Fast dynamics of water droplets freezing from the outside-in Sander Wildeman,1 Sebastian Sterl,1 Chao Sun,2,1 and Detlef Lohse1,3 1Physics of Fluids group, University of Twente, 7500 AE Enschede, The Netherlands 2Centre for Combustion Energy and Department of Thermal Engineering,Tsinghua University, 100084 Beijing, China 3Max Planck Institute for Dynamics and Self-Organization, 37077, Go¨ttingen, Germany (Dated: January 25, 2017) Adropofwaterthatfreezesfromtheoutside-inpresentsanintriguingproblem: theexpansionof wateruponfreezingisincompatiblewiththeself-confinementbyarigidiceshell. Usinghigh-speed imaging we show that this conundrum is resolved through an intermittent fracturing of the brittle ice shell and cavitation in the enclosed liquid, culminating in an explosion of the partially frozen droplet. We propose a basic model to elucidate the interplay between a steady build-up of stresses 7 and their fast release. The model reveals that for millimetric droplets the fragment velocities upon 1 explosion are independent of the droplet size and only depend on material properties (such as the 0 tensile stress of the ice and the bulk modulus of water). For small (sub-millimetric) droplets, on 2 the other hand, surface tension starts to play a role. In this regime we predict that water droplets n with radii below 50µm are unlikely to explode at all. We expect our findings to be relevant in the a modeling of freezing cloud and rain droplets. J 4 2 In mid 17th century, some peculiar tear shaped glass the larger ice particles in clouds [14–16] and of hailstone objectswerebroughttotheattentionoftheRoyalSociety embryos [17, 18]. It is believed that the fragmentation ] n via Prince Rupert of the Rhine [1]. The tears, made by of freezing water droplets can play a role in the rapid y dripping molten glass into cold water, were able to with- glaciation of supercooled clouds and the development of d stand the blow of a hammer when hit on their spherical precipitation [15, 19, 20]. Ice drop bursting has also re- - u head, while they abruptly disintegrated into fine pieces cently been observed for condensation droplets formed l when their delicate tip was tampered with [2]. It was on chilled super-hydrophobic surfaces, where the thrown f . foundthatthetougheningofthesphericalheadrelieson outicefragmentscanspeedupfrostformation[21]. Sim- s c thefactthatglasscontractsasitcools. Inthecoldwater ilarly, solidification expansion and fracturing can affect i theouterlayerofeachliquidglasstearquicklycoolsand the production of small silicon spheres from a melt [22]. s y solidifies, encapsulatingacoreofmoltenglass. Thiscore Although it is clear that the fracturing of solidifying h pulls inwards on the already solid shell as it continues drops plays an important role in many natural and in- p to cool. Nowadays this phenomenon is used to produce dustrial processes, surprisingly little is known about the [ toughened glass. (fast) dynamics that leads up to the final cleaved parti- 1 cle. WiththeexceptionofpreliminaryfootageofLeisner v What is perhaps less well known is what happens in et al. [23], previous observations have been made “after 8 the complementary situation, when, instead of contract- 1 the fact”, or with a too low spatial/temporal resolution ing,thematerialofthedropsexpands uponsolidification. 8 to observe the important details. Qualitatively, it has Onlyahandfulofmaterialshavethisspecialproperty,of 6 been found that the success rate of a freezing drop to which water and silicon are the most common. From 0 explode increases (1) with the strength of the ice shell, . daily experience we know that water freezing up inside a 1 which is influenced, for example, by the presence of dis- closed container can exert extreme pressures on its con- 0 solvedgasesorionsduringthefreezingprocess[6,7,10], 7 tainer walls, as is exemplified by the fracturing of rocks (2) with the degree of radial symmetry of cooling [8– 1 by freezing water inclusions [3] and by the playful “ice 10, 13] and (3) with the size of the droplet [7, 9–11, 24]. : bomb”, in which freezing water bursts out of a thick- v However,asfarasweknow,noattemptshavebeenmade i walled metal flask [4]. For unenclosed water drops, it X sofartomodeltheintricatedynamics(givenquantitative has been found that the combination of expansion and measures of these three factors). r geometrical confinement by surface tension leads to the a formation of a singular tip in the final stage of solidi- In this Letter we present high speed video footage of fication [5]. However, in these experiments the spher- millimetric water droplets freezing from the outside-in, ical symmetry was purposefully broken by cooling the capturing both the intermittent crack formation and the dropletsonlyfromoneside. Symmetric,radiallyinwards, final explosion in detail. We explain the main observa- freezing of water droplets (as in Rupert’s experiment) tionsbymodelingboththeslowfreezingandstressbuild- has been mainly studied in the context of ice forma- up, and the fast dynamics after stress release. Finally, tion in (rain) clouds [6–13]. In these studies it has been we discuss the implications of the models for smaller observed that the self-confinement by a rigid ice shell droplets, such as those found in clouds. can cause such drops to ‘auto-detonate’ and fly apart In each experiment a millimetric droplet of clean, de- into pieces. This can explain (some of) the shapes of gassed water was placed on a glass slide in the center 2 A C 0 t (s) 4000.6 A t = 0 B t = 25 ms C t = 63 ms vacuumP chamber C)84 P Pa) ice nucleation ice shell glass slidiceed plate (°T-40 T Tlv Til Tiv 0.3 (mP -8 0 B D ice spicule supercooled drop AgI (s) ice shell D t = 248.8 ms E t = 259.8 ms F t = 260.2 ms cavities healed T crack cavitation soot substrate ~3.5 m/s FIG. 1. Experimental method. (Suppl. Video 1) A-B, A G t = 1008.4 ms I t = 2190.8 ms dropletofliquidwaterrestsonasuper-hydrophobicsootsur- ice splinters faceinthecenterofasmallvacuumchamber. Hereitbecomes supercooled by evaporative cooling. An iced floor plate con- crack trols the final supercooling of the drop by providing a vapor buffer. C,TraceofchamberpressureP anddroplettempera- ture T (measured by inserting a thermocouple in the drop in cavitation ~1.5 m/s onecase). Themomentoficenucleationandthethreephase equilibrium temperatures involved (ice-liquid, ice-vapor, and H t = 2188.8 ms liquid-vapor)areindicated. D,Oncethedroplethasreached a steady temperature, ice nucleation is induced by touching thedropwithatipofsilveriodide(AgI).(Scalebars: 1mm) of a small vacuum chamber (Fig. 1A). Glass windows in the front and back sides of the chamber allowed for anunobstructedviewofthedroplet(Fig.1B).Toensure thatthedropletkeptitssphericalshape,theglasssurface FIG. 2. High speed recording of an inwards freezing water was made super-hydrophobic by covering it with a layer droplet. (Suppl. Video 2) A, Ice nucleation at the drop sur- face B, Formation of an ice shell, causing the droplet to roll of candle soot [25]. Deposition of a fresh layer of soot over slightly. C, An ice spicule is pushed out from the shell. before each trial effectively isolated the droplet from any D,Spiculeceasestogrow,theiceshellcracksandvaporcav- unintended ice nuclei on the substrate (the soot itself re- ities appear below surface (dashed circles). E, Cavities have mainedinactiveasnucleusdowntodroplettemperatures completely healed. F, Spicule tip explodes with splinter ve- of approximately 15◦C). When the chamber is evacu- locities of 3.5m/s. G, More cracks and cavities appear, ac- − ated, a water droplet here rapidly cools to sub-freezing companied by thrown out ice splinters. H, Droplet appear- temperatures by evaporative cooling and, if nothing is ance just before the final explosion. I, Final explosion with fragment velocities of 1.5m/s. (Scale bar: 1 mm) done to prevent it, would spontaneously freeze [26, 27]. However, unrestricted evaporative cooling gives very lit- tlecontroloverthetemperatureprofileinthedropduring the experiment. To control the droplet temperature, the inducingicenucleationbytouchingthedropletwithatip floor of the vacuum chamber was covered with a layer of of silver iodide (Fig. 1D). Silver iodide is one of the few ice held at a pre-set temperature of T by a cooling cir- materials (besides ice) which readily induces ice nucle- iv cuit embedded in the floor of the chamber. As shown ationatlowsupercooling. Thisisattributedtothesalt’s in Fig. 1C, this ice layer provides a buffer which (by ice-like crystal structure [28]. After nucleation, a small sublimation) keeps the water vapor pressure surround- fractionofthewaterrapidlyfreezes,formingthiniceden- ing the droplet constant at the vapor pressure P (T ) of drites which penetrate the droplet and give it a hazy v iv theicelayer. Thesteadystatetemperatureofthesuper- appearance [7, 8]. Correspondingly, the droplet temper- cooled droplet is in turn determined by the equilibrium atureshootsuptotheice-liquidcoexistencetemperature between this buffer pressure and the vapor pressure of T =0 C(cf. Fig.1C).Oftenthisinitialreleaseoflatent il ◦ liquid water, which occurs at a slightly lower tempera- heatisaccompaniedbyaslightroll-overordisplacement ture T . Once a constant pressure of 340 10Pa (cor- of the droplet, possibly caused by a temporary (asym- lv ± responding to an ice-vapor equilibrium temperature of metric) increase in vapor pressure at the drop’s surface T = 7.0 0.3 C) was reached in the chamber, we let [27]. Upon further evaporative cooling a solid ice shell iv ◦ − ± each droplet equilibrate for at least three minutes before forms which slowly thickens (see Suppl. Mat. (SM) I § 3 foradetailedcalculationoftheslowthickeningoftheice A B ~1 m/s shell with time). Figure 2 shows in detail what happens after a super- cooled drop comes in contact with an ice nucleus. In a few hundredths of a second the droplet is completely en- C D σθθ < σc E σθθ = σc capsulated by a solid ice shell (Fig. 2B). Shortly after, a ice Ri thinicespiculeisseentobepushedoutfromthedroplet, liquid Ro Pi σθθ P(t) presumably at a weak spot in the shell (Fig. 2C). This x(t) growing ice spicule is a first sign of the increase in pres- Til sureinsidethedropletduetotheexpansionofthethick- F Tiv G H ening ice shell. The spicule usually continues to grow to 1 2 m athleenspgitchuloefsatopppsrogxriomwaitnegl,ypornesesudrreopbluetilddsiaumpeategra.in,Otnhcies / RRi o σ/ P c0 Ri / Ro = 0.8 2 σ/ (/ K)e c0.1 Ro= 0.05 m 0.2 mm timeleadingtoasuddenfracturingoftheshell(Fig.2D). 0.5 1 mm 0 -2 0.1 0 During crack formation, vapor cavities are clearly visible 0 1/6 0 π 2π 0 0.5 1 t / τf ωt Ri / Ro below the surface of the ice (dashed circles), indicating a suddenchangefromahighinternalpressuretoapressure low enough for cavitation to occur. As the liquid wa- FIG. 3. Basic picture of the dynamics of an inwards freezing ter in the inclusion continues to freeze and expand these waterdrop. A,Highspeedrecordingofadropletburstinginto twoequalhalves. (Scalebar: 1mm)B,Crosssectionalviews cavities and cracks gradually heal (Fig. 2E). A droplet of the halves of a burst droplet, showing a clear core region generallyundergoesmultipleofsuchfracturingandheal- surroundedbyastrongiceshell. Ahealedcrackemergesfrom ing cycles during the whole freezing process. Often some thecoreandrunsoverthelengthofthedroplet. (Scalebars: smallicesplintersarethrownoffintheseenergeticevents 1 mm) C-E, Cartoon of the freezing and bursting process. (Fig. 2G). Figure 2F shows another interesting source of C,AtemperaturedifferenceT T overtheiceshellmakes il iv − ice splinters: after the first healing phase the tip of the the freezing front Ri gradually proceed inward (see F). D, spicule explodes with fragment velocities of a few meters Expansionofthefreezingwaterleadstoabuild-upofpressure P intheinclusion,whichinturnleadstoanazimuthalstress per second. Finally, about 2 seconds after nucleation, i σ intheiceshell. E,Oncethestressintheiceshellexceeds also the droplet as a whole explodes (Fig. 2I). The ve- θθ σ the shell cracks open. F, Theoretical time evolution of c locities reached by the larger fragments are of the order the freezing front, scaled by τ = (ρL R2)/(κ∆T), where f m o of 1 m/s. Aside from some variations in the sequence κ is the thermal conductivity of the ice shell and L is the m of intermediate events, each droplet in our set of experi- specificlatentheatofmelting(seeSM I).G,Timeevolution § ments showed qualitatively the same behavior and every of pressure in the inclusion during the opening of a crack droplet finally exploded. for three different ratios Ri/Ro 0.1,0.5,0.8. The pressure ∈ rapidlyvariesfromitspositiveinitialvaluetoanegativevalue Oftentheicedropswereseentosplitintotwoapprox- (tension) of the same magnitude. The peak-to-peak pressure imately equal halves (Fig. 3A). Figure 3B shows high- increases as the relative inclusion size shrinks. H, Explosion resolutionphotographsof thecleavedsurfacesof thetwo energy (solid black curve) per unit volume e=ρv2/2 (scaled matching parts of such a droplet. Both halves display by σ2/K) as a function of R /R . A maximum in energy c i o a clear core region, presumably formed by (partially ex- occurs for R /R 0.57. The dashed red lines indicate the i o ≈ pelled) water that was not yet solid at the time of the minimum surface energy (per unit volume) required to tear explosion. The cross-sections reveal some interesting de- apart the liquid in the inclusion for different droplet sizes. Below R 50µm the required surface energy is higher for tails about the freezing history of the droplet, such as o ≈ all R /R . the partially healed crack emerging from the core and i o running over the whole length of the droplet. This basic picture inspired the quantitative model described next. sion of the ice shell, but the order of magnitude will be As depicted in Fig. 3D, the expansion of the freezing thesame[29]. TheinternalpressureP inturnleadstoa i water against the ice shell results in a build-up of pres- tensile stress σ in the ice shell. This stress is maximal θθ sure P in the liquid inclusion. Because water is practi- i at the inner shell surface, where it is given by [30]: cally incompressible, only a small part of it has to freeze to result in a dramatic increase in the pressure. A quick P 1+2(R /R )3 i i o σ = , (1) calculationusingtheelasticbulkmodulusofliquidwater, θθ 2 1 (R /R )3 K 2.2GPa, and the fractional volume increase asso- (cid:18) − i o (cid:19) ≈ ciated with the phase transition, β 0.09, gives a pres- whereR andR denote,respectively,theinnerandouter i o ≈ sureincreaseof∆P Kβ∆V /V 2MPa,whenonlya radius of the ice shell (see Fig. 3C). For a thin shell of i i i ≈ ≈ fraction ∆V /V of 1% of the inclusion volume V freezes. thickness ∆R R this stress can be approximated as i i i o (cid:28) Inrealitythepressureincreasewouldbesomewhatlower σ P R /(2∆R), while for thick shells σ P /2. In θθ i o θθ i ≈ ≈ due the simultaneous compression and outward expan- our model we will assume that once the tensile stress 4 exceeds a critical value σ associated with the tensile Aftertheshellhalvesreachedtheirmaximumvelocity, c strength of the ice, the brittle shell will crack open. Al- theyovershoottheirequilibriumpositionx . Thiscauses e though the tensile strength of ice does not seem to be the pressure in the inclusion to reverse sign, putting the so widely investigated, values between 0.7 and 3.1 MPa liquid under tension. As shown in Fig. 3G, the max- havebeenreported[31]. Thegeometricamplificationfac- imal liquid tension is small for thin shells, but quickly tor R /∆R for thin shells may explain why in the initial increases as the inclusion size shrinks. This puts a first o stageliquidiseasilysqueezedoutintheformofaspicule. constraint on when an ice drop can explode: R /R has i o What remains to be explained is why the droplet first to be small enough to make the pressure fall below the only shows cracks and vapor cavities, while in a later cavitation threshold of the liquid, which for clean water, stage it completely disintegrates. For this we investi- free of gas pockets, can easily be 100MPa into the neg- gate the dynamics of the droplet just after a crack has ative [32, 33]. Of course, for millimetric droplets, the formed. Consider the ideal situation in which a crack micron sized gap formed by the crack will be an effec- instantaneously completely splits the rigid shell in two tive nucleation site, lowering the nucleation threshold equalhalves(Fig.3E).Initially,theliquidintheinclusion to ∆P 2γ/x 0.1MPa [34], where γ 75mN/m e ≈ ≈ ≈ will still be pressurized, exerting a force πR2P on each is the surface tension of water at 0 C. However, for a i i ◦ half, acting to separate them. If the halves move apart a small cloud droplet of say R 10µm, equation (3) pre- o ∼ distancex(t),thevolumeoftheliquidinclusionincreases dicts a hundred fold thinner gap, and a corresponding by dV (t) πR2x(t) and the pressure decreases accord- higher threshold tension. For such droplets the cavi- i ≈ i ingly: dP (t) = KdV /V = (3/4)Kx(t)/R . With tation threshold would form a serious barrier to explo- i i i i − − this,theequationofmotionforeachdroplethalfofmass sion. A second constraint is obtained if we compare m = (2/3)πρR3 and moving over a distance x(t)/2, be- the total elastic energy released when a crack is formed, o comes: E =2 (mv2/2)=(1/2)V P2/K, to the minimum sur- e × i i π d2x 3 x face energy, Eγ = 2πγRi2, required to completely split 3ρRo3dt2 =πRi2 Pi−K4R , (2) the liquid inclusion after cavitation. As shown in Fig. (cid:18) i(cid:19) 3H, this puts another upper limit on the ratio R /R . i o This equation represents an initially compressed mass- This explains why initially, when the shell is relatively spring system, with the solution x(t) = xe(1 cos(ωt)), thin, droplets crack and cavitate without exploding. In − where SM II we show how the above explosion criteria can be § combinedwiththefreezingmodel(SM I)toestimatethe 3 KRi 4PiRi § ω = ; x = . (3) time to explosion in our experiment. The energy com- 2sρRo3 e 3 K parisoninFig.3Halsoshowsthatfordropletswithouter In this the initial internal pressure Pi can be directly re- radii Ro below approximately 50µm (and σc ≈ 3MPa) latedtothetensilestrengthσ byinvertingequation(1). the energy required to create the new surface area is c For a droplet of outer radius R = 1mm, inner radius always higher than the released elastic energy. In our o R = 0.5mm and with a shell strength of σ = 3MPa, modelthesedropletscannotexplodeatall. Thisisconsis- i c one finds typical values of P = 4.2MPa, ω = 1.6MHz tentwithpreviouslabexperiments[7,9–11,24]inwhich i and x =1.3µm. The short time scale of crack opening, a sharp decrease in the drop bursting probability is ob- e 2π/ω = 4µs, explains why even at the high recording served around this droplet size. Also field observations rates in our experiments ( 10kHz) cracks seem to ap- on natural clouds indicate that there exists such a size pear instantaneously. We∼note that this time is signifi- threshold [15, 19, 35]. cantly shorter than the time ∆t µ∆R2/(x2P ) 60µs To summarize, we have shown that millimetric water ∼ e i ≈ required for liquid water with viscosity µ 1.8mPas to droplets that freeze radially inwards undergo a sequence ≈ bepushedoutintothecracks. Thissupportsourneglect of intermittent fracturing and healing events, culminat- of flow in the above considerations. ing in a final explosion with fragment velocities of the Thevelocityreachedbyeachhalfduringanoscillation order of 1 m/s. By modeling the elastic stresses in the cycle is v =ωx /2. Upon inserting equation (3) and us- ice shell and the fast dynamics following its failure, we e ingequation(1)toexpressP intermsofσ ,onediscovers haveunveiledtheimportantcompetingenergiesinvolved i c that the fragment velocity is independent of the size of in this phenomenon. On the one hand the released elas- the droplet and only depends on the material properties tic energy, E σ2R3/K, drives the fragments apart, e ∼ c and the ratio R /R . A maximum in velocity is attained while on the other hand the surface energy, E γR2, i o γ ∼ for R /R 0.57, for which v 0.5 σ2/Kρ. Again glues them together. For millimetric droplets the elas- i o ≈ m ≈ c assuming σ 3MPa, gives v 1m/s, which is close tic energy dominates, leading to the observed fragment c ≈ m ≈ p to the fragment velocities observed in the experiments velocities of the order of v σ2/Kρ 1m/s. Since ∼ c ∼ (see e.g. Figs. 2I and 3A and also Ref. [4]). We now only material properties appear in this expression, mea- p assesswhetherthisvelocityisalwayshighenoughforthe suring fragment velocities of exploding water drops may droplet to completely split. provide a convenient way to probe the tensile strength 5 of ice under different circumstances (for example in the 131, 639 (2005). presence of dissolved gases or ions). Below a critical ra- [16] R. P. Lawson, S. Woods, and H. Morrison, J. Atmos. dius, R Kγ/σ2 50µm, surface energy is found to Sci. 72, 2429 (2015). ∗ ∼ c ∼ [17] C. A. Knight and N. C. Knight, J. Atmos. Sci. 31, 1174 dominate. Our model predicts that below this size, ice (1974). drop explosions become impossible. This finding may be [18] T. Takahashi and N. Fukuta, J. Atmos. Sci. 45, 3288 important in understanding the behavior of freezing rain (1988). and cloud droplets, which often exist in this critical size [19] R. R. Braham, J. Atmos. Sci 21, 640 (1964). range. Although there are some indications of such a [20] R. F. Chisnell and J. Latham, Q. J. R. Meteorol. Soc. threshold in the current literature, a systematic exper- 100, 296 (1974). imental investigation into this may bear fruit. It will [21] J. B. Boreyko, R. R. Hansen, K. R. Murphy, S. Nath, S. T. Retterer, and C. P. Collier, Sci. Rep. 6, 19131 also be interesting to study in more detail the observed (2016). spicule formation and explosion, and the shedding of ice [22] S. Ueno, H. Kobatake, H. Fukuyama, S. Awaji, and splinters during crack formation. Even in the absence H. Nakajima, Mater. Lett. 63, 602 (2009). of a final explosion these phenomena could play a role [23] L. Leisner, T. Pander, P. Handmann, and A. Kiselev, in propagating ice formation throughout a collection of 14th Conference on Cloud Physics (Am. Meteor. Soc., supercooled droplets (cf. Suppl. Video 3). 2014) https://ams.confex.com/ams/14CLOUD14ATRAD/ We thank D. van der Meer and J. Snoeijer for stim- webprogram/Paper250221.html. ulating discussions. This work was funded through the [24] J.LathamandB.J.Mason,Proc.R.Soc.Lond.A260, 537 (1961). NWO Spinoza programme. [25] X. Deng, L. Mammen, H.-J. Butt, and D. Vollmer, Sci- ence 335, 67 (2012). [26] J. A. Sellberg, C. Huang, T. A. McQueen, N. D. Loh, H.Laksmono,D.Schlesinger,R.G.Sierra,D.Nordlund, C. Y. Hampton, D. Starodub, D. P. DePonte, M. Beye, [1] L. Brodsley, C. Frank, and J. W. Steeds, Notes Rec. R. C. Chen, A. V. Martin, A. Barty, K. T. Wikfeldt, T. M. Soc. 41, 1 (1986). Weiss, C. Caronna, J. Feldkamp, L. B. Skinner, M. M. [2] M. M. Chaudhri, Philos. Mag. Lett. 78, 153 (1998). Seibert, M. Messerschmidt, G. J. Williams, S. Boutet, [3] I.VlahouandM.G.Worster,J.Glaciol.56,271(2010). L.G.M.Pettersson,M.J.Bogan, andA.Nilsson,Nature [4] G. Reich, Phys. Teach. 54, 9 (2016). 510, 381 (2014). [5] A.G.Mar´ın,O.R.Enr´ıquez,P.Brunet,P.Colinet, and [27] T. M. Schutzius, S. Jung, T. Maitra, G. Graeber, J. H. Snoeijer, Phys. Rev. Lett. 113, 054301 (2014). M. Ko¨hme, and D. Poulikakos, Nature 527, 82 (2015). [6] E. J. Langham and B. J. Mason, Proc. R. Soc. Lond. A [28] B. Vonnegut, J. Appl. Phys. 18, 593 (1947). 247, 493 (1958). [29] J. D. Eshelby, Proc. R. Soc. A Math. Phys. Eng. Sci. [7] B. J. Mason and J. Maybank, Q.J.R. Meteorol. Soc. 86, 252, 561 (1959). 176 (1960). [30] L.LandauandE.M.Lifshitz,Theory of Elasticity (Vol- [8] D. A. Johnson and J. Hallett, Q.J.R. Meteorol. Soc. 94, ume 7 of A Course of Theoretical Physics) (Pergamon 468 (1968). Press, 1970). [9] P. V. Hobbs and A. J. Alkezweeny, J. Atmos. Sci. 25, [31] J. J. Petrovic, J. Mater. Sci. 38, 1 (2003). 881 (1968). [32] Q. Zheng, D. J. Durben, G. H. Wolf, and C. A. Angell, [10] J.L.BrownscombeandN.S.C.Thorndike,Nature220, Science 254, 829 (1991). 687 (1968). [33] M. E. M. Azouzi, C. Ramboz, J.-F. Lenain, and [11] I. E. Kuhns, J. Atmos. Sci. 25, 878 (1968). F. Caupin, Nat. Phys. 9, 38 (2013). [12] C. Takahashi and A. Yamashita., J. Meteor. Soc. Japan [34] E.N.Harvey,D.K.Barnes,W.D.McElroy,A.H.White- 48, 369 (1970). ley, D. C. Pease, and K. W. Cooper, J. Cell. Comp. [13] R.J.Kolomeychuk,D.C.McKay, andJ.V.Iribarne,J. Physiol. 24, 1 (1944). Atmos. Sci. 32, 974 (1975). [35] P. V. Hobbs and A. L. Rangno, J. Atmos. Sci. 42, 2523 [14] A. V. Korolev, M. P. Bailey, H. J., and G. A. Isaac, J. (1985). Appl. Meteor. 43, 612 (2004). [15] A. L. Rangno and P. V. Hobbs, Q.J.R. Meteorol. Soc. DRAFT Supplementary Material for Fast dynamics of water droplets freezing from the outside-in Sander Wildeman,1 Sebastian Sterl,1 Chao Sun,2,1 and Detlef Lohse1,3 1Physics of Fluids group, University of Twente, 7500 AE Enschede, The Netherlands 7 1 2Centre for Combustion Energy and Department of Thermal Engineering, 0 2 Tsinghua University, 100084 Beijing, China n 3Max Planck Institute for Dynamics and Self-Organization, 37077, G¨ottingen, Germany a J 4 (Dated: January 25, 2017) 2 ] n y d - u l f . s c i s y h p [ 1 v 8 1 8 6 0 . 1 0 7 1 : v i X r a 1 I.STEFANPROBLEMOFARADIALLYINWARDSFREEZINGWATERDROP To estimate the thickness of the ice shell at a particular time t after nucleation, we solve the heat transfer problem of a water droplet that slowly freezes radially inwards. Consider a spherical droplet of outer radius R which at time t has reached an ice shell of thickness o R R (t), where R (t) is the radius of the liquid inclusion. The quasi-steady heat flow o i i − through a spherical shell held at a temperature difference of ∆T = T T is given by (see i o − e.g. [? ]) R R (t) o i q = 4πk∆T , (S1) s R R (t) o i − where k 2 W/(mK) is the thermal conductivity of the ice shell. In our experiment the ≈ estimation of q is greatly facilitated by the fact that the temperature at both the inner and s outershellsurfaceisdeterminedbyawelldefinedphaseequilibrium, ice-liquidandice-vapor, respectively. The ice-vapor buffer on the bottom of the chamber ensures that the vapor pressure in the chamber stays approximately constant during the whole experiment (even whenthevacuumpumpiskepton)sothat∆T isconstantandequalto∆T = T T 7K. il iv − ≈ This heat flow through the shell sets the rate at which latent heat L 330kJ/kg is m ≈ released as the freezing front advances inwards (dR /dt < 0): i dR q = 4πρL R2 i, (S2) f − m i dt where ρ 1000 kg/m3 is the density of water. For simplicity we neglected the 9% de- ≈ crease in density associated with the liquid to ice phase transition and the slight decrease in outer radius due to evaporation (to keep T = T we have dR /dR = (R /R )2(L /L ) o iv o i i o m s ≈ 0.12(R /R )2, where L 2840kJ/kg is the latent heat of sublimation of ice). Setting i o s ≈ q = q , we find the following differential equation for R (t): s f i dR k∆T R i o R = . (S3) i dt −ρL R R m o i − By introducing R˜ = R /R and t˜= t/τ , with the timescale i i o f ρL R2 τ = m o, (S4) f k∆T this expression can be recast in an universal dimensionless form as: ˜ dR 1 ˜ i R = . (S5) i dt˜ −1 R˜ i − 2 Using separation of variables and integrating once the solution to equation (S5) is readily obtained as: 1 1 1 R˜3 R˜2 + = t˜. (S6) 3 i − 2 i 6 It is clear from this relation that the droplet will be completely frozen (R˜ = 0) at t˜= 1/6. i II. ESTIMATION OF THE TIME TO EXPLOSION The model for the thickening of the ice shell can be combined with the criteria for ex- plosion (as derived in the main text) to obtain an estimate of the time it takes for droplet to explode. A lower bound on the explosion time can be obtained by (numerically) finding the ratio R /R for which the stored elastic energy, E = (1/2)V P2/K, at the moment of i o e i i crack formation exceeds the energy, E = 2πγR2, required to create the fresh liquid-vapor γ i interface (for a given droplet size R ). This minimum shell thickness then serves as input o for the freezing model to obtain the corresponding minimal time to explosion. An upper bound is given by the time t = τ /6 for the droplet to freeze completely. f In Fig. S1 the results of this calculation are compared to experimental explosion times. Although the precise explosion time for each drop is random in nature, it is clear that the droplets did indeed always explode within the predicted time frame. In fact, with a few exceptions the explosions occur before or near the time for which the stored elastic energy reaches an optimum, i.e. when R /R 0.57 (dashed line in Fig. S1). i o ≈ 3 10 experiments d n y s) 20 ms u rg ( 8 o e me er b x. en i p a t 6 p m n 0 u o 50 70 μm i s o 4 l p x e 2 lower bound 0 0 1 2 3 droplet radius (mm) FIG. S1. Explosion time as a function of the outer droplet radius R . The upper solid line shows o the theoretical upper bound given by the total freezing time τ /6. The lower solid line represents f the theoretical lower bound given by the time it takes for the shell to reach a thickness for which sufficient internal pressure can build up. The dashed line indicates the time at which the bursting is expected to be the most energetic. Orange dots represent our experimental observations. The inset shows a zoomed in-region near the origin, highlighting the predicted inhibition of ice drop explosions below a critical radius of about 50 micron (vertical dashed line). 4