Molecular Mechanisms Causing Anomalously High
Thermal Expansion of Nanoconfined Water
Stephen H. Garofalini,*[a] Thiruvilla S. Mahadevan,[a] Shuangyan Xu,[b] and
George W. Scherer[b]
with the structure and increased CTE of a region of water ~ 6
Anomalously high thermal expansion is measured in water con-
fined in nanoscale pores in amorphous silica and the molecular thick adjacent to the silica. The structure of water in the first 3
mechanisms are identified by molecular dynamics (MD) simula- of this interface is templated by the atomically rough silica sur-
tions using an accurate dissociative water potential. The experi- face, while the water in the second 3 just beyond the atomical-
mentally measured coefficient of thermal expansion (CTE) of ly rough silica surface sits in an asymmetric potential well and
nanoconfined water increases as pore dimension decreases. The displays a high density, with a structure comparable to bulk
simulations match this behavior for water confined in 30 and water at higher pressure.
70 pores in silica. The cause of the high expansion is associated
1. Introduction
The behavior of nanoconfined water has been studied in Density profiles of water as a function of distance from the
recent years because of its importance in geomorphology, ar- pore center to the silica surface show a 5 thick layer with ele-
vated density in simulations.[4, 5] These, and other simulations
chitectural stones, desalination membranes, and biological
membranes where water is confined by hydrophilic or hydro- of water/silica interfaces (where in some cases the silica is
phobic surfaces. The work presented here is related to the modeled as a quartz surface rather than glassy silica) all show
former. a significant increase in the density of water at the interface.
Early work by Derjaguin et al.[1] showed anomalous expan- The peaks in density range from 20 % above bulk water densi-
ty[4, 5, 9, 13] to 50 %,[14, 15] and even higher.[16, 17]
sion of water confined within 50 pores in silica xerogels.
They determined that the difference in the structure of con- In all of the previous simulations of confined water in silica,
fined water and bulk water disappeared at 70 8C. More recent the water is modeled as rigid or flexible molecules using SPC,
experiments shown here and elsewhere[2] indicate that the dif- SPC/E, or TIP4P potentials and, in most cases, frozen silica. The
ference in coefficient of thermal expansion (CTE) increases with silica surfaces are manually hydroxylated to create the surface
decreasing pore size and the CTE of water confined in 30 silanols expected on real surfaces. All such simulations ignore
pores in silica equals that of bulk water at a temperature the important dissociative chemisorption of water on silica, the
~ 16 8C higher. The latter experiments provide precise data for rupture of strained siloxane bonds by water, the possible pene-
the change in volume with temperature for water confined tration of water and silanol formation into the sub-surface via
within nanopores that constitute a rigorous standard to which openings in the ring-like network structure of silica, and the re-
computational studies could be compared. laxation of the atoms in the silica while interacting with water
The structure of water confined in Vycor, a porous silica molecules. Most importantly, none of these popular water po-
glass containing 40 80 pores, has been studied using both tentials reproduce the liquid equation of state to the level
experimental and computational techniques.[3 11] Neutron dif- needed to evaluate the anomalous expansion of confined
fraction experiments indicate a change in the structure of the water for comparison to experiment.
confined water, and structural changes are observed in MD To achieve a more realistic simulation of the behavior of
simulations.[5, 10] However, there are contradictory results. In water confined in silica nanopores, we employed a newly de-
one case the results are interpreted to mean that water in the veloped dissociative water potential that matches many bulk
center of the pores has a structure similar to bulk water at a water properties, such as the structure, heat of vaporization
temperature 30 8C higher,[9] but another study concludes that
water in the pore interior is similar to bulk water,[7] both of [a] Prof. S. H. Garofalini, Dr. T. S. Mahadevan
which contradict other data that indicates the water confined Department of Materials Science and Engineering
Rutgers University, Piscataway, NJ (USA)
in 150 pores is similar to bulk water at a temperature 15 K
Fax: +1-732-***-****
lower.[12] Similarly, water within 5 of the interface is believed
E-mail: abqkzk@r.postjobfree.com
to have a change in bonding and structure that is said to lead
[b] Dr. S. Xu, Prof. G. W. Scherer
to less bonding[5, 7] or to more bonding.[12] Department of Civil and Environmental Engineering
Princeton University, Princeton, NJ (USA)
1997
ChemPhysChem 2008, 9, 199*-****-**** Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
S. H. Garofalini et al.
(MD = 10.5 kcal mol 1 vs EXP = 10.52 kcal mol 1), diffusion coef- 2.2. Density Distribution and Structure of Nanoconfined
ficient (MD = 2.45 cm2 s 1 vs EXP = 2.3 cm2 s 1), dipole moment Water
(MD = 2.6D vs exp = 2.6D), liquid vapor coexistence curve, and
the liquid equation of state.[18] This potential has been used to The density of atoms as a function of distance perpendicular
study the interactions between a silica-glass surface and to the interfaces for the 30 film at 298 K is shown in
water.[19] Results showed dissociative chemisorption of the Figure 2. The results show three major features that differ from
water, silanol concentration consistent with experiment, and
reaction mechanisms that included the formation of hydroni-
um ions at the interface[19] that are completely consistent with
ab-initio MD simulations.[20] Details of the generation of the po-
tentials and cross-interactions were previously presented.[18, 19]
The new dissociative water potential offers the unique ability
to reproduce the interactions occurring at the confined water/
silica interface, allowing realistic simulation of the change in
volume, V, as a function of temperature, T.
2. Results and Discussion
2.1. Volume Temperature Behavior Figure 2. Density profiles of species as a function of distance perpendicular
to the interfaces. Silica = original Si and O in silica, Remaining H2O = water
MD simulations using plates of amorphous silica separated by molecules remaining after reactions with glass, O water = oxygen in remain-
30 and 70 films of water are performed at several tempera- ing water molecules, H water = hydrogen in remaining water molecules,
tures to evaluate expansion behavior. The water reacted with Ow Si = oxygen originally in water that chemisorbed and bonded to Si, H
OSi = hydrogen attached to any O that is bonded to Si. The latter two spe-
the silica surfaces, as in previous studies of water adsorption
cies indicate silanols.
and reactions on silica surfaces using this dissociative water
potential.[19] Results of the simulations are shown in Figure 1,
all previous studies of confined water in silica. First, there is
only a small (~ 5 %) increase in density of water at the inter-
face; in contrast, all other simulations that used the popu-
lar rigid water potentials show increases in water densities
! 20 %. Second, water molecules react with the silica surface,
forming silanols. These dissociatively chemisorbed water mole-
cules are shown in Figure 2 as the distribution of oxygen from
the water that is attached to a silicon (labeled Ow Si) and hy-
drogen, attached to any oxygen, that is attached to a silicon
(H on silanols and labeled H OSi). Third, undissociated water
molecules migrate into the glass via openings in the normal
ring structure of silica. The reactions and penetration of water
extend about 7 below the outermost surface oxygen.
Figure 1. Normalized volume change with temperature for two pore sizes In one case, it has been inferred that the water in the interi-
and bulk water for experimental data and molecular dynamics (MD) simula- or of nanopores in silica has a structure consistent with bulk
tions.
water 30 8C warmer.[9] To evaluate the structure of our confined
water, O O pair distribution functions (OO PDFs) of oxygen in
along with experimental data for bulk water and water con- water are generated as a function of location parallel to the
fined in silica xerogels with 30 and 70 pores. The simulations water/glass interface for the 30 water system. The results
were previously shown to match V(T) of bulk water.[18] In from the 70 system are very similar. While all data below will
Figure 1, the simulations show the changes in V(T) for confined involve the interface at the lower Z dimension, results are simi-
water consistent with the experimental results for both pore lar for the interface at the upper Z dimension. Starting from
sizes. The experiments clearly show that confinement causes the Z = 20 location on Figure 2 (vertical dashed line), 2D OO
an increase in CTE in comparison to bulk water, but shed no PDFs were generated for water molecules located within
light on the atomistic mechanisms responsible for this behav- volume elements 3 thick (in Z) parallel to the interface for
ior. However, since the MD simulations reproduce the experi- the 30 film. Results are shown in Figure 3 a, along with simi-
mental trends, it is reasonable to use them to investigate the lar data for bulk water. The curve from 35 38 is indicative
mechanisms. Results involving the structure of the simulated of the structure of water in the interior of the pore (away from
systems are presented below primarily for the 30 film, al- the water/silica interface); it is lost behind the bulk water
though the results for interfacial and interior structure apply to curve, indicating that the water in the interior of the pore is
both thicknesses. very similar to bulk water. Other locations in the interior of the
1998 www.chemphyschem.org 2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemPhysChem 2008, 9, 1997 2001
High Thermal Expansion of Nanoconfined Water
Figure 4. Snapshot of a silica glass surface with the size of the oxygen exag-
gerated to emphasize the atomistic roughness of the surface.
from that of bulk water. Owing to the templating effect of the
very stable silica surface, this structure shows little change
with temperature, as shown in Figure 3 b with the OO PDF
from water in the 20 23 range at 298 K and 348 K. The struc-
ture of this region at 298 K is the same in the 30 and 70
films, showing that the pore dimension did not affect this in-
terface structure.
The structure of water farther from the interface (23 26
and 26 29 ) shows smaller deviations from bulk water as the
distance from the interface increases, as shown in Figure 3 a.
Water beyond ~ 7 10 from the interface has a structure simi-
lar to bulk water. Therefore, the behavior causing the anoma-
lously high CTE of confined water is located near the interface.
This accounts for the data in Figure 1: the smaller the pore,
the greater the ratio of the affected interfacial volume to total
pore volume, resulting in a greater influence of the interface
on the properties of the pore.
The water just beyond the end of the silica surface, at 23
26, is clearly different from bulk water at the first minimum
(~ 3.4 3.5 ). This increase in the intensity of the OO PDF at
Figure 3. a) OO pair distribution function (PDF) for 3 regions as a function
the first minimum has been observed for water at elevated
of distance starting from the interface at 20 in Figure 2, plus bulk water
using a summation of 3 regions over the entire bulk. b) OO PDF for temperature as well as water at higher pressure. However,
oxygen in physisorbed water on the silica surface at two temperatures for
higher temperature water shows a decrease in the intensity of
the 30 film, showing little change in structure with temperature, and in
the first maximum and a broadening of that peak that are in-
the 70 film, showing little change with film thickness. c) OO PDF for
compatible with the 23 26 curve. The density profile in
oxygen in the 30 film for the two layers of water closest to the interface
and that for OO in bulk water at 750atm, showing the similarity between Figure 2 shows a 5 % higher density in the 23 26 region
water at higher pressure and the water in the high density region near the
than bulk water, which, on average, is similar to water at a
interface (cf. Figure 2).
pressure near 750 atm. Figure 3 c shows that the OO PDF for
simulated bulk water at 750 atm (75 MPa) and 298 K is very
pore give similar results. This is similar to some studies,[7] but similar to that for the 23 26 region, indicating that the struc-
quite different from another[11] that shows a change in the ture near the interface is similar to that of bulk water at higher
water structure extending from the water/silica interface to the pressure. The effect of pressure on structure has been previ-
ously discussed.[21] However, the similarity to high pressure
center of a 40 pore.
The structure of a glassy silica surface is shown in Figure 4, water is limited.
which is a snapshot of the top of a simulated glass surface in
which the size of the O ions in the drawing is exaggerated to
2.3. Cause of High Expansion in Nanoconfined Water
emphasize the rugosity of the structure. This atomically rough
surface allows for physisorbed water molecules to be coinci- The simulations of nanoconfined water were done at 1 atm,
dent with outer O from the glass surface, as seen in the densi- not at elevated pressure, so the cause for the density increase
ty distributions in Figure 2 at the interfaces. The 20 23 OO near the interface is the interactions with the glass surface, re-
PDF curve, using only oxygen from water molecules, is gener- sulting in an energy minimum in the 23 location. Figure 5
ated from water physisorbed on the atomically rough silica shows the average energy per atom as a function of distance
surface and clearly shows the greatest change in structure perpendicular to the interface from the interface (20 ) into
1999
www.chemphyschem.org
ChemPhysChem 2008, 9, 199*-****-**** Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim
S. H. Garofalini et al.
(23 26 ) is consistent with water at high pressure, based on
the OO PDF. It is known that water at high pressure shows
higher thermal expansion than normal bulk water,[22] so the en-
hanced CTE shown in Figure 1 can be attributed to this high
density region. The change in density as temperature is in-
creased in the confined water shows that the interior region
(35 38, and even thicker interior regions) behaves similarly
to bulk water; however, the density of water near the interface
(23 26 ) decreases faster than bulk water (again, consistent
with the effect of high pressure on expansion). In fact, both
the 20 23 layer and the 23 26 layer decrease in density
Figure 5. Average energy per atom in water molecules as a function of dis-
more rapidly than bulk water. Using 1* as the ratio of the den-
tance perpendicular to the interface (see Figure 2) showing the potential
sity of the layer normalized by the bulk water density at that
well near 23 that attracts water, creating the high density region, and the
asymmetry in the energy curve adjacent to the well. temperature, a plot of ln(1*) vs T is linear, indicating that the
relative density change of a layer to that of the bulk is con-
stant over this T range. The slope of this plot results in a CTE
of the combined layers from 20 26 equal to 5.9 e 4 8C 1.
the interior of the confined water (40 ), showing an energy
minimum near 23 24 . This creates an attractive force to this Using this 6 layer as the relevant high expansion region,
location, resulting in the higher density. The enhancement in times 2 for both interfaces, means that the interface contrib-
binding energy between this high-density region and the inte- utes 40 % to the volume of the 30 film and 17 % to the 70
rior of the confined water film is ~ 1 kcal mole 1 of water mole- film. A calculation of the contribution to CTE of the confined
cules. The curve in Figure 5 also indicates an asymmetry in film from the interfacial layer plus the interior (which has the
forces, with a higher repulsive wall in the Z direction and a bulk water CTE) indicates that the interfacial region must have
a CTE near 6.4 e 4 8C 1 for the 30 film and 6.7 e 4 8C 1 for the
lower barrier in the + Z direction.
The number of hydrogen bonds per oxygen in the different 70 film to account for the CTE of the confined films. These
results are very close to the 5.9 e 4 8C 1 value obtained in the
volume regions and bulk water is shown in Figure 6. A H-bond
interface in the simulations. Thus, the simulations show that
the higher CTE of the 6 of water at the interface creates the
observed higher CTE of the confined water. The volume frac-
tion of the interface in a spherical or cylindrical pore is larger
than in a film, but the experimental systems had a range of
pore sizes (albeit narrow), so more precise comparisons are
not meaningful.
In contrast to hydrostatically compressed water, the interfa-
cial layer is in a highly asymmetrical potential well (Figure 5),
where movement toward the glass surface is energetically
blocked. Consequently, thermal agitation tends to drive water
molecules from the interface toward the interior of the pore,
Figure 6. Number of hydrogen bonds per oxygen in water molecules as a
leading to a net larger volume expansion.
function of region perpendicular to the interface and for bulk water.
3. Conclusions
is defined as having an O H distance less than 2.4 and an
H O O angle less than 308. H-bonds at the interface include We conclude that the anomalously high CTE of nanoconfined
water results from the high expansion of the ~ 6 layer at the
silanols as H-bond donors or acceptors interacting with O in
water at the interface (H-bonds from silanols to any O bonded surface of the pore caused by the local density and asymmetry
to Si are not included). The number of hydrogen bonds per of the potential well. Simulation of this behavior of nanocon-
oxygen in the interior of both the 30 and 70 films is similar fined water requires an interatomic potential that accurately
to that of bulk water, with a slight decrease in the 23 26 reproduces the liquid equation of state for bulk water and
region (which is consistent with water under higher pressure) allows for an accurate representation of the reactions and in-
and a greater decrease at 20 23 due to the presence of the teractions between water molecules and the silica surface. The
silica. density and strength of binding of this layer can be accurately
The structure that we observe in these simulations with the described using the recently developed dissociative potential
dissociative water potential shows results consistent with the for water, enabling simulation results similar to the experimen-
anomalous expansion of confined water that is observed ex- tal data for the expansion of nanoconfined water. The dissocia-
perimentally. This structure deviates from what has been ob- tive potential gives results that are quantitatively, and in some
served in simulations using non-dissociative (rigid or flexible) cases qualitatively, different from previous simulations based
water potentials. The structure in the higher density region on rigid water molecules.
2000 www.chemphyschem.org 2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ChemPhysChem 2008, 9, 1997 2001
High Thermal Expansion of Nanoconfined Water
Methodology [1] B. V. Derjaguin, V. V. Karasev, E. N. Khromova, J. Colloid Interface Sci.
1986, 109, 586 587.
[2] S. Xu, G. C. Simmons, G. W. Scherer, Mater. Res. Soc. Symp. Proc. 2004,
The experimental method has been described in ref. [2]. The 790, 85 91.
molecular dynamics simulations used an amorphous silica [3] P. Gallo, M. Rovere, M. A. Ricci, C. Hartnig, E. Spohr, Phil. Mag. A 1999,
made via a melt-quench process similar to that published in 79, 1923 1930.
[4] E. Spohr, C. Hartnig, J. Mol. Liq. 1999, 80, 165 178.
ref. [19]. In the current case, the silica glass dimensions are
[5] M. A. Ricci, F. Bruni, P. Gallo, M. Rovere, A. K. Soper, J. Phys. Condens.
~ 64 x 64 x 40 in X, Y, and Z, with the water film filling X Matter 2000, 12, A345 A350.
and Y and ~ 30 thick or ~ 70 thick in Z, with periodic boun- [6] P. Gallo, M. Rovere, M. A. Ricci, C. Hartnig, E. Spohr, Europhys. Lett. 2000,
daries in all directions. The simulations were done in modified 49, 183 188.
[7] C. Hartnig, W. Witschel, E. Spohr, P. Gallo, M. A. Ricci, M. Rovere, J. Mol.
NPT (constant number, pressure, temperature) conditions, with
Liq. 2000, 85, 127 137.
constant X and Y dimensions and a constant pressure of 1 atm [8] M.-C. Bellissant-Funel, J. Phys. Condens. Matter 2001, 13, 9165 9177.
applied in the Z dimension, allowing the system to expand or [9] P. Gallo, M. A. Ricci, M. Rovere, J. Chem. Phys. 2002, 116, 342 346.
contract as a function of temperature in the Z direction. Simu- [10] V. Crupi, D. Majolino, P. Migliardo, V. Venuti, M. C. Bellissent-Funel, Mol.
Phys. 2003, 101, 3323 3333.
lations of 1 000 000 moves per temperature were run with a
[11] H. Thompson, A. K. Soper, M. A. Ricci, F. Bruni, N. T. Skipper, J. Phys.
timestep of 1 e 16 s, with volume becoming stabilized within Chem. B 2007, 111, 5610 5620.
the first 200 000 moves. The volume was averaged over the [12] J. Dore, Chem. Phys. 2000, 258, 327 347.
last 40 % of each run. Simulations of bulk silica alone at each [13] A. A. Hassanali, S. J. Singer, J. Phys. Chem. B 2007, 111, 111**-*****.
[14] N. Giovambattista, P. J. Rossky, P. G. Debenedetti, Phys. Rev. E 2006, 73,
temperature were used to generate the V(T) data for silica,
041 604.
which was subsequently subtracted from the total volume of [15] S. H. Lee, P. J. Rossky, J. Chem. Phys. 1994, 100, 3334 3345.
the combined water/silica system in order to generate the [16] T.-D. Li, J. Gao, R. Szoszkiewciz, U. Landman, E. Riedo, Phys. Rev. B 2007,
volume change of the water. 75.
[17] E. J. W. Wensink, A. C. Hoffman, M. E. F. Apol, H. J. C. Berendsen, Lang-
muir 2000, 16, 7392 7400.
Acknowledgements [18] T. S. Mahadevan, S. H. Garofalini, J. Phys. Chem. B 2007, 111, 8919 8927.
[19] T. S. Mahadevan, S. H. Garofalini, J. Phys. Chem. C 2008, 112, 1507 1515.
[20] Y. Ma, A. S. Foster, R. M. Nieminen, J. Chem. Phys. 2005, 122, 144709.
The authors wish to acknowledge funding from the DOE Office of [21] A. G. Kalinichev, Y. E. Gorbaty, A. V. Okhulkov, J. Mol. Liq. 1999, 82, 57
Science, Division of Material Sciences, grant number, DE-FG02- 72.
[22] L. T. Minassian, P. Pruzan, A. Soulard, J. Chem. Phys. 1981, 75, 3064
97ER45642.
3072.
Keywords: glass interfaces molecular dynamics Received: July 21, 2008
nanomaterials water Published online on September 11, 2008
2001
www.chemphyschem.org
ChemPhysChem 2008, 9, 199*-****-**** Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim