;T)\log m(t;T) in the range [0.5τα,5τα][0.5\tau_\alpha,5\tau_\alpha] with the expression μ(t;T)=a(T)+[β(T)−1]log⁡t\mu(t;T)=a(T)+[\beta(T)-1]\log t to estimate β\beta. In fig:S1a we show the TT-dependence of β\beta obtained from this fitting procedure. In fig:S1b we plot the ratio ϵ(t;T)=m(t;T)/μ2/3(t,T)\epsilon(t;T)=m(t;T)/\mu_{2/3}(t,T), with μ2/3=a′(T)−13log⁡t\mu_{2/3}=a'(T)-\frac{1}{3}\log t being the best fit with the expression μ\mu to the log⁡m\log m-dataset, if β=2/3\beta=2/3 is kept constant. This representation documents the approach, as temperature is decreased towards the lowest TTs, to the master-curve behavior with β=2/3\beta=2/3 of the SISF’s described in the main text.

Analytical approximation for the contribution of the tails to the total MSD: The contribution ⟨r2(t)⟩tail\langle r^2(t)\rangle_{\mathrm{tail}} mentioned in the main text is defined as:

⟨r2(t)⟩tail=3⟨x2(t)⟩tail=3∫x0(t)+∞x2P(x,t) dx(4)\langle r^2(t)\rangle_{\mathrm{tail}}=3\langle x^2(t)\rangle_{\mathrm{tail}}=3\int_{x_0(t)}^{+\infty}x^2P(x,t)\,dx \tag*{(4)}

where x0=X0⟨x2⟩x_0 = X_0\sqrt{\langle x^2\rangle} and X0=6X_0 = 6. Note that in the definition of the 1-D displacement function the density distribution is normalized such that ∫0∞P(x,t) dx=1\int_{0}^{\infty} P(x,t)\,dx = 1 (as already done in the main text). Making the change x→Xx \rightarrow X, and using the fact that for X>X0X > X_0 the distribution in XX is given by an exponential, see Figure 4a, we obtain:

⟨r2(t)⟩tail=3⟨x2(t)⟩∫X0+∞X2A(t)exp⁡[−X−X0L(t)] dX(5)\langle r^2(t)\rangle_{\mathrm{tail}} = 3\langle x^2(t)\rangle\int_{X_0}^{+\infty} X^2 A(t)\exp\left[-\frac{X-X_0}{L(t)}\right]\,dX \tag*{(5)}
=⟨r2⟩A(t)L(t)[X02+2X0L(t)+2L2(t)](6)= \langle r^2\rangle A(t)L(t)\left[X_0^2+2X_0L(t)+2L^2(t)\right] \tag*{(6)}
=⟨r2⟩A(t)L(t)X02[1+2L(t)/X0+2L2(t)/X02].(7)= \langle r^2\rangle A(t)L(t)X_0^2\left[1+2L(t)/X_0+2L^2(t)/X_0^2\right]. \tag*{(7)}

Recalling that, by definition, the fraction of particles in the tail of the distribution is νtail=A(t)L(t)\nu_{\mathrm{tail}} = A(t)L(t), we find

⟨r2(t)⟩tail=⟨r2⟩νtail(t)X02[1+ϕ(t)].(8)\langle r^2(t)\rangle_{\mathrm{tail}} = \langle r^2\rangle\nu_{\mathrm{tail}}(t)X_0^2[1+\phi(t)] \quad. \tag*{(8)}

Here, the function ϕ(t)=2L(t)/X0+2L2(t)/X02\phi(t) = 2L(t)/X_0+2L^2(t)/X_0^2 is small since X0X_0 is appreciable larger than L(t)L(t), see Figure 4. Neglecting ϕ(t)\phi(t) and replacing ⟨r2(t)⟩\langle r^2(t)\rangle by its short time value rc2r_c^2 we then obtain, to leading order, ⟨r2⟩tail≃rc2X02νtail(t)\langle r^2\rangle_{\mathrm{tail}} \simeq r_c^2X_0^2\nu_{\mathrm{tail}}(t) for t<τ∗t < \tau^*, as reported in the main text.

References

Supporting Information for

At low temperatures, glass-forming liquids relax in a simple way

Francesco Rusciano, Raffaele Pastore, Francesco Greco, and Walter Kob

References

  1. [1]S. S. Schoenholz, E. D. Cubuk, D. M. Sussman, E. Kaxiras, and A. J. Liu, A structural approach to relaxation in glassy liquids, Nature Physics 12, 469 (2016).arxiv.org/abs/1506.07772
  2. [2]T. S. Grigera and G. Parisi, Fast monte carlo algorithm for supercooled soft spheres, Phys. Rev. E 63, 045102 (2001).pubmed.ncbi.nlm.nih.gov/11308895
  3. [3]A. Ninarello, L. Berthier, and D. Coslovich, Models and algorithms for the next generation of glass transition studies, Phys. Rev. X 7, 021039 (2017).DOI
  4. [4]B. Kou, Y. Cao, J. Li, C. Xia, Z. Li, H. Dong, A. Zhang, J. Zhang, W. Kob, and Y. Wang, Granular materials flow like complex fluids, Nature 551, 360 (2017).DOI
  5. [5]W. Götze, Complex dynamics of glass-forming liquids: A mode-coupling theory, Vol. 143 (OUP Oxford, 2008).DOI
  6. [6]J. Horbach and W. Kob, Relaxation dynamics of a viscous silica melt: The intermediate scattering functions, Phys. Rev. E 64, 041503 (2001).pubmed.ncbi.nlm.nih.gov/11690029
  7. [7]J. P. Garrahan and D. Chandler, Coarse-grained microscopic model of glass formers, Proceedings of the National Academy of Sciences 100, 9710 (2003).DOI
  8. [8]M. Vogel and S. C. Glotzer, Temperature dependence of spatially heterogeneous dynamics in a model of viscous silica, Phys. Rev. E—Statistical, Nonlinear, and Soft Matter Physics 70, 061504 (2004).DOI
  9. [9]D. Coslovich and G. Pastore, Dynamics and energy landscape in a tetrahedral network glass-former: Direct comparison with models of fragile liquids, Journal of Physics: Condensed Matter 21, 285107 (2009).DOI
  10. [10]H. Z. Cummins, W. M. Du, M. Fuchs, W. Götze, S. Hildebrand, A. Latz, G. Li, and N. J. Tao, Light-scattering spectroscopy of the liquid-glass transition in cakno₃ and in the molecular glass salol: Extended-mode-coupling-theory analysis, Phys. Rev. E 47, 4223 (1993).pubmed.ncbi.nlm.nih.gov/9960499
  11. [11]G. Szamel, Mode-coupling theory and beyond: A diagrammatic approach, Progress of Theoretical and Experimental Physics 2013, 012J01 (2013).DOI
  12. [12]L. M. Janssen, Mode-coupling theory of the glass transition: A primer, Frontiers in Physics 6, 97 (2018).DOI
  13. [13]P. Charbonneau, Y. Jin, G. Parisi, and F. Zamponi, Hopping and the Stokes–Einstein relation breakdown in simple glass formers, Proceedings of the National Academy of Sciences 111, 15025 (2014).DOI
  14. [14]K. S. Schweizer and E. J. Saltzman, Entropic barriers, activated hopping, and the glass transition in colloidal suspensions, The Journal of Chemical Physics 119, 1181 (2003).
  15. [15]S. Sastry, P. G. Debenedetti, and F. H. Stillinger, Signatures of distinct dynamical regimes in the energy landscape of a glass-forming liquid, Nature 393, 554 (1998).DOI
  16. [16]W. Kob and H. C. Andersen, Scaling behavior in the β-relaxation regime of a supercooled Lennard-Jones mixture, Phys. Rev. Lett. 73, 1376 (1994).DOI
  17. [17]T. B. Schröder and J. C. Dyre, Solid-like mean-square displacement in glass-forming liquids, The Journal of Chemical Physics 152, 141101 (2020).DOI
  18. [18]D. Coslovich, M. Ozawa, and W. Kob, Dynamic and thermodynamic crossover scenarios in the Kob-Andersen mixture: Insights from multi-cpu and multi-gpu simulations, The European Physical Journal E 41, 1 (2018).DOI
  19. [19]U. K. Nandi, W. Kob, and S. Maitra Bhattacharyya, Connecting real glasses to mean-field models, The Journal of Chemical Physics 154 (2021).DOI
  20. [20]F. Rusciano, R. Pastore, and F. Greco, Fickian non-Gaussian diffusion in glass-forming liquids, Phys. Rev. Lett. 128, 168001 (2022).
  21. [21]W. Kob and H. C. Andersen, Testing mode-coupling theory for a supercooled binary Lennard-Jones mixture I: The van Hove correlation function, Phys. Rev. E 51, 4626 (1995).
  22. [22]M. Ozawa, W. Kob, A. Ikeda, and K. Miyazaki, Equilibrium phase diagram of a randomly pinned glass-former, Proceedings of the National Academy of Sciences 112, 6914 (2015), https://www.pnas.org/doi/pdf/10.1073/pnas.1500730112.DOI
  23. [23]L. Angelani, R. Di Leonardo, G. Ruocco, A. Scala, and F. Sciortino, Saddles in the energy landscape probed by supercooled liquids, Phys. Rev. Lett. 85, 5356 (2000).DOI
  24. [24]K. Broderix, K. K. Bhattacharya, A. Cavagna, A. Zippelius, and I. Giardina, Energy Landscape of a Lennard-Jones Liquid: Statistics of Stationary Points, Phys. Rev. Lett. 85, 5360 (2000).DOI
  25. [25]S. Sengupta, S. Karmakar, C. Dasgupta, and S. Sastry, Breakdown of the Stokes-Einstein relation in two, three, and four dimensions, The Journal of Chemical Physics 138 (2013).DOI
  26. [26]P. Das and S. Sastry, Crossover in dynamics in the Kob-Andersen binary mixture glass-forming liquid, Journal of Non-Crystalline Solids: X 14, 100098 (2022).arxiv.org/abs/2204.04411
  27. [27]M. D. Ediger, Heterogeneous dynamics in supercooled liquids, Annual review of physical chemistry 51, 99 (2000).pubmed.ncbi.nlm.nih.gov/11031277
  28. [28]R. Richert, Heterogeneous dynamics in liquids: Fluctuations in space and time, Journal of Physics: Condensed Matter 14, R703 (2002).
  29. [29]F. Rusciano, R. Pastore, and F. Greco, Rusciano et al. Reply, Phys. Rev. Lett. 131, 119802 (2023).
  30. [30]W. Kob and H. C. Andersen, Testing mode-coupling theory for a supercooled binary Lennard-Jones mixture. II. Intermediate scattering function and dynamic susceptibility, Phys. Rev. E 52, 4134 (1995).
  31. [31]S. Sen and J. J. Lovi, Trends in the temperature dependence of dynamical heterogeneity in strong and fragile supercooled liquids, Journal of Non-Crystalline Solids 638, 123060 (2024).DOI
  32. [32]F. Rusciano, R. Pastore, and F. Greco, Universal evolution of Fickian non-Gaussian diffusion in two-and three-dimensional glass-forming liquids, International Journal of Molecular Sciences 24, 7871 (2023).
  33. [33]J. M. Miotto, S. Pigolotti, A. V. Chechkin, and S. Roldán-Vargas, Length scales in Brownian yet non-Gaussian dynamics, Phys. Rev. X 11, 031002 (2021).
  34. [34]P. Chaudhuri, L. Berthier, and W. Kob, Universal nature of particle displacements close to glass and jamming transitions, Phys. Rev. Lett. 99, 060604 (2007).DOI
  35. [35]R. Pastore, A. Coniglio, and M. Picca-Ciamarra, Dynamic phase coexistence in glass–forming liquids, Scientific reports 5, 11770 (2015).DOI
  36. [36]J. Horbach, W. Kob, and K. Binder, Specific heat of amorphous silica within the harmonic approximation, The Journal of Physical Chemistry B 103, 4104 (1999).
  37. [37]M. Ozawa, A. Ikeda, K. Miyazaki, and W. Kob, Ideal glass states are not purely vibrational: Insight from randomly pinned glasses, Phys. Rev. Lett. 121, 205501 (2018).DOI
  38. [38]J. Bouchaud, A. Comtet, A. Georges, and P. Le Doussal, Anomalous diffusion in random media of any dimensionality, Journal de Physique 48, 1445 (1987).DOI
  39. [39]G. Porpora, F. Rusciano, R. Pastore, and F. Greco, Comparing microscopic and macroscopic dynamics in a paradigmatic model of glass-forming molecular liquid, International Journal of Molecular Sciences 23, 3556 (2022).DOI
  40. [40]M. De Marzio, G. Camisasca, M. Rovere, and P. Gallo, Microscopic origin of the fragile to strong crossover in supercooled water: The role of activated processes, The Journal of Chemical Physics 146 (2017).DOI
  41. [41]K.-U. Hess, D. Dingwell, and E. Rössler, Parametrization of viscosity-temperature relations of aluminosilicate melts, Chemical Geology 128, 155 (1996).DOI
  42. [42]E. Rössler, K.-U. Hess, and V. Novikov, Universal representation of viscosity in glass forming liquids, Journal of Non-Crystalline Solids 223, 207 (1998).DOI
  43. [43]E. Rössler and A. Sokolov, The dynamics of strong and fragile glass formers, Chemical Geology 128, 143 (1996).
  44. [44]R. Pastore, A. Coniglio, and M. Picca-Ciamarra, From cage-jump motion to macroscopic diffusion in supercooled liquids, Soft Matter 10, 5724 (2014).DOI
  45. [45]A. Jiang, J. Shang, X. Liu, Y. Tang, H. K. Kwan, and Y. Zhu, Efficient CSP algorithm with spatio-temporal filtering for motor imagery classification, IEEE Transactions on Neural Systems and Rehabilitation Engineering 28, 1006 (2020).DOI
  46. [46]J. Helfferich, F. Ziebert, S. Frey, H. Meyer, J. Farago, A. Blumen, and J. Baschnagel, Continuous-time random-walk approach to supercooled liquids. I. Different definitions of particle jumps and their consequences, Phys. Rev. E 89, 042603 (2014).pubmed.ncbi.nlm.nih.gov/24827270

Paper details

vlnc:W_b5dhandu
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents