ARTICLE

Low-frequency vibrational modes of stable glassesLijin Wang1,2, Andrea Ninarello3,4, Pengfei Guan1, Ludovic Berthier 3, Grzegorz Szamel2 & Elijah Flenner2

Unusual features of the vibrational density of states D(ω) of glasses allow one to rationalize

their peculiar low-temperature properties. Simulational studies of D(ω) have been restricted

to studying poorly annealed glasses that may not be relevant to experiments. Here we report

on D(ω) of zero-temperature glasses with kinetic stabilities ranging from poorly annealed to

ultrastable glasses. For all preparations, the low-frequency part of D(ω) splits between

extended and quasi-localized modes. Extended modes exhibit a boson peak crossing over to

Debye behavior (Dex(ω) ~ω2) at low-frequency, with a strong correlation between the two

regimes. Quasi-localized modes obey Dloc(ω) ~ω4, irrespective of the stability. The prefactor

of this quartic law decreases with increasing stability, and the corresponding modes become

more localized and sparser. Our work is the first numerical observation of quasi-localized

modes in a regime relevant to experiments, and it establishes a direct connection between

glasses’ stability and their soft vibrational modes

https://doi.org/10.1038/s41467-018-07978-1 OPEN

1 Beijing Computational Science Research Center, 100193 Beijing, China. 2 Department of Chemistry, Colorado State University, Fort Collins, CO 80523, USA.3 Laboratoire Charles Coulomb (L2C), University of Montpellier, CNRS, 34095 Montpellier, France. 4 CNR-ISC, Uos Sapienza, Piazzale A. Moro 2, 00185Roma, Italy. Correspondence and requests for materials should be addressed to P.G. (email: pguan@csrc.ac.cn) or to E.F. (email: flennere@gmail.com)

NATURE COMMUNICATIONS | (2019) 10:26 | https://doi.org/10.1038/s41467-018-07978-1 | www.nature.com/naturecommunications 1

1234

5678

90():,;

Amorphous solids exhibit universal low-temperatureproperties, seen for instance in the heat capacity andthermal conductivity1, that differ remarkably from crystal

physics. These properties are related to the vibrational density ofstates D(ω). For a continuous elastic medium in three dimen-sions, low-frequency excitations are phonons, and the density ofstates follows D(ω)= ADω2, where AD is given by Debye theory2.A well-known universal feature of amorphous solids is an excessin vibrational modes over the Debye prediction that results in apeak in D(ω)/ω2 at an intermediate frequency, called the bosonpeak3–6.

More recently, another source of ‘excess modes’ has beenidentified in computer simulations of model glasses7–12. It iscomposed of quasi-localized low-frequency modes with a densityobeying Dloc(ω) ~ ω4. Quasi-localized modes are observed atfrequencies significantly lower than the boson peak and the linkbetween the two phenomena is not immediate, despite someindications that they may be connected8,13. The quartic law waspredicted long ago using phenomenological models14,15, reana-lyzed over the years16–18, and remains the focus of intenseresearch19,20. These predictions differ from two recent mean-fieldapproaches21,22, which predict instead a universal non-Debyebehavior that is quadratic in all spatial dimensions, also reportednumerically23. Interest in the low-frequency localized modesextends beyond connections to theoretical models and the bosonpeak. It was suggested that these modes are correlated withirreversible structural relaxation in the supercooled liquid state24,and that the spatial distribution of these soft modes is correlatedwith rearrangements upon mechanical deformation and plasti-city25–28. Localized defects are also central to theoreticaldescriptions of glass properties at cryogenic temperatures29,30.

Recent numerical insights were obtained for glasses that arevery different from the ones studied experimentally, since they areprepared with protocols operating on timescales that differ fromexperimental ones by as many as ten orders of magnitude31. It istherefore unknown whether any of the vibrational, thermal, ormechanical properties derived from earlier computational studyof the density of states is experimentally relevant. For example, itwas reported9,32 that Dloc(ω) ~ ωβ with β ranging from 3 to 4depending on the glass’s stability, with β= 4 for the two moststable simulated glasses created by cooling at a constant rate. Itremains unclear, however, whether β would be different forglasses with stability comparable to that of the experimentalglasses.

Our main achievement is to extend studies of the vibrationaldensity of states of computer glasses to an experimentally relevantregime of glass stability for the first time. To this end, we build onthe recent development of a Monte Carlo method that allows usto equilibrate supercooled liquids down to temperatures belowthe experimental glass transition33–35 to prepare glasses thatcover an unprecedented range of kinetic stability, from extremelypoorly annealed systems to ultrastable glasses. We thus match thelarge gap between previous numerical findings and the experi-mental regime36. Recent studies have shown that that such stableglasses may differ qualitatively from ordinary computerglasses35,37,38. For example, qualitatively different yielding beha-vior of well-annealed glasses compared to that of poorly annealedglasses was reported in ref. 38. Since rearrangements uponmechanical deformation are correlated with the spatial distribu-tion of soft modes, this result suggested that the density of statescould also evolve dramatically with the stability.

ResultsSystem preparation. We prepare glasses by instantaneouslyquenching supercooled liquids equilibrated at parent temperature

Tp to T= 0, so that Tp uniquely controls the glass stability. Wefind that the low-frequency part of the vibrational density ofstates changes considerably when Tp varies, thus offering a directlink between soft vibrational modes and kinetic stability. Fol-lowing earlier work8,10, we divide modes into extended and quasi-localized ones. As found for high parent temperature glasses7–10,the density of states of the quasi-localized modes follows Dloc=A4ω4, with the same quartic exponent for all glass stabilities. Ourwork thus establishes the relevance of earlier findings aboutquasi-localized modes and their effect on the density of states inthe experimentally relevant regime of glass stability. In addition,we find that the overall scale A4 decreases surprisingly rapidlywhen Tp decreases, showing that the density of the quasi-localizedmodes is highly sensitive to the glass stability. This rapid decreasecontrasts with the modest changes found for other structuralquantities, such as mechanical moduli, sound speed, and Debyefrequency. Quasi-localized modes also become sparser andincreasingly localized at low Tp, and so the identification of softlocalized modes as relevant glassy defects controlling the physicsof amorphous solids becomes more convincing near the experi-mental glass transition. Our results also suggest that ultrastableglasses contain significantly fewer localized excitations thanordinary glasses, which appears consistent with recent experi-ments39–41.

We simulate a polydisperse glass forming system in threedimensions, which is a representative glass-forming computermodel33. We use the swap Monte Carlo algorithm to prepareindependent equilibrated configurations at parent temperaturesTp ranging from above the onset temperature of slow dynamicsTo ≈ 0.200, down to Tp= 0.062, which is about 60% of the mode-coupling temperature Tc ≈ 0.108 (Tc marks a crossover toactivated dynamics and corresponds typically to the lowesttemperature accessed by standard molecular dynamics). Impor-tantly, our lowest Tp is lower than the estimated experimentalglass temperature Tg ≈ 0.07233, and no previous computationalstudy has explored such range of glass stability. In addition, wealso use a very high parent temperature which we refer to asTp=∞. We then probe vibrational properties of zero-temperature glasses produced by an instantaneous quench fromequilibrated configurations at different Tp. The specific simulationdetails are provided in Methods.

Macroscopic properties. We begin by presenting macroscopicproperties of the glasses as a function of the parent temperatureTp. The inherent structure energy EIS is directly related to themobility of the particles42, and thus we show EIS in Fig. 1a as anindicator of the increased stability of the glass. EIS deviates fromits high-temperature plateau when Tp becomes smaller thanthe onset temperature, and decreases further with decreasingTp43. Similarly, the bulk modulus B decreases modestlywith decreasing Tp (Fig. 1b). By contrast, the shear modulus Gin Fig. 1c remains nearly temperature-independent until the-mode-coupling temperature, which is below the onset tempera-ture, and then the shear modulus increases with decreasing Tp.Associated with the increase in the shear modulus is a decrease inthe Debye level AD ¼ 3=ω3

D, where the Debye frequencyωD ¼ 18π2ρð Þ= c�3

l þ 2c�3t

�� �1=3. The decrease of AD is mainly

controlled by the increase of the shear modulus since the trans-verse speed of sound ct ¼

ffiffiffiffiffiffiffiffiG=ρ

pis 2.4–2.6 times smaller than the

longitudinal speed of sound cl. The overall relative variations ofmechanical moduli and Debye frequency are, however, relativelymild given the broad range of glass stabilities covered in Fig. 1.

Classification of quasi-localized and extended modes. Byexamining the participation ratio P(ω) as a function of ω at

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-018-07978-1

2 NATURE COMMUNICATIONS | (2019) 10:26 | https://doi.org/10.1038/s41467-018-07978-1 | www.nature.com/naturecommunications

different parent temperatures (see Fig. 2), we observe all thefeatures that characterize the Tp-dependence of the density ofstates. A value of P(ω)= 1 indicates a mode where all the particlesparticipate equally, a value of P(ω)=N−1 indicates a mode whereonly one particle participates, and a value of P(ω)= 2/3 indicatesa plane wave. The sharp peaks in P(ω) at low frequencies are dueto the phonon modes, with the first peak corresponding to thefirst allowed transverse phonon at ωt= ct2π/L, L being the boxlength. An increase in ωt indicates an increase in ct ¼

ffiffiffiffiffiffiffiffiG=ρ

p. The

low-frequency modes can be naturally divided into quasi-localized modes (small P) and extended modes (large P)through an appropriate thresholding procedure8,10, this decom-position becoming sharper as L increases and Tp decreases. Thevalue P0= 0.006 is appropriate, as shown in Fig. 2, but wechecked that our results are not qualitatively affected by a rea-sonable change of P0. As Tp decreases, phonon modes shift tolarger frequencies, as expected from the evolution of themechanical moduli, whereas quasi-localized modes becomeincreasingly localized and well-separated from the phonons. Wealso checked that our results hold for small system sizes whereallowed phonon modes are shifted to much higher frequencies7.

Properties of quasi-localized modes. We examined the densityof states for the quasi-localized modes Dloc(ω), which are shownin Fig. 3a for a few representative Tp. At low frequencies, Dloc(ω)=A4ω4 for each parent temperature with a prefactor A4 thatdepends on the glass stability. We show the resulting A4(Tp) inFig. 3b. The prefactor A4 stays nearly constant for high enoughTp, but decreases sharply when Tp decreases below the mode-coupling temperature Tc. This observation is robust againstchanging the system size. The decrease of A4 at low Tp correlateswell with the evolution of shear modulus and Debye level inFig. 3c, d. We note that a study of less stable glasses32 found anincrease in the lowest frequency of quasi-localized modes withdecreasing parent temperature, which, under certain assumptions,may be related to the change of A4 reported here. A major resultof our study is that the quartic law governing Dloc(ω) is obeyedirrespective of the glass stability, thus extending the validity ofprevious findings to the experimentally relevant regime.

In Fig. 3c we show the probability distribution for finding amode with a participation ratio P for the modes with P < P0 for N= 48,000 particles. With decreasing Tp, the distribution becomes

narrower and the peak position shifts to smaller P values. We findthat the average participation ratio decreases with decreasing Tp,which is evident from Fig. 3c. This confirms that these modesbecome more localized with decreasing parent temperature,which had been observed for less stable glasses9,13,32. Since thedensity of states is a function of the structure of the quenchedsystem, we conclude that subtle local structural changes occur forTp below Tc that strongly affect soft vibrational motion in thequenched glass.

To visualize the increasing mode localization, we define a‘softness’25 for particle i as AðiÞ ¼ ð1=MÞPM

l¼1 jel;ij, where thesum is taken over the M= 40 lowest frequency quasi-localizedmodes for one inherent structure (we have checked that ourconclusions hold when we take other values of M= 5–40). Thesoftness quantifies the vibrational amplitude of low-frequencyquasi-localized modes. In the snapshots of Fig. 4, particles arerepresented with a size proportional to A(i) for (a) Tp= 0.200 and(b) Tp= 0.062. For the highest Tp, clusters contributing tolocalized modes are relatively numerous, quite extended, andstrongly coupled to their environment. At the lowest Tp, eachcluster is localized around just a few particles, there are muchfewer clusters, and they offer a stronger contrast with the

0.28 4.5

4.0

3.5

4×10–3

5×10–3

0.24EIS

AD

G

B

0.20

20.0

19.0

18.0

17.00 0.1 0.2 0.3

Tp

0 0.1 0.2 0.3

Tp

a

b

c

d

Fig. 1 Macroscopic properties. a Inherent structure energy EIS; b Bulk modulus B; c Shear modulus G; d Debye level AD. In all panels, the vertical dashed-dotted, dashed, and dotted lines mark the positions of Tg, Tc, and To, respectively. Data shown are for N= 48,000 (circles), 96,000 (squares), and192,000 (triangles)

100

10–1

10–2

10–3

P (

ω)

10–4

10–5

10–6

0.2ω ω0.4 0.6 0.8 0.2

N = 48KN = 96KN = 192K

0.4 0.6 0.81 1

a b

Fig. 2 Classification of modes. Participation ratio P(ω) as a function offrequencies ω combined from systems with N= 48,000, 96,000, and192,000 at parent temperatures Tp= 0.200 in (a) and Tp= 0.062 in(b). The horizontal line marks the threshold P0= 0.006 between extendedand quasi-localized modes

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-018-07978-1 ARTICLE

NATURE COMMUNICATIONS | (2019) 10:26 | https://doi.org/10.1038/s41467-018-07978-1 | www.nature.com/naturecommunications 3

immobile background. To quantify these observations, wemeasured the probability distribution of A(i) (Fig. 4c). Thesedistributions show a power-law tail at large A values, PðAiÞ ¼λðTpÞA�α

i with α ≈ 3.7. At low Tp the tail is well separated fromthe core of the distribution at small A, and mobile particles withlarge A are better defined. There is also a pronounced decay of theprobability of finding large A values at low Tp since λ(0.2)/λ(0.062) ≈ 4.3, which indicates a greater than four fold decrease inthe number of soft particles with large vibrational amplitudes.The interpretation of quasi-localized modes as relevant glassydefects controlling mechanical and thermal properties of glasses istherefore more convincing for stable glasses than it is forconventional computer glasses.

Properties of extended modes. Next, we examine the density ofstates of extended modes, Dex(ω), with a participation ratiogreater than P0. In Fig. 5a, b we show the reduced density of statesDex(ω)/ω2 for two parent temperatures. For each temperature, theDebye level is reached at low enough ω and a boson peak isobserved at larger frequencies. Using our localization criterion,we find that modes near the boson peak are not localized, but thisdoes not imply that they have a phononic character. The bosonpeak narrows slightly with decreasing Tp. The Debye level, theboson peak location, height, and width all change modestly as Tpis varied over the entire range studied. The changes observed inour study agree qualitatively with those found by Grigera et al. 4.

In Fig. 5c we examine scaling properties of the density of statesof extended modes. We rescale ω by the boson peak frequency,

ωBP, and plot the rescaled density of states Dex/(ADω2). Weobserve an excellent collapse on the low-frequency side of theboson peak. This shows that in this frequency range the reducedfrequency dependence has a universal shape, as reported before44.Second, the collapse also shows that the height of the boson peakcorrelates with the Debye level AD. These results agree withexperiments on molecular glass formers45–47. However, some ofthe same experiments report that the boson peak position scalesas the Debye frequency45,47, which is not consistent with ourresults. We also find that a scaling of ωBP with the bulk modulussuggested in ref. 48 is inconsistent with our results. Note that westudy the evolution of the boson peak as a function of thepreparation temperature, while experiments sometimes examinethe temperature evolution of the boson peak for a given glasspreparation. We also note that a correlation between the bosonpeak and quasi-localized modes has been proposed by studyingsystems at different pressures around the unjamming transition49.

Since the boson peak occurs in a different frequency range thanthe ω4 scaling of Dloc(ω), it is not clear that there could be arelationship between the boson peak and the low-frequencyquasi-localized modes. Simulations close to jamming suggest thatA4 � ω4

BP8, but we do not find that this relation holds with

changing Tp. An alternative possibility can be obtained fromdimensional analysis, where a characteristic frequency for quasi-localized modes can be defined as A�1=5

432. We find that A�1=5

4 �ωBP for glasses with Tp < Tc (Fig. 6), but this relation does nothold for glasses created with Tp > Tc. We note that ωBP is constantfor Tp > Tc, see the inset to Fig. 6, and only changes for Tp < Tc.Again we find that Tc marks a change in the behavior of D(ω).Given the relatively small changes in both ωBP and A�1=5

4 over our

10–3 Tp = 0.062

Tp = 0.075

Tp = 0.085

Tp = 0.100

Tp = 0.120

Tp = 0.200

Tp = 0.062 N = 48K

Tp = 0.075 N = 48K

Tp = 0.100 N = 48K

Tp = 0.200 N = 48K

10–4

10–5

1000

800

600

D (

P)

400

200

00 0.001 0.002

P

0.003 0.004

0.2 0.4 0.60.81 0 0.1 0.20.00

0.02

0.04

A4

TP

Tg Tc

ω

Dlo

c (ω

)

a b

c

Fig. 3 Density of states and spatial localization of quasi-localized modes.a Density of states Dloc(ω) for quasi-localized modes for N= 48,000, withfits to Dloc= A4ω4. b Tp dependence of A4 for N= 48,000 (circles),96,000 (squares), and 192,000 (triangles), with the mode-couplingtemperature Tc and the estimated experimental glass temperature Tgindicated. c Probability distribution of the participation ratio for quasi-localized modes in the frequency-range of the ω4 scaling for various Tp

104

103 TP = 0.062

TP = 0.200102

101

P(A

i)

100

10–1

10–2

10–3

10–4

10–3 10–2 10–1

Ai

a b

c

Fig. 4 Softness of quasi-localized modes. Snapshots obtained for Tp=0.200 (a) and Tp= 0.062 (b) with N= 192,000. The particles are shownwith their radius given by the vibrational amplitude A(i) calculated from thelowest quasi-localized modes. c The probability distribution of A(i) for Tp=0.200 and Tp= 0.062. For lower parent temperatures there is a smallerfraction of the particles with larger A(i), and thus the modes are morelocalized

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-018-07978-1

4 NATURE COMMUNICATIONS | (2019) 10:26 | https://doi.org/10.1038/s41467-018-07978-1 | www.nature.com/naturecommunications

entire range of parent temperatures studied, it is not clear that apower law is the proper relationship between these quantities andfurther work is needed to verify it.

DiscussionIn summary, we report the first characterization of the vibrationaldensity of states of computer glasses prepared over a range ofglass stability that bridges the gap between ordinary simulationsand experimental studies. At low-frequency extended and quasi-localized modes coexist, and both types of modes evolve differ-ently when the glass stability is varied. We find a relatively mildtemperature dependence of extended modes, with a strong cor-relation between the Debye level and the boson peak. By contrast,quasi-localized modes evolve more strongly when Tp decreasesbelow the mode-coupling temperature, but their density of statesis always described by Dloc � A4ω

4. Unexpectedly, the tempera-ture dependence of the prefactor A4(Tp) is more interesting thanthe value of the quartic exponent, which is insensitive to thedegree of annealing.

The increasing localization of the modes implies that subtle yetsignificant changes occur in the local structure of the glass thatare not reflected in the pair correlation function, which is nearlyidentical for parent temperatures below Tc. Since soft modes havebeen linked to irreversible relaxation24 and rearrangements undershear25–28, it follows that the reduction of these soft modes canhave significant implications for glassy dynamics. In turn thisreduction indicates that there are fewer soft spots, which shouldincrease the strength of the glass. This hypothesis is supported bythe observation that the decrease in Dloc(ω) mirrors the increaseof the shear modulus, and also correlates very well with theevolution of the ductility of the produced glasses38,50. Since wecan now equilibrate amorphous systems at temperatures lowenough so that they do not flow, another perspective would be toanalyze the density of states at finite temperatures through theFourier transform of the velocity autocorrelation function51, or bydiagonalizing the covariance matrix of displacements52. Futurestudies should examine the difference between these proceduresto provide insights into thermal anharmonicities of stable glasses,and more generally into their low-temperature transportproperties.

MethodsSimulations. We simulate a polydisperse model glass former of sizes between N=48,000 and 450,000 particles with equal mass at a number density ρ= 1.033. Theinteraction between two particles i and j is given by VðrijÞ ¼ σ ij

rij

� �12þvðrijÞ when

their separation rij � rcij ¼ 1:25σij and zero otherwise. We use

vðrijÞ ¼ c0 þ c2rijσ ij

� �2þc4

rijσ ij

� �4, where the coefficients c0, c2, and c4 ensure the

continuity of V(rij) up to the second derivative at the cutoff rcij . The probability ofparticle diameters σ is P(σ)= A/σ3, where σ∈[0.73,1.63] and we use a non-additivemixing rule, σ ij ¼ σ i þ σj

2 ð1� 0:2jσi � σ jjÞ. For N ≤ 192,000 we use the swap MonteCarlo algorithm to prepare independent equilibrated configurations at parenttemperatures Tp ranging from above the onset temperature of slow dynamics (To ≈0.200) down to Tp= 0.062, which is about 60% of the mode-coupling temperature(Tc ≈ 0.108), and is lower than the estimated experimental glass temperature (Tg ≈0.072)33. In addition, we also use a very high parent temperature, which we refer toas Tp=∞. Due to very long equilibration times for systems of N > 192,000 particleswe only study systems with N > 192,000 for Tp=∞.

Density of states calculation. Following equilibration at a temperature Tp, zero-temperature glasses are produced by instantaneously quenching equilibrium con-figurations to their inherent structures using the Fast Inertia Relaxation Enginealgorithm53. We then calculate the modes by diagonalizing the Hessian matrixusing Intel Math Kernel Library (https://software.intel.com/en-us/mkl/) andARPACK (http://www.caam.rice.edu/software/ARPACK/). We calculate all thenormal modes for the 48,000 particle systems, but only the low-frequency part ofthe spectrum in systems with N > 48,000. We characterize the modes through thedensity of states DðωÞ ¼ 1

3N�3

P3N�3l¼1 δðω� ωlÞ and the participation ratio

0.1 1 100.0

1.0

2.0

3.0

0.02

0.01

0.00

0.1 1 10

Tp = 0.062

Tp = 0.085

Tp = 0.100

Tp = ∞

N = 48KN = 96KN = 192KN = 250KN = 300KN = 350KN = 400KN = 450K

Dex

(ω)/

ω2

Dex

(ω)/

(AD

ω2 )

ω1 10

ω

ω/ωBP

a b

c

Fig. 5 Density of states for extended modes. Reduced density of states forextended modes, Dex(ω)/ω2, at parent temperatures Tp=∞ (a) and 0.062(b), for systems with different sizes. The black line in each panel indicatesthe Debye level AD while the vertical arrow marks the frequency, whereDex(ω)/ω2 starts to deviate from AD. c Rescaled version of the same datausing the Debye level (vertical axis) and the boson peak frequency ωBP

(horizontal axis)

1.6

2.0

2.4

2.8

3.2

0.6

0.8

1

1.2

1.4

1.6

ωB

P

Tp = 0.11 > Tc

A4–1

/5

0.6 0.8 1.61 1.2 1.4ωBP

0 0.1 0.2 0.3Tp

Tp = 0.1 < Tc

Fig. 6 Scaling of the boson peak frequency. The characteristic frequencyA�1=54 versus the boson peak position ωBP. The line is a fit A�1=5

4 � ωBP forglasses whose Tp < Tc, which is the parent temperature range where we seean increase in ωBP with decreasing Tp, see inset where the vertical dashed-dotted, dashed lines mark the positions of Tg and Tc, respectively

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-018-07978-1 ARTICLE

NATURE COMMUNICATIONS | (2019) 10:26 | https://doi.org/10.1038/s41467-018-07978-1 | www.nature.com/naturecommunications 5

PðωlÞ ¼PN

i¼1jel;i j2

� �2

NPN

i¼1jel;i j4

, where el,i is the polarization vector of particle i in mode l with

frequency ωl. For a mode localized to one particle P(ω)=N−1, and for an idealplane wave P(ω)= 2/3. The phonon modes occur at discrete frequencies, and carehas to be taken in the binning procedure to calculate the density of states ofextended modes, Dex(ω). To perform this calculation, we determine the phononfrequencies from the peak positions of the participation ratio versus frequency, andtune the bin size to smooth Dex(ω).To obtain the shear modulus G and the bulkmodulus B we use the method described in ref. 54.

Data availabilityAll data will be available from the authors upon request.

Received: 28 August 2018 Accepted: 4 December 2018

References1. Zeller, R. C. & Pohl, R. O. Thermal conductivity and specific heat of

noncrystalline solids. Phys. Rev. B 4, 2029–2041 (1971).2. Kittel, C. Introduction to Solid State Physics. 7th edn (Wiley, New York, 1996).3. Nakayama, T. Boson peak and terahertz frequency dynamics of vitreous silica.

Rep. Prog. Phys. 65, 1195–1242 (2002).4. Grigera, T. S., Martin-Mayor, V., Parisi, G. & Verrocchio, P. Phonon

interpretation of the boson peak in supercooled liquids. Nature 422, 289–292(2003).

5. Inoue, K. et al. Low-energy excitations in amorphous polymers. J. Chem. Phys.95, 5332–5340 (1991).

6. Buchenau, U., Wischnewski, A., Ohl, M. & Fabiani, E. Neutron scatteringevidence on the nature of the boson peak. J. Phys. Condens. Matter 19, 205106(2007).

7. Lerner, E., Düring, G. & Bouchbinder, E. Statistics and properties of low-frequency vibrational modes in structural glasses. Phys. Rev. Lett. 117, 035501(2016).

8. Mizuno, H., Shiba, H. & Ikeda, A. Continuum limit of the vibrationalproperties of amorphous solids. Proc. Natl Acad. Sci. USA 114, E9767–E9774(2017).

9. Lerner, E. & Bouchbinder, E. Effect of instantaneous and continuous quencheson the density of vibrational modes in model glasses. Phys. Rev. E 96, 020104(R) (2017).

10. Shimada, M., Mizuno, H. & Ikeda, A. Anomalous vibrational properties in thecontinuum limit of glasses. Phys. Rev. E 97, 022609 (2018).

11. Kapteijns, G., Bouchbinder, E. & Lerner, E. Universal nonphononic density ofstates in 2D, 3D, and 4D glasses. Phys. Rev. Lett. 121, 055501 (2018).

12. Angelani, L., Paoluzzi, M., Parisi, G. & Ruocco, G. Probing the non-Debyelow-frequency excitations in glasses through random pinning. Proc. NatlAcad. Sci. USA 115, 8700–8704 (2018).

13. Lerner, E. & Bouchbinder, E. Frustration-induced internal stresses areresponsible for quasilocalized modes in structural glasses. Phys. Rev. E 97,032140 (2018).

14. Buchenau, U., Galperin, Yu. M., Gurevich, V. L. & Schober, H. R. Anharmonicpotentials and vibrational localization in glasses. Phys. Rev. B 43, 5309 (1991).-5045.

15. Buchenau, U. et al. Interaction of soft modes and sound waves in glasses. Phys.Rev. B 46, 2798–2808 (1992).

16. Schober, H. R. & Oligschleger, C. Low-frequency vibrations in a model glass.Phys. Rev. B 53, 11469–11480 (1996).

17. Gurevich, V. L., Parshin, D. A. & Schober, H. R. Anharmonicity, vibrationalinstability, and the boson peak in glasses. Phys. Rev. B 67, 094203 (2003).

18. Schirmacher, W., Ruocco, G. & Scopigno, T. Acoustic attenuation in glassesand its relation with the boson peak. Phys. Rev. Lett. 98, 025501 (2007).

19. Benetti, F. P. C., Parisi, G., Pietracaprina, F. & Sicuro, G. Mean-field model forthe density of states of jammed soft spheres. Phys. Rev. E 97, 062157 (2018).

20. Stanifer, E., Morse, P. K., Middleton, A. A. & Manning, M. L. A simplerandom matrix model for the vibrational spectrum of jammed packings.Preprint at http://arXiv.org/abs/1804.04074 (2018).

21. DeGiuli, E., Laversanne-Finot, A., Düring, G., Lerner, E. & Wyart, M. Effectsof coordination and pressure on sound attenuation, boson peak and elasticityin amorphous solids. Soft Matter 10, 5628–5644 (2014).

22. Franz, S., Parisi, G., Urbani, P. & Zamponi, F. Universal spectrum of normalmodes in low-temperature glasses. Proc. Natl Acad. Sci. USA 112,14539–14544 (2015).

23. Charbonneau, P., Corwin, E. I., Parisi, G., Poncet, A. & Zamponi, F. Universalnon-Debye scaling in the density of states of amorphous solids. Phys. Rev. Lett.117, 045503 (2016).

24. Widmer-Cooper, A., Perry, H., Harrowell, P. & Reichman, D. R. Irreversiblereorganization in a supercooled liquid originates from localized soft modes.Nat. Phys. 4, 711–715 (2008).

25. Schoenholz, S. S., Liu, A. J., Riggleman, R. A. & Rottler, J. Understandingplastic deformation in thermal glasses from single-soft-spot dynamics. Phys.Rev. X 4, 031014 (2014).

26. Chen, K. et al. Measurement of correlations between low-frequencyvibrational modes and particle rearrangements in quasi-two-dimensionalcolloidal glasses. Phys. Rev. Lett. 107, 108301 (2011).

27. Manning, M. L. & Liu, A. J. Vibrational modes identify soft spots in a sheareddisordered packing. Phys. Rev. Lett. 107, 108302 (2011).

28. Zylberg, J., Lerner, E., Bar-Sinai, Y. & Bouchbinder, E. Local thermal energy asa structural indicator in glasses. Proc. Natl Acad. Sci. USA 114, 7289–7294(2017).

29. Phillips, W. A. Tunneling states in amorphous solids. J. Low Temp. Phys. 7,351–360 (1972).

30. Anderson, P. W., Halperin, B. I. & Varma, C. M. Anomalous low-temperaturethermal properties of glasses and spin glasses. Philos. Mag. 25, 1–9 (1972).

31. Berthier, L. & Biroli, G. Theoretical perspective on the glass transition andamorphous materials. Rev. Mod. Phys. 83, 587–645 (2011).

32. Lerner, E. & Bouchbinder, E. A characteristic energy scale in glasses. J. Chem.Phys. 148, 214502 (2018).

33. Ninarello, A., Berthier, L. & Coslovich, D. Models and algorithms for the nextgeneration of glass transition studies. Phys. Rev. X 7, 021039 (2017).

34. Berthier, L., Coslovich, D., Ninarello, A. & Ozawa, M. Equilibrium sampling ofhard spheres up to the jamming density and beyond. Phys. Rev. Lett. 116,238002 (2016).

35. Fullerton, C. & Berthier, L. Density controls the kinetic stability of ultrastableglasses. Europhys. Lett. 119, 36003 (2017).

36. Berthier, L. et al. Configurational entropy measurements in extremelysupercooled liquids that break the glass ceiling. Proc. Natl Acad. Sci. USA 114,11356–11361 (2017).

37. Scalliet, C., Berthier, L. & Zamponi, F. Absence of marginal stability in astructural glass. Phys. Rev. Lett. 119, 205501 (2017).

38. Ozawa, M., Berthier, L., Biroli, G., Rosso, A. & Tarjus, G. Random criticalpoint separates brittle and ductile yielding transitions in amorphous materials.Proc. Natl Acad. Sci. USA 115, 6656–6661 (2018).

39. Pérez-Castañeda, T., Rodrguez-Tinoco, C., Rodrguez-Viejo, J. & Ramos, M. A.Suppression of tunneling two-level systems in ultrastable glasses ofindomethacin. Proc. Natl Acad. Sci. USA 111, 11275–11280 (2014).

40. Liu, X., Queen, D. R., Metcalf, T. H. J., Karel, E. & Hellman, F. Hydrogen-freeamorphous silicon with no tunneling states. Phys. Rev. Lett. 113, 025503(2014).

41. Ramos, M. A. & Pérez-Castañeda, T. Do tunneling states and boson peakpersist or disappear in extremely stabilized glasses? J. Low Temp. Phys. 41,412–418 (2015).

42. Helfferich, J., Lyubimov, I., Reid, D. & de Pablo, J. J. Inherent structure energyis a good indicator of molecular mobility in glasses. Soft Matter, 12,5898–5904 (2016).

43. Sastry, S., Debenedetti, P. G. & Stillinger, F. H. Signatures of distinctdynamical regimes in the energy landscape of a glass-forming liquid. Nature393, 554–557 (1998).

44. Shintani, H. & Tanaka, H. Universal link between the boson peak andtransverse phonons in glass. Nat. Mater. 7, 870–877 (2008).

45. Monaco, A. et al. Density of vibrational states of a hyperquenched glass. Phys.Rev. Lett. 96, 205502 (2006).

46. Baldi, G. et al. Connection between boson peak and elastic properties insilicate glasses. Phys. Rev. Lett. 102, 195502 (2009).

47. Monaco, A. et al. Effect of densification on the density of vibrational states ofglasses. Phys. Rev. Lett. 97, 135501 (2006).

48. Rufflé, B. et al. Scaling the temperature-dependent boson peak of vitreoussilica with the high-frequency bulk modulus derived from Brillouin scatteringdata. Phys. Rev. Lett. 104, 067402 (2010).

49. Shimada, M., Mizuno, H., Wyart, M. & Ikeda, A. Spatial structure of quasi-localized vibrations in nearly jammed amorphous solids. Preprint at http://arXiv.org/abs/1804.08865 (2018).

50. Ketkaew, J. et al. Mechanical glass transition revealed by the fracturetoughness of metallic glasses. Nat. Commun. 9, 3271 (2018).

51. Keyes, T. Instantaneous normal mode approach to liquid state dynamics. J.Phys. Chem. A 101, 2921–2930 (1997).

52. Henkes, S., Brito, C. & Dauchot, O. Extracting vibrational modesfrom fluctuations: a pedagogical discussion. Soft Matter 8, 6092–6109(2012).

53. Bitzek, E., Koskinen, P., Gähler, F., Moseler, M. & Gumbsch, P. Structuralrelaxation made simple. Phys. Rev. Lett. 97, 170201 (2006).

54. O’Hern, C. S., Silbert, L. E., Liu, A. J. & Nagel, S. R. Jamming at zerotemperature and zero applied stress: The epitome of disorder. Phys. Rev. E 68,011306 (2003).

ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-018-07978-1

6 NATURE COMMUNICATIONS | (2019) 10:26 | https://doi.org/10.1038/s41467-018-07978-1 | www.nature.com/naturecommunications

AcknowledgementsWe thank J.-L. Barrat for discussions. L.W., E.F., and G.S. acknowledge funding fromNSF DMR-1608086. This work was also supported by a grant from the Simons foun-dation (No. 454933 L.B.). L.W. and P.G. acknowledge support from the National NaturalScience Foundation of China (No. 51571011), the MOST 973 Program (No.2015CB856800), and the NSAF joint program (No. U1530401). We acknowledge thecomputational support from Beijing Computational Science Research Center.

Author contributionsL.B., G.S., and E.F. designed the project. L.W. and A.N. performed numerical simula-tions. L.W., A.N., L.B., P.G., G.S., and E.F. contributed to the analysis of data. L.W., L.B.,G.S., and E.F. wrote the paper.

Additional informationCompeting interests: The authors declare no competing interests.

Reprints and permission information is available online at http://npg.nature.com/reprintsandpermissions/

Journal peer review information: Nature Communications thanks Miguel Ramos andthe other anonymous reviewer(s) for their contribution to the peer review of this work.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims inpublished maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attri-bution 4.0 International License, which permits use, sharing, adaptation,

distribution and reproduction in any medium or format, as long as you give appropriatecredit to the original author(s) and the source, provide a link to the Creative Commonslicense, and indicate if changes were made. The images or other third party material in thisarticle are included in the article’s Creative Commons license, unless indicated otherwise ina credit line to the material. If material is not included in the article’s Creative Commonslicense and your intended use is not permitted by statutory regulation or exceeds thepermitted use, you will need to obtain permission directly from the copyright holder. Toview a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

© The Author(s) 2019

NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-018-07978-1 ARTICLE

NATURE COMMUNICATIONS | (2019) 10:26 | https://doi.org/10.1038/s41467-018-07978-1 | www.nature.com/naturecommunications 7