Water Mobility in the Interfacial Liquid Layer of Ice/Clay Nanocomposites

Abstract At solid/ice interfaces, a premelting layer is formed at temperatures below the melting point of bulk water. However, the structural and dynamic properties within the premelting layer have been a topic of intense debate. Herein, we determined the translational diffusion coefficient Dt of water in ice/clay nanocomposites serving as model systems for permafrost by quasi‐elastic neutron scattering. Below the bulk melting point, a rapid decrease of Dt is found for charged hydrophilic vermiculite, uncharged hydrophilic kaolin, and more hydrophobic talc, reaching plateau values below −4 °C. At this temperature, Dt in the premelting layer is reduced up to a factor of two compared to supercooled bulk water. Adjacent to charged vermiculite the lowest water mobility was observed, followed by kaolin and the more hydrophobic talc. Results are explained by the intermolecular water interactions with different clay surfaces and interfacial segregation of the low‐density liquid water (LDL) component.


Introduction
Interactions of water molecules and mineral surfaces play amajor role in environmental sciences.Atthe involved water/ solid interfaces,t he hydrogen bonding network between neighboring water molecules is disturbed. Interface-induced phase transitions are consequences of such modifications of the intermolecular force balance. [1] Depending on the specific system, interface-induced order and disorder has been observed. [2,3] Forice in particular, interface-induced premelting is found by experiments and simulations. [4] In this case,an amorphous liquid layer near interfaces emerges below the bulk crystal/liquid phase transition. Thep resence of this nanoscopic liquid layer and its properties have important implications for friction [5] as well as for macroscopic geophysical processes,a sr eviewed by Dash et al. [6] Therefore, amolecular-level understanding of the structure and dynamics within the liquid layer is highly desirable.
In 1859, Faraday proposed the existence of al iquid-like layer at free ice surfaces. [7] Since then, the growth law of this liquid layer, that is,i ts thickness d(TÀT m )v s. temperature T below the bulk melting point T m ,w as extensively studied using various experimental techniques [8][9][10][11][12] and molecular dynamics simulations. [13,14] However,u nlike for the free ice surface [15,16] only relatively little is known about thermodynamic,s tructural, and dynamic properties of this premelting layer at ice/solid interfaces.T his includes materials parameters such as density,l atent heat of melting, or viscosity and thermodynamic response functions such as compressibility, heat capacity,ordielectric constants. [17] Moreover,properties on the molecular level, such as pair correlation functions, coordination number, rotational motion, energy transfer rates between adjacent water molecules connected by hydrogen bonds and their ability to form complexes and hydration shells,m ight deviate from bulk. However,d ue to al ack of experimental data, theoretical models [18,19] rely on bulk values for the latent heat of melting or ion solubilities. By X-rayreflectivity, Engemann et al.f ound aq uasiliquid layerw itha20 %h igherd ensity compared to bulk water. [10] Recently, high-density liquid waterw as observed at the interfaceb etween watera nd high-pressurei ce IIIo rV Ib y opticalm icroscopy. [20] This indicatest hatt he interfacialp remeltingl ayer ands upercooled bulk watere xhibit remarkable structural differences.Furthermore,suchdensity increase also suggests changesi nt he waterm obilityw ithint he premelting layer. Fort he viscosity, as trong increaser elative to supercooled bulk waterwas found. [21] Whilefrictionforce measurementsa ti ce/quartzi nterfacesg avea ni ncreaseb ym oret han oneo rder of magnitude, [22] quasi-elasticn eutron scattering (QENS) [23] on graphitizedc arbonb lack indicateda ni ncrease by less than afactoroftwo.These observations areparticularly important for the viscoelastic properties of partially frozen ice/solid composites,serving as model systems for permafrost.
Intense attention has also been paid to experimental observations of water diffusion in swollen clays. [24] Depending on the water/solid interactions,m olecular dynamics simulations of interfacial water adjacent to hydrophobic and hydrophilic surfaces indicate ar elative increase or decrease of the diffusion constant, respectively. [25] However,particle-tracking studies in electric field gradients, [26] nuclear magnetic resonance (NMR), [27] and QENS [24] provide no clear evidence that these results are transferable to the water mobility within the interfacial premelting layer. Therefore,d espite its importance,the understanding of the surface and interfacial melting of ice,i np articular the molecular scale dynamics within the premelting layer, is still under debate.
Hydrogen exhibits al arge incoherent neutron scattering cross section. [28] This makes QENS an ideal technique to provide information on the dynamics of liquid water. [29,30] In addition to bulk measurements,QENS is also an ideal tool to quantitatively study water dynamics in wet clay minerals, [31][32][33][34] surface melting of adsorbed multilayer films, [35][36][37] or interfacial ice melting in powders with high surface to volume ratio. [23] Herein, we studied the translational water diffusion coefficient D t in the interfacial premelting layer of ice/clay nano-composites.T he clay platelets have ap lanar geometry, large surface-to-volume ratios,a nd am olecular scale surface roughness.T his ensures that the formation of the premelting layer is governed by intrinsic interfacial premelting rather than the Gibbs-Thomson effect. TheG ibbs-Thomson effect is ac hange in the melting point due to curvature of the interface.Therefore,itstrongly affects premelting in ice nanocrystals, [38] spherical nano-powders, [39] or nano-pores. [40] Experiments Te mperature dependent QENS experiments (stability AE 0.01 8 8C) were carried out from À100 8 8C. Subsequently,t he temperature was gradually increased to above the bulk melting point. To elucidate the influence of the water/solid interactions,w ec ompare QENS results from the charged hydrophilic clay vermiculite,t he uncharged hydrophilic clay kaolin, and the more hydrophobic clay talc.Detailed sample preparation can be found in the SI. Thep reparation of the clay minerals and their morphology characterization by scanning electron microscopy and atomic force microscopy has been reported previously. [41] Quantitative analysis of nitrogen adsorption isotherms using aB runauer-Emmett-Te ller (BET) slit model gave as pecific surface area of 10.5 m 2 g À1 (vermiculite), 10.2 m 2 g À1 (kaolin), and 4.9 m 2 g À1 (talc) ( Figure S1). Thew ater content was determined by thermogravimetric analysis (TGA/DSC 3 + ,Mettler Toledo) to be 33.6 wt %( vermiculite), 17.6 wt %( kaolin), 17.4 wt % (talc I), and 16.2 wt %( talc II), respectively.F or the QENS measurements,wet clay samples of approx. 0.5 mm thickness were contained in flat, rectangular aluminum cells.

Results and Discussion
Representative QENS spectra from the kaolin/water composite sample at q = 1.25 À1 in the temperature range between À100 8 8Ca nd + 2.7 8 8Ca re shown in Figure 1a.W ith increasing temperature,t he intensity in the broad wings (quasi-elastic peak) around the elastic peak at E = 0meV increases.This indicates the gradual growth of the liquid layer as temperature approaches the bulk melting point of water. Figure 1b shows the spectra at À1.3 8 8Cf or four different momentum transfers q between 0.45 À1 and 1.65 À1 .A s expected for translational diffusive motion, with increasing q the wings broaden [Eq. (1) and Ref. [42]].S imilar behavior was observed in QENS spectra of talc/water and vermiculite/ water samples ( Figure S2 and S3).
To quantitatively evaluate the dynamics of water molecules and extract the translational diffusion coefficient D t of the interfacial liquid layer, amodel-free approach was used to consistently analyze all the QENS spectra with momentum transfers q = 0.45 À1 to 1.65 À1 at agiven temperature. [42] In this approach, the quasi-elastic signal is composed of two Lorentzian functions L r (q,w)and L t (q,w). They represent fast rotational and slow translational jump diffusion of water molecules.T he q-dependent linewidth G t (q)o ft he narrow (slow) Lorentzian spectral component L t (q,w)was constrained by Equation (1), assuming aconstant apparent jump length l = 0.77 according to Qvist et al. [42] ( Figure S6). Details of the fit functions and analysis procedure are summarized in the SI.
To emphasize the QENS signal from the interfacial premelting layer, difference spectra were calculated by subtraction of the signals recorded on the completely frozen samples at À100 8 8C. Fitting of these difference spectra results in as table analysis procedure.F or all datasets,t he measured patterns are perfectly reproduced. Figure 2s hows the experimental data (blue points) and the calculated spectra I(E)(red curve) for the ice/kaolin composite sample at À1.3 8 8Ca nd 1.25 À1 momentum transfer. Theg reen curve in Figure 2a shows the elastic component. Its Gaussian line shape is given by the instrumental resolution. Figure 2b summarizes the contributions to the quasi-elastic signal. Fore ach temperature,t he FWHM G r of the broad (fast) Lorentzian component (yellow curve) was constrained to ac ommon value within a q series.Using this procedure,robust parameters that exhibit ac onsistent temperature variation were obtained for all ice/clay composite samples. Figure 3summarizes the translational diffusion coefficient D t of the premelting water fraction extracted from QENS measurements on the talc,k aolin, and vermiculite samples (symbols) in an Arrhenius plot. Va lues of supercooled liquid bulk water, determined by Qvist et al. [42] (black curve), are given for comparison.
Below the bulk melting point of water, as ignificant slowdown of the translational diffusion is observed in all clay composites.T his effect is most pronounced for the charged hydrophilic vermiculite.I nc ontrast, for the uncharged talc as maller slowdown of less than 11 %i sf ound. While the observed reduction of D t is highly significant, its values clearly show that for all cases studied here the interfacial premelting layer is liquid. These values are similar to the results reported by Maruyama et al. for hydrophobic graphitized carbon black. [23] At low temperatures,t hat is, T m ÀT > 4K (Figure 3, region III), the values of D t extracted from the QENS data indicate aflattening of the curves.Inthis temperature region, an effective interfacial premelting layer thickness smaller than 1.4 nm and 2.0 nm was determined by high-energy X-ray diffraction [41] for ice/vermiculite and ice/kaolin samples, respectively.T hese values approach characteristic structural water dimensions.Athickness of 2.0 nm corresponds to approx. seven 2.8 monolayers of liquid water [43] or three times the 0.736 nm lattice constant of ice Ih along its c axis, that is,perpendicular to the basal plane. [44] On the other hand, this thickness is significantly larger compared to swollen clays where only one or two water monolayers are intercalated in between clays. [31][32][33][34] For0< T m ÀT < 4K (Figure 3, region II), astrong increase of D t with increasing temperature T is observed. Experimental studies showed that, at these temperatures,the premelting layer is rapidly thickening. [6,[9][10][11]41] Therefore,w ith increasing  temperature the geometric confinement effect on the water molecules inside the premelting layer will decrease.
Above its bulk melting point T m (Figure 3, region I), the ice confined between clay platelets is totally molten. From the water fraction in the clay composites and specific surface area we estimate an average water layer thickness of 97 nm (vermiculite), 42 nm (kaolin), 86 nm (talc I), and 79 nm (talc II) between the clay platelets.Arelatively small decrease of the translational water diffusion coefficients D t by approx. 12 %i so bserved for talc and kaolin samples (Figure 3, region I). This arises from the contribution of the small fraction of less mobile interfacial water.
It is expected that the water mobility within the interfacial premelting layer is affected by the interactions between premelting water molecule and the clay surfaces.Bare kaolin surfaces have as urface energy of 171 mN m À1 with ar atio of 40 %d ispersive and 60 %n ondispersive interactions. [45] The positive spreading coefficient of 76 mN m À1 ,calculated by the Fowkes method, [46] reflects the strong hydrophilic nature of kaolin surfaces.Incontrast, the attraction of water molecules to the more hydrophobic talc is significantly smaller. At low temperature,t hat is,t hin premelting layers (Figure 3, region III), the mobility of water molecules in the premelting layer follows the trend: D vermiculite < D kaolin < D talc .T herefore,t he more hydrophilic the clay mineral, the stronger the D t decrease within the premelting layer.
TheN etz group investigated the water dynamics near hydrophobic and hydrophilic membranes by molecular dynamics simulations. [25] Fordistances between water molecules and the hydrophilic membrane smaller than 1nm, the water diffusion coefficient D t decreases dramatically.O nt he other hand, near hydrophobic surfaces as mall increase of D t was found. However,f or larger wall separations D t quickly approaches its bulk value.T hese findings are consistent with our observation that at T m ÀT < 4KD t increases rapidly.Since none of our clays are strongly hydrophobic, the weak increase of the diffusion constant near hydrophobic interfaces,p redicted by simulations, [25] is not found.
Then ano-confinement effect on water dynamics in reverse micelles was investigated by the Fayer group using ultrafast infrared pump-probe spectroscopy. [47][48][49] Fors mall reverse micelles with diameters d 2.5 nm, they observed spectral signatures from as ingle water ensemble only.T hese results suggest that in the temperature region III (Figure 3) there is only one liquid water ensemble present in the nanoscopic premelting layer of clay composites.T his observation is consistent with the flattening of the D t curve in region III. However,f or larger spherical pores or slit pores with sizes 2.5 nm d 5.5 nm, two water ensembles have been found. [47][48][49] Thefirst one comprises the core,the second one includes the water molecules adjacent to the interface. Likewise,w ith increasing premelting layer thicknesses,c ontributions from the fast-translational water diffusion apart from the clay surface will gradually start to dominate the average QENS signal. This readily explains the rapid increase of D t observed in region II of Figure 3.
Theproperties and nature of the premelting layer formed at ice surfaces and interfaces is atopic of intense debate.Smit et al. found that the sum-frequency generation spectra (SFG) from the ice premelting layer and supercooled bulk water are indistinguishable. [50] Therefore,they deduced that the surface of ice is more like supercooled liquid water down to 245 K. [50] However,u sing the same experimental technique Sµnchez et al. [12] found that the SFG response from ice surfaces at 270 Kisdifferent compared to supercooled water at the same temperature,b ut more similar to that of ice at 243 K. This indicates that the premelting water forms stronger hydrogen bonds than supercooled bulk water.
While SFG spectra probe the vibrational states of the outermost water molecules adjacent to interfaces,X -ray reflectivity (XRR) is sensitive to density and thickness.F rom XRR experiments on ice/SiO 2 interfaces,E ngemann et al. [10] deduced the presence of apremelting layer afew nanometers thick with adensity of 1.2 gcm À3 .T his density is significantly different from that of liquid bulk water. However,inaddition to ordinary liquid water av ariety of liquid water and amorphous ice structures was discovered at low temperatures and/or high pressures: [10,51,52] low-density liquid water (LDL, 1 = 0.92 gcm À3 ), high-density liquid water (HDL, 1 = 1.15 gcm À3 ), low-density amorphous ice (LDA, 1 = 0.94 gcm À3 ), and high-density amorphous ice (HDA, 1 = 1.17 gcm À3 ). Therefore,Engemann et al. proposed astructural relationship between the premelting layer and high-density liquid or amorphous water structures.
We now compare our results in Figure 3with the temperature dependence of D t for normal and supercooled (238-273 K) liquid bulk water, amorphous solid water (ASW:LDA, 150-160 K), crystalline ice, [53] and HDA. [54] Fora ll clay composites, D t is much lower than for supercooled water. On the other hand, values for D t are four or five orders of magnitude higher than that for ASW/ice (Figure 3i n Ref. [53]) and HDA. [54] Therefore,w ec onclude that the interfacial premelting layer is liquid rather than an amorphous solid. This is consistent with the inelastic neutron scattering results by Zanotti et al. [55] In this work, interfacial liquid water was also found when heating Vycor samples with adsorbed water monolayers from 77 to 280 K. [55] At 240 K, they observed that the interfacial liquid water changes from alow-density to ahigh-density liquid. All D t values shown in Figure 3are obtained at temperatures higher than 240 K, that is, T À1 < 4.17 10 À3 K À1 .A ccording to Zanotti et al., [55] this would mean that the premelting water is HDL with almost constant D t values,which is contrary to our result. However, supercooled liquid bulk water is composed of spatially and temporally fluctuating LDL and HDL mixtures. [56,57] Moreover, in deeply supercooled water droplets,the Nilsson group recently observed aW idom transition by using femtosecond X-ray laser pulses. [52] Even at ambient conditions above the Widom line,LDL fluctuations exist in primarily HDL water. And this LDL component can adsorb to solid interfaces. Additionally,n eutron diffraction and MD simulations by Soper and Ricci [58] and Mishima and Stanley [59] have shown that LDL is more "structured" and HDL is more "liquidlike". This implies as lower diffusion for LDL compared to HDL. Therefore,wesuggest that the premelting layer close to solid surfaces contains more "structured" LDL. This interpretation is also consistent with the results obtained from SFG by Sµnchez et al. [12] Based on the above discussion, we propose ap icture for the structural evolution of ice confined in porous clay minerals (Figure 4). Upon heating, ice Ih confined in clay minerals forms ap remelting liquid. Up to T m ÀT = 4K,t he thickness of the interfacial liquid layer is around 2nmorless. This liquid consists primarily of the more "structured" LDL water since HDL fluctuations are suppressed adjacent to ice and clays.A th igher temperatures,a nH DL fraction with faster translational diffusion starts to emerge in regions where water molecules are further apart from the clay and ice surfaces.Spatial and temporal fluctuations between the LDL and HDL components lead to ac ontinuous increase of the diffusion constant as shown in Figure 3r ather than as harp transition.
This interfacial premelting mechanism has the following implications:T he very low friction observed at ice/solid interfaces is explained by the presence of athin film of liquid water. Aside from ice melting caused by energy dissipation from friction, this liquid film is also caused by the intrinsic interfacial premelting layer studied in this work. Frictional forces are controlled by surface roughness and the viscosity of the interfacial water h i = k B T/(6pD t r). Thel atter is directly linked to the diffusion by the Stokes-Einstein relation. Experiments determining the interfacial water viscosity from frictional forces yield values for h i up to afactor of 20 higher than that of supercooled bulk water h b . [22] In contrast, our QENS study gives values h i /h b 2. This quotient is about one order of magnitude smaller than calculated from shear forces in friction experiments assuming ideally smooth interfaces. This indicates the importance of surface roughness for aq uantitative description of friction at ice/solid interfaces. Therefore,o ur results contribute to the understanding of friction at ice/solid interfaces. Furthermore,t he viscosity within the interfacial premelting layer can affect the mechanical properties of permafrost. In such partially frozen ice/mineral composites,m aterials properties are not only determined by the sum of the individual components.F or composite engineering materials, it is known that interfaces connecting filler and matrix can be significant. Likewise,t he viscosity and thickness of the interfacial premelting layer affects the adhesion between mineral particles and ice.T herefore,t he premelting layer is expected to affect the viscoelastic properties of permafrost. Moreover,interfaces can be important for impurity migration in composite materials.F or instance,t he mobility of plant nutrients and contaminants inside minerals and ice crystals is very low.Therefore,their diffusion along interfaces can be the dominating transport mechanism, despite the small volume fraction of the interfacial premelting layer. As the temperature approaches the melting point, the interfacial layer is thickening.Inthe context of global warming, the viscosity of the premelting layer could therefore gain increasing relevance.

Conclusion
Thet emperature dependence of the translational selfdiffusion coefficient D t of water within the premelting liquid layer of three different clay/ice nanocomposites was studied by QENS.Three distinct temperature regions were observed. At low temperature T m ÀT > 4K(region III), areduced water mobility within the premelting layer compared to supercooled bulk water was obtained. It is suggested that the water molecules within this premelting layer with less than 2nm thickness form an LDL structure.I nt his region, D t exhibits ac lear trend with the water-substrate interaction strength. Them obility slowdown is most pronounced for the charged hydrophilic vermiculite,f ollowed by the kaolin and more hydrophobic talc samples.A tT m ÀT < 4K (region II), in all clays the D t value of the premelting liquid strongly increases with temperature.T his effect is explained by the decreasing contribution of water molecules located in direct vicinity of the solid/liquid interface as the fraction of HDL is increasing in the premelting layer.
These results will help to understand friction at ice/solid interfaces.F urthermore,r elevance is seen for phenomena related to the water mobility in partially frozen soils. Examples include the transport of guest molecules such as plant nutrients or contaminants within the premelting layer and geochemical reactions such as ion exchange processes at ice/mineral interfaces. . Schematic picture of the structural evolution of ice confined in clay mineral materials when warming it up. The red line shows the growth law of the premelting layer thickness for kaolin, adapted from ah igh-energy X-ray diffraction study using the same kaolin batch. [41] The light-blue area represents ice Ih, the dark-blue area LDL, and the orange area bulk-like liquid water composed of an LDL + HDL mixture. Spatial and temporal fluctuationsinto LDL and HDL phases appear on either side of the transition (inset).