Emergence and percolation of rigid domains during colloidal glass transition

Xiunan Yang+,1, 2 Hua Tong+,3 Wei-Hua Wang∗,4, 2, 5 and Ke Chen∗1, 2, 5

1Beijing National Laboratory for Condensed Matter Physics and Key Laboratory of Soft Matter Physics,Institute of Physics, Chinese Academy of Sciences, Beijing 100190, People’s Republic of China

2University of Chinese Academy of Sciences, Beijing 100049, People’s Republic of China3Department of Fundamental Engineering, Institute of Industrial Science,University of Tokyo, 4-6-1 Komaba, Meguro-ku, Tokyo 153-8505, Japan

4Institute of Physics, Chinese Academy of Sciences, Beijing 100190, People’s Republic of China5Songshan Lake Materials Laboratory , Dongguan, Guangdong 523808, China

(Dated: February 25, 2019)

Using video microscopy, we measure local spatial constraints in disordered binary colloidal sam-ples, ranging from dilute fluids to jammed glasses, and probe their spatial and temporal correlationsto local dynamics during the glass transition. We observe the emergence of significant correlationsbetween constraints and local dynamics within the Lindemann criterion, which coincides with theonset of glassy dynamics in supercooled liquids. Rigid domains in fluids are identified based onlocal constraints, and demonstrate a percolation transition near glass transition, accompanied bythe emergence of dynamical heterogeneities. Our results show that the spatial constraints insteadof the geometry of amorphous structures is the key that connects the complex spatial-temporalcorrelations in disordered materials.

A liquid solidifies when sufficiently cooled. Under near-equilibrium conditions, crystals form, with distinctivelydifferent structures and mechanical properties to the liq-uid phase. When rapidly quenched, on the other hand,a supercooled liquid undergoes glass transition and be-comes an amorphous solid with apparently disorderedstructures. For the glass transitions, two fundamentalquestions remain. The first one is is there a qualita-tive transition point between the liquid and solid phasesduring the glass transition?. Glasses obviously fit ourexperiences with solids. Experimentally, however, thereis no definitive signal for the emergence of rigidity, de-spite more than 10 orders of magnitude increase in vis-cosity during the glass transition. The other question iswhat structural orders, if any, are associated with theunusual dynamical phenomena and the rise of rigidityduring the glass transition?. Many studies attempt toconstruct structural parameters based on local geometryto distinguish slow rigid domains from more mobile fluidregions in glasses [1–5], but have yet to find any universalsignatures. In condensed matters, particularly in solids,the role of the structure is to confine the motion of atoms,thus maintain rigidity. From this point of view, a solidlose its rigidity when the motions of consisting atoms canno longer be adequately constrained. A perfect exampleis the Lindemann criterion for the melting of crystals,which is found to be accurate in almost all crystallinematerials [6, 7]. A crystal melts when the vibrationalfluctuations of atoms reach the order of 0.1 of the latticeconstant. The Lindemann criterion is independent of thesymmetry of the underlying structures of the solids, thusmay be employed to determine the liquid-solid transitionin glass forming materials [8–20]. In meta-stable struc-tures, the vibrational fluctuations of atoms are primarilydetermined by local structures, thus the confinement ex-perienced by individual particles can be employed as a

structural parameter when local geometry is too intri-cate to analyze.

In this Letter, we employ local Debye-Waller factorto measure the local constraints in colloidal liquids andglasses, and investigate its correlations to local dynam-ics during the glass transition. Temporal correlationsbetween particle constraints and local dynamics revealthe emergence of structural relaxation barriers that giverise to finite rigidity in the system, as the tempera-ture decreases. A common Lindemann-like length scaleis identified by comparing the configurational changeswhen the system overcomes the relaxation barriers andstarts behaving like fluids. The rise of rigidity and theonset of glassy dynamics are both shown to coincidewith the percolation of rigid domains identified by theLindemann-like length scale. Dynamical heterogeneityincreases sharply when rigid domains percolate the sys-tem, and then decreases when the system becomes over-whelmingly solid. Our results suggest that a Lindemann-like criterion can be applied in amorphous materials todetermine the transition between liquid and solid states,and the glass transition is the growth and percolation ofrigid domains in supercooled liquids.

The samples consist of binary mixtures of poly-N-isopropylacrylamide (PNIPAM) particles [21, 22] hermet-ically sealed between two coverslips, forming a monolayerof disordered packing. To avoid crystallization, the di-ameter ratio between large and small particles is cho-sen to be 1:1.4, with the number ratio close to 1. ThePNIPAM particles are thermo-sensitive which allows thein-situ tuning of the packing fractions using an objec-tive heater (BiOptechs). PNIPAM spheres are best de-scribed as hard spheres with soft shells [21, 23]. At highpacking fractions, PNIPAM particles are compressible tosome extent, allowing observation of dynamical phenom-ena above the hard sphere jamming transition. The di-

arX

iv:1

710.

0815

4v3

[co

nd-m

at.s

oft]

22

Feb

2019

2

Figure 1. Correlation between soft mode and Debye-Waller factor in jammed packings. a, Correlationbetween αi and Ψn

i as a function of the fraction of the lowest frequency modes n/2N included in Ψni at different

packing fractions. Inset: correlations between α and Ψ30 at different packing fractions. The noise level is about 0.02.b, Real space distribution of cooperatively rearranging regions (white dots) and Ψ30 at φ = 0.850 (colored contours),

normalized by the average value. c, Rank correlations between Ψ2Ni and αi(τ) as a function of τ for different

packing fractions. The noise level is about 0.02. Inset: MSDs at different φ.

ameters of the particles are measured by dynamical lightscattering to be 1 and 1.4 µm at 22 ◦C. The total numberof particles in the field of view is about 3500. To covera wide range of packing fractions, two groups of samplesare seperately prepared. The packing fractions are be-tween 0.890 to 0.850 (jammed solids) for the first group,and between 0.56 and 0.84 (unjammed liquids) for thesecond group. Here we use the 2D jamming packing frac-tion of hard spheres of 0.85 to indicate that no sponta-neous topological rearrangements are observed in samplesof higher packing fractions during the time window avail-able to our experiments [24]. Before data acquisition, thesamples are equilibriated on microscope stage for 3 hours.The particle configurations are recorded by digital videomicroscopy at 30 to 110 frames/s, and the particle trajec-tories are extracted by particle-tracking techniques [25].Combined optical and tracking error of particle fluctua-tions is estimated to be less than 0.01 µm by measuringthe MSD of fixed particles at different packing fractions.For jammed samples, the phonon modes are extractedusing the covariance matrix analysis [26–29]. The co-variance matrix analysis measures the phonon modes ofa shadow system with the same configurations and in-teractions as the colloids in experiment, but without thedamping.

Spatial constraints felt by individual particles can bemeasured by either the lowest energy barrier for dis-placements or positional fluctuations. In jammed solidswith stable configurations, the lowest energy barrier isdirectly related to the soft phonon modes [30]. We em-ploy a soft mode parameter Ψ for individual particles,proposed by Tong and Xu [31] based on equipartition

hypothesis. For particle i, Ψ2Ni =

∑2Nj=1

1ω2

j|~ej,i|2, where

ωj is the vibrational frequency of mode j and ~ej,i isthe polarization vector of particle i in mode j, N is thenumber of particles in a two-dimensional glass. Ψ2N

i isbiased toward the lower frequency modes, as the con-

tributions from high frequency modes to Ψ2Ni diminish

rapidly with frequency. Ψ2Ni removes the ambiguities in

soft modes selections, and can be proven to be statis-tically proportional to the single particle Debye-Wallerfactor αi in meta-stable glasses (See supplementary forderivations [29]). αi = 〈[~ri(t) − ~ri(0)]2〉, where ~ri(t) isthe position of particle i at time t, and 〈.〉 denotes thetrajectory average [31, 32]. Debye-Waller factor is of-ten employed as a dynamical parameter. On short timescales when topological rearrangement is infrequent, lo-cal Debye-Waller factor is primarily determined by localstructures, thus can be employed as a structural param-eter as well. Previous experiments and simulations haveshown that short-time local positional fluctuation is agood predictor of long-time dynamics in the supercooledand glass regime [32, 33].

The high correlations between soft mode Ψ and Debye-Waller factor α are experimentally demonstrated injammed colloidal glasses. Figure 1a plots the Spearman’srank correlation between Ψn

i and αi as a function of thefraction of the lowest frequency modes n

2N included injammed colloidal glasses. The correlation to local dy-namics comes predominantly from the lowest frequencymodes, as the bottom 0.5% of modes (∼ 30 for our sys-tem) achieve a correlation over 0.8. The inset of Fig-ure 1a plots the correlation between Ψ30

i and αi at differ-ent packing fractions, which shows that in jammed solids,positional fluctuations of inidividual particles can be welldecribed by a handful of soft modes. Figure 1b shows thespatial distribution of cooperatively rearranging regions(CRRs) composed of the top 10% fastest particles (whitecircles) [34] and Ψ30

i (colored contours). It is clear thatregions with higher concentration of soft modes are spa-tially correlated with fast local dynamics.

In jammed glasses, soft modes can be accounted for byshort-time fluctuations of particle positions. Figure 1cplots the correlation between αi(τ) and Ψ2N

i as a function

3

Figure 2. Structure-dynamics correlation duringglass transition. a, Measured mean square

displacements at different packing fractions. Dashedline indicates free diffusive motions. b, Spearman’s rank

correlation between αi(τmax) and D2min(∆t) as a

function of ∆t. The vertical arrows indicate the ∆tact

when the correlations start to decay. Inset: the timesequence for measuring α(τ) and D2

min(∆t). c, φdependence of the activation time ∆tact, and the α andβ relaxation time. The dashed line indicates the onsetof glassy dynamics. d, MSDs dependence of correlationCαi(τmax)−D2

min(∆t) for each packing fraction. Verticaldotted line indicates the Lindemann criterion. The

noise level is about 0.02.

of the time window τ in which αi is measured. The cor-relation increases rapidly for small τ values and reaches∼ 0.8 at ∼ 1s, within the β-relaxation time scale (∼ 10s)defined by the middle of the plateau in the log-log plotof the mean square displacements (Figure 1c, inset) [32].The high correlations between short-time αi(τ) and Ψ2N

i

suggest that the local structures can be adequately ex-plored at relatively short periods of time. Further in-creasing of τ only slightly improves the correlation to softmodes. For comparison, it requires more than 1000s ofvideo microscopy measurements to properly extract thevibrational modes from the same jammed colloidal sam-ples using covariance matrix analysis [27]. Thus short-time particle Debye-Waller factor can be employed as aneffective soft mode parameter in colloidal systems belowjamming [35], where direct measurements of spatial dis-tribution of soft modes are difficult.

We now apply αi(τ) in unjammed colloidal liquids tomeasure local mechanical constraints. The MSDs of theunjammed samples are plotted in Figure 2a. As theDebye-Waller factor in unjammed fluids may vary withtime, αi(τ) is no longer time averaged, instead it is calcu-lated for each segment of trajectory in a time window ofτ . To identify the relevant time scales over which localstructures have the most influence over future dynam-

ics in liquids, we measure the temporal correlations be-tween αi(τ) and local dynamics measured by non-affinedisplacement D2

min(∆t) [36, 37] after the preceding struc-tures are measured. D2(t1, t2) =

∑n

∑i

[rin,t2 − ri0,t2 −∑j

(δij + εij)× (rjn,t1 − rj0,t1

)]2 , where rin,t is the ith (x or

y) component of the position of the nth particle at timet, and the δij+εij that minimize D2 are calculated basedon rin,t. D

2min measures the particle level nonaffine strain,

i.e., the minimum mean square difference between actualrelative displacements of particle to its neighbors and therelative displacements that they would have if they werein a region of uniform strain. Correlations between αi(τ)and D2

min(∆t) depend on both the window τ in whichstructural information is collected, and the timescale ofthe dynamics after αi is measured, ∆t. We choose theτ = τmax that yields the highest correlations to D2

min [29].τmax is thus the proper time scale to identify structuresthat have the highest predictability for dynamics in liq-uids; and it naturally emerges from correlation measure-ments. For observation window shorter than τmax, insuf-ficient structural information is collected, and for muchlonger time windows, relevant information will eventu-ally be lost in structural relaxations. In our experiments,τmax is found to be in the vicinity of β-relaxation timeτβ [29], consistent with our results from jammed solids.The β-relaxation time and α-relaxation time of the sam-ples are extracted by fitting the intermediate scatteringfunction with a two-step stretched exponential function(the Kohlrausch-Williams-Watts function) [29, 38, 39].For liquids with only one-step relaxations, the fitting ofthe function yields two nearly identical relaxation times.

Figure 2b plots the correlation between αi(τmax) andD2

min(∆t) as a function of ∆t. The correlations are av-eraged over all available trajectories. At low packingfractions, the correlation between αi(τmax) and local dy-namics is low, and decays almost immediately after theαi(τmax) is measured. This short memory in dynamicsreflects a nearly flat potential energy landscape wherestructural relaxations are facilitated by free diffusion andcollisions between particles. The energy landscape be-comes more rugged as the packing fraction increases, andan activation mechanism begins to emerge [40]. At higherpacking fractions, the correlation between αi(τmax) andD2

min first increases with ∆t then decreases after reach-ing a peak value at ∆tact. This delayed correlation peakbetween local constraints and structural relaxations sig-nifies the emergence of rearranging barriers, hence finiterigidity of the system, with ∆tact being the average timerequired for thermal fluctuations to overcome the barriersfor structural relaxations. When the packing fraction isfurther increased, the energy barrier also increases, withhigher peak correlation values.

The rise of the relaxation barriers coincides with theseparation of α- and β-relaxation time scales in liq-

4

Figure 3. Structure evolution during glass transition. a-f, Spatial distribution of αi(τmax) at different packingfractions, binarized by the Lindemann criterion. Red colors are fluid regions with αi(τmax) larger than the

Lindemann criterion; blue colors are rigid regions with αi(τmax) below the Lindemann criterion. g, Average αi(τmax)and the percolation of rigid regions during glass transition. Left axis: average αi(τmax) normalized by Lindemann

criterion, as a function of φ. The dashed line indicates the onset of glassy dynamics shown in Figure 3c . The blacksquares (data 1) are measured from the same dataset as in Figure 1 and Figure 2. To extend the range of the plot,we include measurements from an additional dataset (data 2, red circles). Error bars represent standard deviations.

Right axis: The probability of rigid regions percolating the field of view(blue triangles). The probability iscalculated as the fraction of the configurations with rigid regions percolating the field of view in all measured

configurations. h, Peak value of dynamical susceptibility, χ∗4 as a function of φ.

uids [41]. Figure 2c plots the measured τα , τβ and ∆tact

in unjammed colloidal samples. The increase of the relax-ation time is modest near the jamming point compared tostandard hard sphere systems [42, 43], due to the softnessof PNIPAM spheres. Below φ = 0.69, ∆tact is short, andthe τα and τβ are close. Without obvious peaks, ∆tact ischosen to be the the point where the correlation betweenαi(τ) and D2

min starts to decay, as indicated by verticalarrows in Figure 2b. Around φ = 0.69 where a delayedcorrelation peak appears, the α- and β-relaxation timesbegin to separate. ∆tact becomes significantly larger thanτβ when the packing fraction is further increased. As theαi(τ) is measured on the time scale of τmax (close to τβ),a ∆tact greater than τβ allows the prediction of long-timedynamics with short-time structural information.

Temporally, local dynamics in liquids begin to decouplefrom earlier structures after ∆tact. An interesting ques-tion is that do the average positional fluctuations of theparticles reach a common length scale when the systembegins to behave like a fluid, as in the case of the meltingof crystals. In Figure 2d, we replot the Cαi−D2

minas a

function of system MSDs. For all the packing fractions,the correlation begins to decay around 20% of the smallparticle diameter d indicated by the dashed line, close tothe Lindemann criterion for the melting of crystals [6],despite orders of magnitude differences in relaxation timescales between these liquid samples. We can thus define

L = 0.2d as the equivalent melting criterion for glasses,and generalize the Lindemann criterion from the meltingof crystals to the transition between solid and fluid phasesin amorphous materials [7–20] where the dichotomy be-tween solid and fluid phases has been ambiguous. For agiven time window, structures that evolve less than the Lare considered solid-like or rigid, while structures evolvemore than the L are considered fluid-like.

Before applying the Lindemann-like criterion locallyto identify rigid or fluidic domains, a proper observationtime window needs to be determined. In the original Lin-demann theory for crystals, the vibrational fluctuationsof atoms around equilibrium positions are considered.For glasses, atoms can be considered primarily vibratingin cages on the β-relaxation time scale. However, insteadof arbitrarily imposing the β-relaxation time, we employthe τmax, which naturally emerges as the time scale mostpertinent to future dynamics from inter-correlation mea-surement, as the observation window for the identifica-tion of rigid regions. Independent measurements confirmthat the τmax in different samples are very close to themeasured β-relaxation times [29]. Using the time win-dow of τmax, we identify solid-like domains in unjammedsamples whose αi(τmax) are below the Lindemann crite-rion, and fluid regions with higher αi(τmax) during theglass transition. Figure 3a-f plot the snapshots of spatialdistribution of αi(τmax) at different packing fractions, bi-

5

narized by the Lindemann criterion. A bond percolationbased on the particle positions is used after we clusterrigid particles from the nearest neighbors which are deter-mined from the first minimum of the radial distributionfunction. At low packing fractions, the system is mostlyfluid-like (red color) with small pockets of solid-like re-gions (blue color). The rigid regions grow with the pack-ing fraction and begin to percolate the system aroundφ = 0.69 until complete solidification near the jammingpoint. Key features of percolation phase transition arerecovered by analyzing the distributions of the size andshape of the solid-like clusters [29]. The percolation prob-ability of rigid regions and the averaged αi(τmax) of thesystem shows a sharp transition around φ = 0.69, asplotted in Figure 3g.

The growth and percolation of the rigid regions in cool-ing liquids provide a microscopic origin for the onset ofglassy dynamics shown in Figure 2c and the dynamicalheterogeneity. At low packing fractions (high tempera-tures), isolated rigid structures are created and relaxedby a one-step fluctuation-relaxation process. The sizeand the fraction of rigid regions both increase as thesamples are further cooled. At a critical packing frac-tion ( φ = 0.69 in our experiments), the rigid regionsbecome connected and percolate the system [44, 45]. Be-fore the percolation, isolated rigid domains exist in theliquid. However, unconnected rigid clusters cannot ren-der the whole system rigid, as they are simply floatingin a continuous phase of flowing liquid. Only after thepercolation, the ability of the spanning network of rigiddomains to resist small stresses gives rise to finite rigidityof the whole system [44, 45]. For the relaxation dynam-ics, before percolation, the rigid domains are formed andrelaxed locally through fluctuations in the liquid, witha single relaxation time. After the percolation, whilethe liquid relaxation process remains in the liquid phase,the relaxation of the system-wide rigid network is muchharder than isolated rigid clusters, which results in amuch longer relaxation time, namely, the α-relaxationtime. The percolating rigid network also impedes longdistance diffusions of particles. Under spatial confine-ment, particles are forced to rearrange locally throughcooperative motions, or β-relaxation [34, 46, 47]. The de-coupling of relaxation times signals the transition fromlocal relaxation process to a correlated relaxation pro-cess [48, 49]. Dynamical heterogeneity naturally emergesfrom the competition between these two different relax-ation mechanisms [50]. The peak of the dynamical sus-ceptibility χ∗

4 first increases around φ = 0.69 and thendecreases near the jamming point (φj ∼ 0.85) when thewhole system becomes homogeneously rigid [51], as plot-ted in Figure 3h (for the measurements of χ∗

4, see thesupplementary materials [29, 52, 53]).

In summary, by measuring the local constraintsin colloidal liquids and glasses, we directly observethe emergence and growth of structure-dynamics cor-

relations in supercooled liquids, which depend on aLindemann-like length scale in configurational changes.The glass transition is then shown to be the growth andpercolation of the rigid regions in supercooled liquids,which can be employed to explain the slowing-downand the dynamical heterogeneity [50, 54]. Although ourresults are obtained from a quasi-2D hard sphere col-loidal system, the method to identify solid-like regions influids can be easily generalized to other glassy systems.Following the melting analogy, the rigid clusters in glasstransition are similar to the crystalline nuclei duringcrystallization. But unlike the nuclei that span thesystem by growing from boundaries, the rigid clustersgain stability by forming a percolating network acrossthe system. These clusters are also natural candidatesfor low-entropy droplets in random first-order transitiontheories for their slower dynamics [9]. We thus speculatethe percolation of rigid domains during glass transitioncan also be observed in 3D glasses [43, 55–58] or insystems with different interactions, while the specificpath leading to the percolation or the evolution of theconnected rigid network after it may be different, whichwill be an interesting topic for future simulation orexperimental studies. Our results are strong evidencethat local constraints are a useful parameter to connectstructure to dynamics in glassy systems compared topurely geometric or topological metrics. However, thisdiscovery does not render the geometric structures irrel-evant. It is obvious that the spatial constraints in glassesdepend sensitively on local configurations, althoughspecific dependence may vary greatly from system tosystem. It is only through the lens of the constraintscan the correlations between structures and dynamics indisordered systems be clearly demonstrated. In addition,local constraints naturally include multi-body effects ofamorphous structures that are difficult to quantify fromanalyzing the geometric structures alone. A direct linkbetween conventional geometric structures and glassydynamics may be established by searching for local andnon-local configurations that contribute the most tolocal constraints in glassy materials [59].

+ X.Y., and H.T. contributed equally to this work.∗ kechen@iphy.ac.cn∗ whw@iphy.ac.cn

[1] F. Spaepen, A microscopic mechanism for steady stateinhomogeneous flow in metallic glasses. Acta Metall. 25,407 (1977).

[2] H. Tanaka, T. Kawasaki, H. Shintani, and K. Watanabe,Critical-like behaviour of glass-forming liquids. NatureMater. 9, 324 (2010).

[3] H. W. Sheng, W. K. Luo, F. M. Alamgir, J. M. Bai, and

6

E. Ma, Atomic packing and short-to-medium-range orderin metallic glasses. Nature 439, 419 (2006).

[4] K. F. Kelton et al. First X-Ray scattering studies on elec-trostatically levitated metallic liquids: demonstrated in-fluence of local icosahedral order on the nucleation bar-rier. Phys. Rev. Lett. 90, 195504 (2003).

[5] Y.-C. Hu, F.-X. Li, M.-Z. Li, H.-Y. Bai, and W.-H.Wang, Five-fold symmetry as indicator of dynamic ar-rest in metallic glass-forming liquids. Nat. Commun. 6,8310 (2015).

[6] F. A. Lindemann, The calculation of molecular Eigen-frequencies. Phys. Z. 11, 609 (1910).

[7] S. Alexander, Amorphous solids: their structure, latticedynamics and elasticity. Phys. Rep. 296, 65 (1998).

[8] H. B. Yu, R. Richert, R. Maaβ, and K. Samwer, UnifiedCriterion for Temperature-Induced and Strain-DrivenGlass Transitions in Metallic Glass. Phys. Rev. Lett. 115,135701 (2015).

[9] X. Xia and P. G. Wolynes, Fragilities of liquids predictedfrom the random first order transition theory of glasses.Proc. Natl. Acad. Sci. U.S.A. 97, 2990 (2000).

[10] C. Angell, Formation of Glasses from Liquids andBiopolymers. Science 267, 1924 (1995).

[11] W. Gotze, Recent tests of the mode-coupling theory forglassy dynamics. J. Phys.: Condens. Matter 11, A1(1999).

[12] R. Hall and P. Wolynes, The aperiodic crystal pictureand free energy barriers in glasses. J. Chem. Phys. 86,2943 (1987).

[13] Stillinger, A topographic view of supercooled liquids andglass formation. Science 267, 1935 (1995).

[14] R. Berry and B. Smirnov, Phase transitions and adjacentphenomena in simple atomic systems. Phys. Usp. 48, 345(2005).

[15] F. Starr et al., What do we learn from the local geometryof glass-forming liquids? Phys. Rev. Lett. 89, 125501(2002).

[16] V. Novikov and A. Sokolove, Universality of the dynamiccrossover in glass-forming liquids: A magic relaxationtime. Phys. Rev. E 67, 031507 (2003).

[17] V. Novikov et al., Strong and fragile liquids in percola-tion approach to the glass transition. Europhys. Lett. 35289(1996).

[18] J. Dyre, Colloquium: The glass transition and elasticmodels of glass-forming liquids. Rev. Mod. Phys. 78, 953(2006).

[19] J. Onuchic, Z. Luthey-Schulten and P. Wolynes, THE-ORY OF PROTEIN FOLDING: The Energy LandscapePerspective. Annu. Rev. Phys. Chem. 48, 545 (1997).

[20] L. Larini, Ottochian, C. De Michelle and D. Leporinin,Universal scaling between structural relaxation and vi-brational dynamics in glass-forming liquids and poly-mers. Nat. Phys. 4 42 (2008).

[21] P. J. Yunker et al. Physics in ordered and disordered col-loidal matter composed of poly(N-isopropyl acrylamide)microgel particles. Rep. Prog. Phys. 77, 056601 (2014).

[22] K. Chen et al. Phonons in two-dimensional soft colloidalcrystals. Phys. Rev. E 88, 022315 (2013).

[23] Y. Han, N. Y. Ha, A. M. Alsayed, and A. G. Yodh, Phys.Rev. E 77, 041406 (2008).

[24] Z. Zhang, N. Xu, D. T. N. Chen, P. Yunker, A. M. Al-sayed, K. B. Aptowicz, P. Habdas, A. J. Liu, S. R. Nagel,and A. G. Yodh, Nature 459, 230 (2009).

[25] J. C. Crocker and D. G. Grier, Methods of digital video

microscopy for colloidal studies J. Colloid Interface Sci.179, 298 (1996).

[26] S. Henkes, C. Brito, and O. Dauchot, Extracting vibra-tional modes from fluctuations: a pedagogical discussion.Soft Matter 8, 6092 (2012).

[27] K. Chen et al. Low-Frequency Vibrations of Soft Col-loidal Glasses. Phys. Rev. Lett. 105, 025501 (2010).

[28] K. Chen et al. Measurement of Correlations betweenLow-Frequency Vibrational Modes and Particle Rear-rangements in Quasi-Two-Dimensional Colloidal Glasses.Phys. Rev. Lett. 107, 108301 (2011).

[29] See Supplemental Material at http://link.aps.org/ sup-plemental/ for a discussion of additional experimentaldetails.

[30] N. Xu, V. Vitelli, A. J. Liu, and S. R. Nagel, Anharmonicand quasi-localized vibrations in jammed solids - Modesfor mechanical failure. Europhys. Lett. 90, 56001 (2010).

[31] H. Tong and N. Xu, Order parameter for structural het-erogeneity in disordered solids. Phys. Rev. E 90, 010401(2014).

[32] A. Widmer-Cooper and P. Harrowell, Predicting theLong-Time Dynamic Heterogeneity in a Supercooled Liq-uid on the Basis of Short-Time Heterogeneities. Phys.Rev. Lett. 96, 185701 (2006).

[33] R. Pastore, G. Pesce, A. Sasso, and M. Pica Ciamarra,Cage Size and Jump Precursors in Glass-Forming Liq-uids: Experiment and Simulations. J. Phys. Chem. Lett.8, 1562 (2017).

[34] G. Adam and J. H. Gibbs, On the Temperature De-pendence of Cooperative Relaxation Properties in Glass-Forming Liquids. J. Chem. Phys. 43, 139 (1965).

[35] R. L. Jack, A. J. Dunleavy, and C. P. Royall,Information-Theoretic Measurements of Coupling be-tween Structure and Dynamics in Glass Formers. Phys.Rev. Lett. 113, 095703 (2014).

[36] X. Yang, R. Liu, M. Yang, W.-H. Wang, and K.Chen, Structures of local rearrangements in soft colloidalglasses. Phys. Rev. Lett. 116, 238003 (2016).

[37] M. L. Falk, and J. S. Langer, Dynamics of viscoplasticdeformation in amorphous solids. Phys. Rev. E 57, 7192(1998).

[38] J. Mattsson et al. Soft colloids make strong glasses. Na-ture 462, 83 (2009).

[39] T. Kawasaki and H. Tanaka, Structural evolution in theaging process of supercooled colloidal liquids. Phys. Rev.E 89, 062315 (2014).

[40] V. Lubchenko, Theory of the structural glass transi-tion: a pedagogical review. Advances in Physics 64, 283(2015).

[41] P. G. Debenedetti and F. H. Stillinger, Supercooled liq-uids and the glass transition. Nature 410, 259 (2001).

[42] R. Pastore, G. Pesce, and M. Caggioni, Differential Vari-ance Analysis: a direct method to quantify and visualizedynamic heterogeneities. Sci. Rep. 7, 43496 (2017).

[43] S. Vivek et al., Long-wavelength fluctuations and theglass transition in two dimensions and three dimensions.Proc. Natl. Acad. Sci. U.S.A. 114, 1850 (2017).

[44] M. H. Cohen and G. S. Grest, Liquid-glass transition, afree-volume approach. Phys. Rev. B 20, 1077 (1979).

[45] D. J. Jacobs and M. F. Thorpe, Generic Rigidity Per-colation: The Pebble Game. Phys. Rev. Lett. 75, 4051(1995).

[46] S. Albert et al., Fifth-order susceptibility unveils growthof thermodynamic amorphous order in glass-formers. Sci-

7

ence 352, 1308 (2016).[47] C. Crauste-Thibierge, C. Brun, F. Ladieu, D. LHote, G.

Biroli, and J.-P. Bouchaud, Evidence of growing spatialcorrelations at the glass transition from nonlinear re-sponse experiments. Phys. Rev. Lett. 104, 165703 (2010).

[48] R. Pastore, A. Coniglio, A. de Candia, A. Fierro, andM. Pica Ciamarra, Cage-jump motion reveals universaldynamics and non-universal structural features in glassforming liquids. J. Stat. Mech. 5, 054050 (2016).

[49] R. Pastore, G. Pesce, A. Sasso, and M. P. Ciamarra,Many facets of intermittent dynamics in colloidal andmolecular glasses. Colloids Surf. A 532, 87 (2017).

[50] D. Long, and F. Lequeux, Heterogeneous dynamics atthe glass transition in van der Waals liquids, in the bulkand in thin films. Eur. Phys. J. E 4, 371 (2001).

[51] P. Ballesta, A. Duri, and L. Cipelletti, Unexpected dropof dynamical heterogeneities in colloidal suspensions ap-proaching the jamming transition. Nature Phys. 4, 550(2008).

[52] R. Candelier, O. Dauchot, and G. Biroli, Building Blocksof Dynamical Heterogeneities in Dense Granular Media.Phys. Rev. Lett. 102, 088001 (2009).

[53] Z. Zhang, P. J. Yunker, P. Habdas, and A. G. Yodh, Co-operative Rearrangement Regions and Dynamical Het-erogeneities in Colloidal Glasses with Attractive VersusRepulsive Interactions. Phys. Rev. Lett. 107, 208303(2011).

[54] G. Biroli, J.-P. Bouchaud, K. Miyazaki, and D. R. Reich-

man, Inhomogeneous Mode-Coupling Theory and Grow-ing Dynamic Length in Supercooled Liquids. Phys. Rev.Lett. 97, 195701 (2006).

[55] P. Harrowell, Nonlinear physics: Glass transitions inplane view. Nature Phys. 2, 157 (2006).

[56] B. Illing et al., MerminWagner fluctuations in 2D amor-phous solids. Proc. Natl. Acad. Sci. U.S.A. 114, 1856(2017).

[57] H. Shiba et al.,Unveiling Dimensionality Dependence ofGlassy Dynamics: 2D Infinite Fluctuation Eclipses Inher-ent Structural Relaxation. Phys. Rev. Lett. 117, 245701(2016).

[58] H. Tong and H. Tanaka, Revealing Hidden StructuralOrder Controlling Both Fast and Slow Glassy Dynamicsin Supercooled Liquids. Phys. Rev. X. 8, 011041 (2018).

[59] S. S. Schoenholz, E. D. Cubuk, D. M. Sussman, E. Kaxi-ras, and A. J. Liu, A structural approach to relaxationin glassy liquids. Nature Phys. 12, 469 (2016).

Acknowledgements We thank Walter Kob, Pe-ter Harrowell, Rui Liu, Mingcheng Yang, ChenhongWang, and Maozhi Li for helpful discussions. Thiswork was supported by the MOST 973 Program (No.2015CB856800). K. C. also acknowledges the supportfrom the NSFC (No. 11474327).