Journal of Molecular Liquids 290 (2019) 111213

Contents lists available at ScienceDirect

Journal of Molecular Liquids

journal homepage: www.elsevier.com/locate/molliq

Percolation transitions of physically and hydrogen bonded clusters in

supercritical water
Dimitrios T. Kallikragas, Igor M. Svishchev ⁎
Department of Chemistry, Trent University, 1600 West Bank Drive, Peterborough, ON K9L 0G2, Canada

a r t i c l e i n f o a b s t r a c t

Article history: Molecular Dynamics simulations were employed to locate the densities and pressures at which infinitely span-
Received 11 February 2019 ning networks form for both physically and hydrogen bonded water clusters along the 650, 700, 750 and
Received in revised form 11 June 2019 800 K supercritical isotherms. The SPC/E water model was used and the percolation thresholds were determined
Accepted 18 June 2019
by examining the cluster size distributions. Average hydrogen bond numbers for the entire system are reported
Available online 19 June 2019
and are near the critical number of 1.55 expected at the hydrogen bonded percolation threshold. Percolating
Keywords:
physical clusters appeared at lower densities than infinite hydrogen bonded networks. Physical and hydrogen
Supercritical water bonded percolation thresholds were plotted on the P-ρ and P-T planes. Neither of the lines of percolation transi-
Clusters tions match the extension of the vapor-liquid coexistence line into the supercritical region. Empirical relation-
Percolation ships were obtained for each transition threshold as functions of temperature as well as density. Activation
volumes of diffusion increase rapidly until they reach their maximum at the threshold densities where water
molecules are connected via percolating network of hydrogen bonds.
© 2019 Elsevier B.V. All rights reserved.

1. Introduction into the supercritical region [20]. However, this was later retracted
[21]. Woodcock proposed that there exists a supercritical mesophase
In recent years, the degree of hydrogen bonding as well as cluster bounded by percolation transition loci [22]. Below the critical tempera-
formation in supercritical water (SCW) has been studied through ture, this line of percolation transitions becomes the familiar liquid–
Raman and IR spectroscopy, neutron scattering as well as computer vapor two-phase coexistence line. The densities for the percolation
simulation methods [1–9]. Continuous tetrahedral network of hydrogen transitions in SCW given by Woodcock, appear to be much lower than
bonds observed at ambient conditions disappears in the supercritical those suggested by the experimental work of Bernabei et al.
state, yet various hydrogen bonded structures can exist even at very The thermodynamic response functions (e.g. the isothermal com-
high temperatures [4–7,10]. pressibility and isobaric heat capacity) exhibit extrema in the supercrit-
Clustering of water molecules and percolation transition at super- ical region [23,24]. It has been argued that the lines connecting the
critical temperatures and pressures has been mostly studied by compu- extrema of the response functions converge on approaching the critical
tational methods [1–3,11–16]. Kalinichev and Churakov conducted an point in a single line, the so-called Widom line [23]. Potentially, these
MD study with a detailed analysis of the topology of water clusters in extrema may be associated with the formation of an infinite cluster.
SCW and found that clusters form in predominantly chain like arrange- However, as argued by Nezbeda and Škvor [14], the loci of the response
ments [13,17]. Experimentally, Bernabei et al. have used neutron function extrema exhibit density/pressure dependence quite different
diffraction to identify the percolation transition in SCW, though for the from that of the percolation line. Imre et al. identify as many as seven
limited number of state points [18]. Based on extensive Molecular “Widom lines” corresponding to the maxima of the thermodynamic
Dynamics (MD) and Monte Carlo (MC) simulations, Campi et al., pre- response functions, all of which converge on the critical point [25].
dicted that infinitely spanning clusters can exist in the supercritical From a phenomenological perspective, the Widom line is frequently
phase of simple liquids, and that a threshold exists along a percolation evoked to mark a transition from gas like to liquid like regions in the
line beginning at the critical point [19]. The percolation transition for phase diagram. In a supercritical state, such a transition is not sharply
supercritical water has also been numerically examined by Partay defined and placing it on a phase diagram involves identifying a density
et al. [2,20,21]. Partay initially asserted that the threshold density for at which the behaviour changes. As an example, Corradini et al., identify
percolation follows an extension of the vapor-liquid coexistence line this transition as when the viscosity and inverse in the diffusion coeffi-
cients have an inflection point in the temperature coordinate [26].
⁎ Corresponding author. In this study we compare the threshold densities of physically bound
E-mail address: [email protected] (I.M. Svishchev). versus hydrogen bonded percolating clusters along the 650, 700, 750

https://doi.org/10.1016/j.molliq.2019.111213
0167-7322/© 2019 Elsevier B.V. All rights reserved.
2 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213

and 800 K supercritical isotherms. Our goal is to similarly differentiate a We have located the transitions for both physical and hydrogen
regime in the supercritical phase not necessarily of gas like or liquid like bonded clusters along four supercritical isotherms at a broad range of
behaviour of SCW but of different solvation and reaction environments pressures. This is to provide data on not only when one can expect to
dependent on water clustering and charge transfer conditions. It has find percolating clusters, but whether they will be predominantly phys-
been found that the geometry of H-bonded molecular arrangements, ically bound or hydrogen bonded in nature. The coordinates of the per-
namely the O\\H…O angle and bond length, remains invariant over a colating thresholds have been plotted on the pressure-temperature (P-
wide range of densities from a dilute gas-like to a highly compressed su- T) and pressure-density (P-ρ) planes of the phase diagram for the sim-
percritical fluid at a given temperature [27]. We use the geometric crite- ulated water model. This allows for such a transition to be easily visual-
rion for a hydrogen bond and where the distance between the oxygen ized and compared to the thermodynamic quantities such as the
and hydrogen of a neighbouring molecule is less than 2.5 Å [28–30] maximum of the response functions or viscosity and diffusion cross-
which corresponds to the first peak of the O…H radial distribution func- overs [26]. Hypothetically, these transition lines may as well point to
tion (RDF), and we define a physically bound cluster as having an the boundaries of the supercritical mesophase [33]. Empirical fits for
oxygen oxygen distance of up to 3.5 Å [31]. Using the definition of H- the percolation thresholds have been obtained. Through the principle
bonding from the O…H pair-distribution is similar to using the O…O dis- of corresponding states, these transition lines can be plotted for real
tance and the O…H\\O bond angle distributions, as we have found water or for any other water model.
throughout this and other studies [20]. In the case of physical clusters The localization of water molecules in clusters can affect their mass
defined by the O…O neighbor distance of 3.5 Å, but without a selective transport characteristics. We have calculated the activation volumes of
bond angle, any orientation is possible. In a nutshell, molecules in a diffusion for all state points simulated. It appears that clustering has a
physical cluster are predominantly H-bonded, but some are not. A typ- limiting effect on the activation volume marked approximately by the
ical such arrangement of water molecules is show in Fig. 1. These phys- hydrogen bonded percolation transition points.
ical clusters are more amorphic in nature and are a consequence of The rest of the paper is organised as follows: Section 2 provides the
electrostatic and van der Waals interactions holding the cluster together simulation details, Section 3 the results and discussion and Section 4
upon collisions of the molecules. Since the molecules in a physical or presents our conclusions.
collision-induced cluster can have any orientation to their nearest
neighbours, they will form at lower densities than are required for hy- 2. Simulation details
drogen bonded clusters at a given temperature. Analogous to this, per-
colating physically bound clusters should also appear at lower Classical MD simulations were used to obtain the percolation thresh-
densities than hydrogen bonded percolating clusters. Pugnaloni has in- olds of SCW along the 650, 700, 750 and 800 K isotherms with densities
corporated cluster lifespan in defining cluster types for a pure Lennard- ranging from 0.1795 to 1.0471 g cm−3 (0.006 to 0.035 Å−3). Dozens of
Jones fluid and found that regions of different clustering regimes exist simulations were run at each temperature with system sizes of 500
within the phase diagram, similar to what we describe below [32]. and 3000 water molecules. The percolation threshold was identified
From a practical view point, identifying the transitions from physically by the cluster size distribution, P(n):
bound to hydrogen bonded clusters can help better understand electro-
chemical, charge transfer processes facilitated by interconnected hydro- P ðnÞ ≈ n−τ ;
gen bond networks, e.g. in corrosion phenomena. In the case of
physically bound clusters, solvation based chemical phenomena are ex-
pected to be of influence. Detailed information on the densities and NðnÞ
P ðnÞ ¼ X ; ð1bÞ
pressures at which to expect these rapid changes in solvation, wetting, NðnÞ
corrosion, etc., may help in chemical synthesis and reaction engineering. n
where P(n) is the mean number of clusters of size n, and τ is the critical
exponent, which is equal to 2.19 at the percolation threshold as de-
scribed by Campi [19], as well as Stauffer and Aharony [34]. Eq. (1b)
shows the numerical calculation of the probability of cluster formation
P(n) used in Fig. 3. Here N(n) is the mean number of clusters of size n,
and the summation term in the denominator is a normalization factor.
Cluster size distribution has been demonstrated to accurately describe
the percolation threshold consistent with results from fractal dimension
analysis as well as cluster size distribution and spanning probability and
thus is used herein for this study [19,20,31]. P(n) has the added benefit
of being insensitive to system size. The purpose of our work was to lo-
cate the threshold densities for physical and hydrogen bonded percola-
tion. Hence not all the state points were analyzed for P(n), and along
some isotherms only state points near the expected threshold density
were simulated for cluster formation. However, diffusion coefficients
and hydrogen bond numbers were obtained for the entire range of
densities.
The SPC/E model of water was used as it presents an accurate pair
potential for aqueous solutions ranging from the ambient liquid to the
supercritical phase and is found to accurately reproduce many of the
properties of water at elevated temperatures and pressures, including
dielectric constant and liquid-vapor coexistence properties [35–42].
The parametrization of the equation of state from the Helmholtz energy
Fig. 1. A schematic picture of a physical cluster. In a physical cluster the distance between for the SPC/E water model allows us to accurately plot the coexistence
the oxygen atoms of two water molecules is smaller than 3.5 Å. Molecules in a physical
cluster are predominantly H-bonded, but some are not. Molecules are hydrogen bonded
lines and spinodals on the P-ρ and P-T phase diagrams [42]. Moreover
(shown by dotted lines) if the closest pair distance between the oxygen and hydrogen the simulation results from any potential can be scaled to those of real
atoms is smaller than 2.5 Å. water through the principle of corresponding states, making the choice
 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213 3

of model often a choice of practicality. Although there has been some 3. Results and discussion
discrepancy in earlier studies as to the values of critical point parame-
ters provided by the SPC/E formulation, recent detailed analysis places 3.1. Largest cluster size
the critical point for the SPC/E water at 640 K and 0.28 g cm−3, which
corresponds to a pressure of 16.4 MPa [42]. Fig. 2 shows the largest cluster size observed at the simulated densi-
An NVT ensemble was used, with the number of particles, simula- ties for the systems with 500 molecules. Fig. 2a) shows the largest phys-
tion cell volume and temperature remaining constant. Periodic ically bound clusters and b) shows the largest hydrogen bonded
boundary conditions were used and temperature was maintained clusters. As expected, at higher temperatures clusters tend to be smaller
by the Nose-Hoover thermostat [43] while the intramolecular due to the increased thermal energies of the molecules. However signif-
motions were constrained using the SHAKE algorithm [44]. The icantly large clusters can be seen even at 800 K. At the lowest density of
equations of motion were integrated with the Verlet algorithm 0.1795 g cm−3, physically bound cluster sizes ranged from 121 water
with a 1 fs time-step [45]. Long range Coulombic interactions were molecules at 800 K to 215 water molecules at 650 K. These sizes grew
handled via Ewald summation [46]. Intermediate averaging was steadily with increasing density and converge to approach a physically
performed every 1 ps, and after the equilibration period the simula- bound cluster consisting of all 500 waters, reaching 468 molecules for
tions were run for a 1 ns trajectory. 650 K and 448 molecules at 800 K. The difference in physically bound
The simulations were performed with the MDynaMix MD simula- cluster sizes with respect to temperature diminishes as density in-
tion code [47]. All simulations were run on the Shared Hierarchy Aca- creases and density becomes the major factor in physical cluster forma-
demic Resource Computing Network (SHARCNET), a consortium of tion as observed in Fig. 2a). At higher supercritical densities, the
Ontario universities and colleges operating a network of high perfor- proximity of the molecules results in larger clusters, with even those
mance computer clusters. at high temperatures, being only slightly smaller than their lower

a)

500
450
400
350
Largest Cluster
(N Waters)

300
250
200
150
100
50
0
0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50
Density
(g cm-3)
Physical Clusters 650K Physical Clusters 700K
Physical Clusters 750K Physical Clusters 800K

b)

500
450
400
350
Largest Cluster
(N Waters)

300
250
200
150
100
50
0
0.15 0.25 0.35 0.45 0.55 0.65 0.75
Density
(g cm-3)
Hydrogen Bonded Clusters 650K Hydrogen Bonded Clusters 700K
Hydrogen Bonded Clusters 750K Hydrogen Bonded Clusters 800K

Fig. 2. Largest cluster formed vs. density for the system with 500 water molecules. Panel a) shows physically bound clusters and b) shows hydrogen bonded clusters. 650 K is shown by
circles ●, 700 K by ×, 750 K by squares ■, and 800 K is shown by triangles ▲. The dashed and dotted lines are a guide for the eye only.
4 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213

temperature counterparts. There was only a minor difference of 20 mol- with a largest physical cluster of 770 molecules at 0.1795 g cm−3 and
ecules at a density of 0.4637 g cm−3, as compared to 94 molecules at 2297 molecules at 0.391 g cm−3 observed at 650 K. Although the clus-
0.1795 g cm−3, between 650 and 800 K. ters are larger, they consist of a smaller percentage of the total waters
The same trend was observed for hydrogen bonded clusters shown available in the simulation cell. For example, at 650 K and
in Fig. 2 b) although the largest cluster size is markedly smaller. At the 0.1795 g cm−3, 72% of 500 water molecules were incorporated into
lowest density of 0.1795 g cm−3 a maximum cluster size of 120 mole- the physical cluster while with 3000 waters this number dropped to
cules was observed for 650 K and only 49 waters at 800 K. The steric re- 26%. This difference diminishes at higher density with 94% of the avail-
quirements for molecules to align along the H\\O bond necessary to able molecules forming the physical cluster when 500 molecules are
form a hydrogen bond restricts the clusters from growing as large as simulated at 0.4637 g cm−3, and 76% at 0.3291 g cm−3 with 3000 sim-
in the case of physical kinetically formed clusters. ulated waters molecules.
Over the range of densities analyzed for hydrogen bonded clusters,
the convergence seen in Fig. 2a) is absent and the largest clusters re- 3.2. Probability of the largest cluster
main more differentiated in size. At the higher density of
0.4936 g cm−3 389 water molecules were observed in the largest cluster Fig. 3 shows the probability of the formation of the largest observed
at 650 K. To reach sizes comparable to physical clusters at 800 K, density clusters for the systems with 500 water molecules. The probability was
had to increase significantly to 0.7479 g cm−3 before a cluster size of calculated with Eq. (1b). Fig. 3a) shows those for the physically bound
464 molecules was observed. clusters and b) shows the hydrogen bonded clusters. The largest cluster
The largest cluster size for the systems with 3000 water molecules formed is taken from Fig. 2 and thus is the largest clusters observed at
follows the same trend as in Fig. 2. The cluster sizes are much larger different densities, with the smaller clusters being at lower densities

a)

20
18
16
14
Probability

12
(10-7)

10
8
6
4
2
0
100 150 200 250 300 350 400 450 500
Largest Cluster
Physical Clusters 650K (N Waters) Physical Clusters 700K
Physical Clusters 750K Physical Clusters 800K

b)

10
9
8
7
Probability

6
(10-7)

5
4
3
2
1
0
0 100 200 300 400 500
Largest Cluster
(N Waters)
Hydrogen Bonded Clusters 650K Hydrogen Bonded Clusters 700K
Hydrogen Bonded Clusters 750K Hydrogen Bonded Clusters 800K

Fig. 3. Probability of the formation of the largest cluster with 500 simulated water molecules. a) Physically bound clusters, b) hydrogen bonded clusters. 650 K is shown by circles ●, 700 K
by ×, 750 K by squares ■, and 800 K is shown by triangles ▲.
 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213 5

than the largest. The probability of finding the largest cluster for the sys- bonded clusters, the average number of hydrogen bonds per molecule
tems with 500 water molecules is on the order of 10−7. Probability in- is estimated to be 1.55 [13].
creases with density, and larger clusters are more likely in dense
systems than smaller clusters in the lower density systems. 3.4. Percolation threshold densities
In Fig. 2a) and b), the probability of cluster formation is on the same
order of magnitude for both physically bound and hydrogen bonded The percolation threshold density was identified by the cluster size
clusters, the largest physically bound clusters being formed with a distribution, P(n), as discussed in Section 2. Fig. 5 shows the probability
slightly greater probability than their hydrogen bonded counterparts. distribution for the 700 K isotherm. Physical clusters for the systems
In all cases excluding the outliers on the graph, probability of forming with 500 molecules are shown in a), with 3000 molecules in b) and
the largest cluster is invariant with temperature and is only dependant hydrogen bonded clusters are shown in c) and d) for the systems with
on water density. The systems with 3000 molecules show the same 500 and 3000 water molecules, respectively. Only a selection of six
trend. In all cases with both system sizes, probability is independent of distributions is shown around the threshold densities for 700 K and
temperature and remains only a function of water density, with larger the results are typical for the other 650, 750 and 800 K simulations.
clusters more probable in systems with higher density, than the largest Our study differs from previous works in that the mean of the probabil-
although relatively smaller clusters forming in lower density systems. ity of a cluster of size n was plotted. Also we have used a greater number
of water molecules, 500 and 3000, and have analyzed 20,000 configura-
tions yielding better statistical averages than previously reported.
3.3. Number of hydrogen bonds [20,21] The system is said to be percolating when the critical exponent
is equal to 2.19 and a log-log plot of P(n) should have the mean cluster
The total number of hydrogen bonds on each water molecule for the probabilities fall on the line. An infinite cluster in at least one dimension
entire system is shown in Fig. 4. The number of hydrogen bonds was cal- is anticipated when truncation effects due to the finite size of the simu-
culated by integrating the oxygen to hydrogen radial distribution func- lation cell, cause a peak in the distribution [2,3,18–21,31]. We define the
tions (RDF) to the first minimum at 2.5 Å although the position of this system as percolating when the probability distribution falls on the line
minimum shifted slightly at higher temperatures. As expected the defined by the critical exponent and first deviates above the line even
total number of hydrogen bonds is smallest for the highest temperature slightly, indicating the first appearance of infinite spanning clusters.
of 800 K and increased as temperature decreased. At the lowest density Our analysis of the 700 K isotherm gives results for H-bonded clusters
of 0.1795 g cm−3, 1.22 hydrogen bonds per molecule were observed consistent with those obtained by Partay et al. through the analysis of
and this increased to maximum of 1.9 at 650 K. These numbers in- fractal dimension and cluster weight average [21]. P(n) was found to
creased steadily and converged to 1.9 for all temperatures when a den- be unaffected by system size for physical clusters. Threshold densities
sity of 1.0471 g cm−3 was reached. This suggests that at lower densities for hydrogen bound clusters were only 0.015 g cm−3 higher with
dimers and trimers exist and any spanning clusters are formed by chain 3000 molecules compared with 500, at 650 and 700 K. One may treat
like arrangements of water molecules that span the simulation cell ei- this small discrepancy as an estimation error as both distributions in
ther in one or two dimensions, representative by two bonds per mole- Fig. 5c) and d) appear very similar. From Fig. 5a) and b), one can see
cule. At the high-density limit occurring at high pressures of 896.6 and that all six of the systems are close to the threshold density of
1263 MPa at 650 K and 800 K respectively, the SCW is in a more liquid 0.2692 g cm−3, evident by when truncation effects first begin to appear,
like state and the average water molecule in the system has just under denoted by an asterisk in the figure. This threshold was identical for
2 hydrogen bonded neighbours. Our percolation threshold as discussed both system sizes. Fig. 5c) and d) shows similar results although the
below is in fairly close agreement with the previous work of Kalinichev threshold density for hydrogen bonded clusters is greater than those
and Churakov in which at the percolation threshold for hydrogen for physical clusters. Here the threshold appears to be at

2.0

1.9

1.8

1.7
Total Hydrogen Bonds

1.6

1.5

1.4

1.3

1.2

1.1

1.0
0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85 0.95 1.05
Density (g cm-3)

650 K 700 K 750 K 800 K

Fig. 4. Total number of hydrogen bonds per water molecule for the entire system as a function of density. 650 K is shown by circles ●, 700 K by ×, 750 K by squares ■, and 800 K is shown by
triangles ▲. The dashed and dotted lines are a guide for the eye only.
6 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213

0.4338 g cm−3 with 500 waters and 0.4487 g cm−3 for 3000 waters. percolating clusters first appear between 0.3739 g cm−3 for 500 waters
Again, this is a very small difference and may simply be due to the am- and 0.3889 g cm−3 for 3000 waters at 650 K, and at 700 K appear at
biguity introduced by inspecting P(n). 0.4338 and 0.4487 g cm−3. Above this the results were identical and
Table 1 shows the percolation threshold densities for all the systems at 800 K hydrogen bonded percolating clusters first appear at
studied. The threshold densities for both physical and hydrogen bound 0.5086 g cm−3.
clusters increased with system temperature. In all cases the threshold
density was lower for the physically bound clusters, and an exponential 3.5. Phase diagrams
relation was found between threshold pressure and both density and
temperature and is given below. Percolating physical clusters appear The identification of the percolation threshold on thermodynamic
between the range of 0.2393 to 0.3141 g cm−3 and hydrogen bonded phase diagrams provide insight into the locus of the densities and

a)

b)

Fig. 5. Probability distributions P(n) for the 700 K isotherm. Panel a) shows P(n) for physically bound clusters and b) shows hydrogen bonded clusters, both for the systems with 500 water
molecules. Panel c) shows P(n) for physically bound clusters and d) shows hydrogen bonded clusters both for the system with 3000 molecules. The critical exponent of 2.19 is plotted as a
straight line and the systems where infinite clusters are first observed are denoted by an asterisk.
 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213 7

c)

d)

Fig. 5 (continued).

pressures of the transitions. Fig. 6 shows the four supercritical isotherms obtained here [2,21]. The transition line for physically bound clusters
simulated, as well as the physical and hydrogen bonded percolation occurs at lower densities and pressures than that of hydrogen bound
transitions on the P-ρ diagram, along with the coexistence curve and clusters denoted by stars and diamonds respectively. The line for phys-
spinodals. The data is plotted in reduced units using the critical point pa- ically bound percolating clusters occurs at around the critical density
rameters for SPC/E water [42]. It can be seen that both the physical and whereas that for hydrogen bonded clusters at a higher density than
hydrogen bonded percolation transition occur on separate lines and in- the critical density.
crease in pressure and density when temperature is increased. The hy- Fig. 7 shows the percolation transitions on the P-T plane and in-
drogen bonded percolation transition densities of 0.404 and cludes the spinodals and an extension of the coexistence line into the
0.410 g cm−3 obtained by Partay for the 700 K isotherm transitions supercritical region. The Clapeyron equation was fitted to the coexis-
are shown and are in very close agreement with the transition density tence data from the SPC/E water model in the subcritical phase and
8 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213

Table 1
Percolation threshold densities and pressures for both systems of 500 and 3000 water molecules.

500 waters

Physical clusters H-bonded clusters

Temp (K) Threshold density Threshold pressure (MPa) Temp (K) Threshold density Threshold pressure (MPa)
−3 −3 −3 −3
N (Å ) ρ (g cm ) N (Å ) ρ (g cm )

650 0.0080 0.2393 18.62 650 0.0125 0.3739 20.88

700 0.0090 0.2692 30.22 700 0.0145 0.4338 44.07
750 0.0100 0.2992 45.08 750 0.016 0.4787 75.88
800 0.0105 0.3141 60.64 800 0.017 0.5086 114.3

3000 waters

Physical clusters H-bonded clusters

Temp (K) Threshold density Threshold pressure (MPa) Temp (K) Threshold density Threshold pressure (MPa)
−3 −3 −3 −3
N (Å ) ρ (g cm ) N (Å ) ρ (g cm )

650 0.0080 0.2393 18.62 650 0.013 0.3889 21.23

700 0.0090 0.2692 30.22 700 0.015 0.4487 45.93
750 0.0100 0.2992 45.08 750 0.016 0.4787 75.88
800 0.0105 0.3141 60.64 800 0.017 0.5086 114.3

was extrapolated into the supercritical phase. the density fluctuation, accompanying the Widom line. This is still a
matter of some speculation, however. Our line for the compressibility
1 maximum is calculated from an analytical equation of state for SPC/E
ln P r ¼ −a ð2Þ
Tr water [42], fitted to the simulated PVT database in the wide range of
state points. It is important to note, that the accuracy of the simulations
Eq. (2) is the Clapeyron equation where Pr is the pressure in reduced near the critical point is affected by the density fluctuations being at a
units and Tr is system temperature in reduced units. The a term corre- maximum and exceeding the size of the simulation cell. It is thus diffi-
sponds to the slope of the line and was found with a linear fit to the co- cult to identify with certainty, via simulations, the percolation thresh-
existence data obtained by Plugatyr [42]. The physical and hydrogen olds in this vicinity.
bonded percolation transitions are close together just above the critical Conceivably, these percolation transitions may be related to the
point and diverge as temperature increases. Contrary to previous analy- pseudo spinodals arising from the extensions of the liquid-vapor
sis [20] we found that the percolation transition of hydrogen bonded spinodals, marking a pseudo phase boundary between kinetically labile
clusters does not coincide with the extension of the vapor-liquid coex- physical clusters and stable hydrogen bonded cluster formation. There
istence line. The extension of the coexistence line falls in between the is currently no comprehensive analysis of the free energy surface (e.g.
physical and hydrogen bonded transition lines and may be correlated of the higher order derivatives for the pressure) for the SPC/E water
to the conditions that mark the transition between the physical and hy- model, and we leave this theoretical problem for a future project. It is
drogen bonded cluster types themselves, rather than the threshold for certainly worth developing a sound theoretical formulation for a contin-
hydrogen bonded clusters alone. In the P-T plane shown in Fig. 7, the uous extension of the thermodynamic (or kinetic) [48,49] spinodals
percolation transition lines appear to converge roughly on the com- into a region of supercritical states. For empirical fits to the transition
pressibility maximum and ultimately on the critical point itself. The for- points for two clustering regimes, see below. From a phenomenological
mation of infinite clusters near the critical point is likely correlated to point of view, the thermodynamic response functions tend to diverge,

Fig. 6. Phase diagram in P-ρ plane showing the physical and hydrogen bonded threshold densities. 650 K is shown by circles ●, 700 K by ×, 750 K by squares ■, and 800 K is shown by
triangles ▲. The threshold for physically bonded clusters is shown by stars and the threshold for hydrogen bonded clusters is denoted by diamonds along their respective isotherms.
Partay's thresholds are shown by the bolded x's along the 700 K isotherm. The coexistence curve is plotted as a solid black line with the spinodals in black dashing. The compressibility
maximum is shown in black dotted dashing.
 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213 9

Fig. 7. Phase diagram in P-T plane showing the extension of the vapor-liquid coexistence line and physical and hydrogen bonded threshold densities. The threshold for physically bonded
clusters is shown by stars and the threshold for hydrogen bonded clusters is denoted by diamonds. The coexistence curve is plotted as a solid black line with its extension in gray. The
spinodals in black dashing and the compressibility maximum is shown in black dotted dashing.

as the spinodal is approached indicating the limit of the mechanical sta- justification of the “pseudo spinodals” in the supercritical phase exists
bility of the system. However, above the critical point, the loci of the re- (e.g. “pseudo spinodals” being the limits of mechanical or kinetic stabil-
sponse function extrema seem to exhibit density/pressure dependence ity of various cluster arrangements). The topic of Widom lines and the
quite different from that of the percolation lines which makes such an maxima of the thermodynamic response functions is still a matter of de-
analysis disconcerting [23,24]. We may comment that the thermody- bate in the recent literature. It would be interesting indeed if some gen-
namic properties for supercritical fluids continue to be a focus of consid- uine correlation can be found between percolation and these other
erable speculation and argument [23,24]. physical properties of water.

3.6. Empirical fits

a)
An exponential function of the form

y ¼ a ebx ð3Þ

was fitted to the percolation lines for physical and hydrogen bound clus-
ters in both the P-ρ and P-T planes. Table 2 shows the parameters a and
b, as well as the R2 values for all four sets of data. In Eq. (3), y is the re-
duced pressure and x is either reduced density or reduced temperature.
The exponential equation fits the data quite well, being valid from
the critical point to at least 800 K. Analytical fits to thresholds densities
and pressures are shown in Fig. 8, with P-ρ plot given in a) and P-T plot
given in b). This is the first time to our knowledge that these two differ-
ent percolation transition lines have been identified in such a detailed
manner. Beyond the benefit of being able to predict whether there
will be physical (or hydrogen bonded) connectivity in the SCW me-
dium, which has practical implications in charge transfer (e.g. electro- b)
chemical corrosion) or solvation phenomena, these lines can be
plotted on any model phase diagram which will hopefully aid in further
theoretical analysis.
Exploring the extensions of the so called “pseudo spinodals” near the
critical point, such as by using Taylor expansions of the partial deriva-
tives of the thermodynamic state functions, may prove useful [48,49].
We leave it to a future work. Currently no meaningful physical

Table 2
Empirical fitting parameters for percolation thresholds for physical and hydrogen bonded
clusters.

aeb x Pr vs ρr Pr vs Tr

Physical clusters H-bonded clusters Physical clusters H-bonded clusters

a 2.6055 × 10−2 1.0169 × 10−2 1.0862 × 10−2 2.5340 × 10−3

Fig. 8. Empirical exponential fits to percolation thresholds in a) P-ρ plane and b) in P-T
b 4.3358 3.5391 4.6787 6.3513
plane. The physical and hydrogen bonded percolation thresholds are shown by stars and
R2 0.9991 0.9998 0.9985 0.9972
diamonds respectively.
10 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213

percolation transition loci for the two cluster types. Empirical fits were
obtained for the transition lines and provide a quick means by which
to differentiate the clustering regimes for corrosion, synthesis and
other applications. It can be seen on the P-T plot that the percolation
lines converge approximately on the compressibility maximum. These
lines may arguably be related to the supercritical mesophase that
marks a dynamical clustering region that separates the gas and liquid
like states of a supercritical fluid. The activation volume of diffusion
reaches a maximum near the threshold pressures where water mole-
cules are connected via percolating network of hydrogen bonds. Future
work may benefit from further examinations of the thermodynamic
response functions and identification of the “pseudo spinodals” in the
supercritical phase.
Fig. 9. Activation volumes of diffusion as a function of reduced pressure for all four
simulated isotherms.
Author contributions
3.7. Activation volumes of diffusion
DTK and IMS conceived and performed the simulations, and co-
Cluster formation is the dynamical process, and it is ultimately wrote the manuscript.
related to the translational mobility of water molecules [50]. We have
analyzed the self-diffusion coefficient, D, of water molecules along the
Acknowledgements
four chosen isotherms. The activation volumes of diffusion were calcu-
lated for each isotherm using the expression
The authors are grateful for the financial support of the Generation
IV Energy Technologies Program, funded by Natural Resources Canada,
V d ¼ −RT ½∂ ln D=∂P T ; ð4Þ
Canadian Nuclear Laboratories, and Natural Sciences and Engineering
Research Council of Canada [Grant Number: RGPIN-2015-05321]. We
where R is the gas constant, T is the temperature and P is the pressure
would also like to thank Professor Bill Atkinson for valuable discussion.
[51]. Fig. 9 shows Vd for all isotherms, as a function of reduced pressure.
Vd is larger at higher temperatures and increases relatively rapidly
References
until the pressure for the formation of hydrogen bonded clusters is
reached, where it begins to increase slower and changes very little in [1] N. Yoshii, S. Okazaki, A large-scale and long-time molecular dynamics study of su-
magnitude beyond these pressures. At 650 K Vd increases from percritical Lennard-Jones fluid. An analysis of high temperataure clusters, J. Chem.
Phys. 107 (1997) 2020–2033.
0.084 m3 mol−1 at a reduced pressure of 1.073 and reaches [2] L. Partay, P. Jedlovszky, I. Brovchenko, A. Oleinikova, Formation of mesoscopic water
0.088 m3 mol−1 at reduced pressure of 1.270. Vd at 700 K begins at networks in aquous systems, Phys. Chem. Chem. Phys. 9 (2007) 1341–1346.
0.090 m3 mol−1 and reaches 0.094 m3 mol−1 at reduced pressure of [3] M. Bernabei, M.A. Ricci, Percolation and clustering in supercritical aqueous fluids, J.
Phys. Condens. Matter 20 (2008), 494208.
2.681. This rapid increase up to the hydrogen bond threshold pressure [4] P. Postorino, R.H. Tromp, M.A. Ricci, A.K. Soper, G.W. Neilson, The interatomic struc-
was also seen for 750 and 800 K isotherms. At 800 K Vd increased ture of water at supercritical temperatures, Nature 366 (1993) 668–670.
form 1.02 m3 mol−1 at a reduced pressure of 2.417 g cm−3 and reaches [5] Y.E. Gorbaty, A.G. Kalinichev, Hydrogen bonding in supercritical water. 1. Experi-
mental results, J. Phys. Chem. 99 (1995) 5336–5340.
0.107 m3 mol−1 at a reduced pressure of 6.954. Beyond these pressure [6] Y.E. Gorbaty, R.B. Gupta, The structural features of liquid and supercritical water, Ind.
and corresponding densities, Vd increases only marginally even at Eng. Chem. Res. 37 (1998) 3026–3035.
reduced pressures in excess of 55 times the critical pressure. [7] P. Jedlovszky, J.P. Brodholt, F. Bruni, M.A. Ricci, A.K. Soper, R. Vallauri, Analysis of the
hydrogen-bonded structure of water from ambient to supercritical conditions, J.
This may indicate some sort of transition in the change in volume Chem. Phys. 108 (1998) 8528–8540.
upon diffusion when water begins to form a spanning cluster. Near [8] C.J. Sahle, C. Sternemann, C. Schmidt, S. Lehtola, S. Jahn, L. Simonelli, S. Huotari, M.
the critical point Vd increases rapidly with pressure but upon the forma- Hakala, T. Pylkkänen, A. Nyrow, K. Mende, M. Tolan, K. Hämäläinen, M. Wilke, Mi-
croscopic structure of water at elevated pressures and temperatures, PNAS 110
tion of infinitely interconnected hydrogen bonded systems, this activa-
(2013) 6301–6306.
tion volume stops increasing and remains almost constant as pressure is [9] Q. Sun, Q. Wang, Hydrogen bonded networks in supercritical water, J. Phys. Chem. B
increased over an order of magnitude. When water molecules are in a 118 (2014) 11253–11258.
hydrogen bonded cluster, they become “locked” into the bond network. [10] P.H.-L. Sit, C. Bellin, B. Barbiellini, D. Testemale, J.-L. Hazemann, T. Buslaps, N.
Marzari, A. Shukla, Hydrogen bonding and coordination in normal and supercritical
Here Vd nears its maximum corresponding to the volume change for water from X-ray inelastic scattering, Phys. Rev. B 76 (2007).
entering, leaving or skipping locations in the bonded network. There is [11] E. Johansson, K. Blton, P. Ahlstrom, Simulations of vapor water clusters at vapor-
little influence on this by an increase in pressure as this only serves to liquid equilibrium, J. Chem. Phys. 123 (2005), 024504.
[12] A.G. Kalinichev, J.D. Bass, Hydrogen bonding in supercritical water. 2. Computer sim-
force more molecules in to the structural arrangement, eventually ulations, J. Phys. Chem. A 101 (1997) 9720–9727.
reaching a tetrahedrally coordinated system. [13] A.G. Kalinichev, S.V. Churakov, Thermodynamics and structure of molecular clusters
in supercritical water, Fluid Phase Equilib. 183–184 (2001) 271–278.
[14] J. Škvor, I. Nezbeda, Percolation line, response functions, and Voronoi polyhedra
4. Conclusions analysis in supercritical water, Condens. Matter Phys. 15 (2012) 1–8.
[15] I. Skarmoutsos, J. Samios, Local density inhomogeneities and dynamics in supercrit-
The transition from physically bound to hydrogen bonded clusters ical water: a molecular dynamics simulation approach, J. Phys. Chem. B 110 (2006)
21931–21937.
was examined along the 650, 700, 750, and 800 K supercritical water [16] I. Skarmoutsos, E. Guardia, J. Samios, Local structural fluctuations, hydrogen bonding
isotherms. Physically bound clusters form more readily than hydrogen and structural transitions in supercritical water, J. Supercrit. Fluids 130 (2017)
bonded clusters, and the probability of formation of the largest cluster 156–164.
[17] A.G. Kalinichev, S.V. Churakov, Size and topology of molecular clusters in supercrit-
is insensitive to temperature, being determined by water density. Near
ical water: a molecular dynamics simulation, Chem. Phys. Lett. 302 (1999) 411–417.
the critical point, infinite clusters span the simulation cell in one or [18] M. Bernabei, A. Botti, F. Bruni, M.A. RIcci, A.K. Soper, Percolation and three-
two dimensions. At extreme pressures 3D spanning clusters are formed. dimensional structure of supercritical water, Phys. Rev. E 78 (2008), 021505.
The lines of percolation transitions for both types of clusters do not [19] X. Campi, H. Krivine, N. Sator, Percolation line of stable clusters in supercritical
fluids, Phys. A 296 (2001) 24–30.
coincide with the extension of the vapor liquid coexistence line. The [20] L. Partay, P. Jedlovsky, Line of percolation in supercritical water, J. Chem. Phys. 123
extension of the coexistence line lies approximately in between the (024502) (2005).
 D.T. Kallikragas, I.M. Svishchev / Journal of Molecular Liquids 290 (2019) 111213 11

[21] L. Partay, P. Jedlovszky, I. Brovchenko, A. Oleinikova, Percolation transition in super- [37] A.A. Chialvo, P.T. Cummings, Microstructure of ambient and supercritical water. Di-
critical water: a Monte Carlo simulation study, J. Phys. Chem. B 111 (2007) rect comparison between simulation and neutron scattering experiments, J. Phys.
7603–7609. Chem. 100 (1996) 1309–1316.
[22] L.V. Woodcock, Observations of a thermodynamic liquid–gas critical coexistence [38] J.R. Errington, K. Kiyohara, K.E. Gubbins, A.Z. Panagiotopoulos, Monte Carlo simula-
line and supercritical fluid phase bounds from percolation transition loci, Fluid tion of high-pressure phase equilibria in aqueous systems, Fluid Phase Equilib.
Phase Equilib. 351 (2013) 25–33. 150–151 (1998) 33–40.
[23] P. Gallo, D. Corradini, M. Rovere, Widom line and dynamical crossovers as routes to [39] T.M. Hayward, I.M. Svishchev, Equations of state and phase coexistence properties
understand supercritical water, Nat. Commun. 5 (2014) 1–6. for simulated water, Fluid Phase Equilib. 182 (2001) 65–73.
[24] M. Raju, D.T. Banuti, P.C. Ma, M. Ihme, Widom lines in binary mixtures of supercrit- [40] J. Alejandre, D.J. Tildesley, Molecular dynamics simulation of the orthobaric densities
ical fluids, Sci. Rep. 7 (2017) 3027. and surface tension of water, J. Chem. Phys. 102 (1995) 4574–4583.
[25] A.R. Imre, U.K. Deiters, T. Kraska, I. Tiselj, The pseudocritical regions for supercritical [41] J.R. Errington, A.Z. Panagiotopoulos, A fixed point charge model for water optimized
water, Nucl. Eng. Des. 252 (2012) 179–183. to the vapor–liquid coexistence properties, J. Phys. Chem. B 102 (1998) 7470–7475.
[26] D. Corradini, M. Rovere, P. Gallo, The Widom line and dynamical crossover in super- [42] A. Plugatyr, I.M. Svishchev, Accurate thermodynamic and dielectric equations of state
critical water: popular water models versus experiments, J. Chem. Phys. 143 (2015). for high-temperature simulated water, Fluid Phase Equilib. 227 (2009) 145–151.
[27] A.G. Kalinichev, Universality of hydrogen bond distributions in liquid and supercrit- [43] S. Nose, A molecular dynamics method for simulations in the canonical ensemble,
ical water, J. Mol. Liq. 241 (2017) 1038–1043. Mol. Phys. 100 (2002) 191–198.
[28] R. Kumar, J.R. Schmidt, J.L. Skinner, Hydrogen Bonding Definitions and Dynamics in [44] J.P. Ryckaert, G. Ciccotti, H.J.C. Berendsen, Numerical integration of the Cartesian
Liquid Water, 2007 126. equations of motion of a system with constraints: molecular dynamics of n-
[29] A.A. Chialvo, P.T. Cummings, Hydrogen bonding in supercritical water, J. Chem. Phys. alkanes, J. Comput. Phys. 23 (1977) 327–341.
(1994) 101. [45] L. Verlet, Computer “experiments” on classical fluids. I. Thermodynamical properties
[30] M. Mezei, D.L. Beveridge, Theoretical studies of hydrogen bonding in liquid water of Lennard-Jones molecules, Phys. Rev. 159 (1967) 98–103.
and dilute aqueous solutions, J. Chem. Phys. 74 (1981). [46] M.P. Allen, D.J. Tildesley, Computer Simulations of Liquids, Clarendon Press, New
[31] A. Oleinikova, I. Brovchenko, A. Geiger, B. Guillot, Percolation of water in aqueous so- York, 1989.
lution and liquid–liquid immiscibility, J. Chem. Phys. 117 (2002). [47] A.P. Lyubartsev, A. Laaksonen, M.DynaMix — a scalable portable parallel MD simula-
[32] L.A. Pugnaloni, C.M. Carlevaro, M.G. Valluzzi, F. Vericat, Continuum percolation of tion package for arbitrary molecular mixtures, Comput. Phys. Commun. 128 (2000)
long lifespan clusters in a simple fluid, J. Chem. Phys. 129 (2008), 064510. 565–589.
[33] D.M. Heyes, L.V. Woodcock, Critical and supercritical properties of Lennard–Jones [48] S.B. Kiselev, I.G. Kostyukova, Spinodal and kinetic boundary of metastable region, J.
fluids, Fluid Phase Equilib. 356 (2013) 301–308. Chem. Phys. 98 (1993) 6455–6464.
[34] D. Stauffer, A. Aharony, Introduction to Percolation Theory, Taylor & Francis, [49] P.G. Debenedetti, Metastable Liquids Concepts and Principles, Princeton University
London; Bristol, PA, 1994. Press, New Jersey, 1996.
[35] H.J.C. Berendsen, J.R. Grigera, T.P. Straatsma, The missing term in effective pair po- [50] A.G. Kalinichev, Molecular-dynamics and self-diffusion in supercritical water, Ber.
tentials, J. Phys. Chem. 91 (1987) 6269–6271. Bunsenges. Phys. Chem. 97 (1993) 872–876.
[36] Y. Guissan, B. Guillot, A computer simulation study of the liquid–vapor coexistence [51] K. Krynicki, C.D. Green, D.W. Sawyer, Pressure and temperature dependence of self
curve of water, J. Chem. Phys. 98 (1993) 8221–8235. diffusion in water, Faraday Discuss. 66 (1978) 199–208.