cond-mat0703372/pmw.tex
1: \documentclass[twocolumn,showpacs,superscriptaddress]{revtex4}
2: \usepackage{graphicx}
3: \usepackage{amsmath}
4: 
5: \begin{document}
6: 
7: \title{Gas-Liquid Coexistence in the Primitive Model for Water}
8: \author{Flavio Romano\footnote{Author to which correspondence should be 
9: addressed; electronic address {\tt flavio.romano@gmail.com}}}
10: \affiliation{Dipartimento di Fisica, Universit\`a di Roma {\em La Sapienza}, 
11: Piazzale A. Moro 2, 00185 Roma, Italy}
12: \author{Piero Tartaglia}
13: \affiliation{Dipartimento di Fisica and INFM-CNR-SMC, Universit\`a di Roma 
14: {\em La Sapienza}, Piazzale A. Moro 2, 00185 Roma, Italy} 
15: \author{Francesco Sciortino}
16: \affiliation{Dipartimento di Fisica and INFM-CNR-SOFT, Universit\`a di Roma 
17: {\em La Sapienza}, Piazzale A. Moro 2, 00185 Roma, Italy}
18: %\ead{francesco.sciortino@phys.uniroma1.infn.it}
19: 
20: \begin{abstract}
21: We evaluate the location of the gas-liquid coexistence line and of the 
22: associated critical point for the primitive model for water (PMW), introduced 
23: by Kolafa and Nezbeda [J. Kolafa and I. Nezbeda, Mol. Phys. {\bf 61}, 161 
24: (1987)]. Besides being a simple model for a molecular network forming liquid, 
25: the PMW is representative of patchy proteins and novel colloidal particles 
26: interacting with localized directional short-range attractions. We show that 
27: the gas-liquid phase separation is metastable, i.e. it takes place in the 
28: region of the phase diagram where the crystal phase is thermodynamically 
29: favored, as in the case of particles interacting via short-range attractive 
30: spherical potentials. Differently from spherical potentials, we do not observe 
31: crystallization close to the critical point. The region of gas-liquid 
32: instability of this patchy model is significantly reduced as compared to 
33: equivalent models of spherically interacting particles, confirming the 
34: possibility of observing kinetic arrest in an homogeneous sample driven by 
35: bonding as opposed to packing.
36: \end{abstract}
37: 
38: %61.20.Qg Structure of associated liquids: electrolytes, molten salts, etc.
39: %81.16.Dn Self-assembly
40: %82.70.Dd (colloids) , and 82.70.Gg (gels) 
41: %61.20.Ja Computer simulation of liquid structure
42: %64.60.Fr Equilibrium properties near critical points, critical exponents
43: %64.70.Fx Liquid-vapor transitions
44: %64.60.My Metastable phases
45: %61.20.-p Structure of liquids
46: %61.20.Gy Theory and models of liquid structure
47: %61.20.Ne Structure of simple liquids
48: \pacs{61.20.Ja,61.20.Gy,61.20.Ne}
49: 
50: \maketitle
51: 
52: \section{Introduction}
53: This article presents a detailed numerical study of the critical point and 
54: gas-liquid coexistence of a model introduced several years ago by Kolafa and 
55: Nezbeda~\cite{Kol87a} as a primitive model for water (PMW). The water molecule 
56: is described as a hard-sphere with four interaction sites, arranged on a 
57: tetrahedral geometry, which are meant to mimic the two hydrogen protons and 
58: the two oxygen lone-pairs of the water molecule. The PMW has been studied in 
59: detail in the past, since it is both a valid candidate for testing theories of 
60: bond association~\cite{Wer84a,Wer84b,Gho93a,Sea96a,Dud98a,Pee03a,Kal03a} as 
61: well as a model able to reproduce the thermodynamic anomalies of 
62: water~\cite{Kol87a,Nez89a,Nez90a,Veg98a,pwmnoi}. The Wertheim 
63: theory~\cite{Wer84a,Wer84b} has been carefully compared to numerical studies, 
64: suggesting a good agreement between theoretical predictions and numerical data 
65: in the temperature $T$ region where no significant ring formation is 
66: observed~\cite{Veg98a,pwmnoi}. More recently, the slow dynamics of this model 
67: has been studied using a newly developed code for event-driven dynamics of 
68: patchy particles~\cite{pwmnoi}. It has been shown that, at low density, there 
69: are indications of a gas-liquid phase separation. At intermediate density, the 
70: system can be cooled down to the smallest temperatures at which equilibration 
71: is feasible with present day computational facilities without any signature of 
72: phase separation. In this region, dynamics progressively slows down following 
73: an Arrhenius law, consistently with the expected behavior of strong 
74: network-forming liquids.
75: 
76: This primitive model, besides its interest as an elementary model for water, 
77: is also representative of a larger class of models which are nowadays receiving 
78: considerable interest, namely models of particles interacting with localized 
79: directional short-range attractions. This class of models includes, apart from 
80: network forming molecular systems, also 
81: proteins~\cite{Lom99a,Sea99c,Ker03a,doye} and newly designed colloidal 
82: particles~\cite{Man03a}. Recently, the phase diagram of patchy particles has 
83: been studied~\cite{bianchiprl} in the attempt to estimate the role of the
84: surface patchiness in dictating their dynamic and thermodynamic behavior. It 
85: has been suggested that the number of possible bonds $M$, as opposed to the 
86: fraction of surface with attractive interactions, is the key ingredient in 
87: determining the width of the unstable region of the phase 
88: diagram~\cite{Zac05a,Sci04bCPC,zaccalungo}. A study of the evolution of the 
89: critical point on decreasing $M$ shows a clear monotonic 
90: trend~\cite{bianchiprl} and, more importantly, in the direction of decreasing 
91: critical packing fraction. Such a trend is not observed in spherical 
92: potentials on decreasing the attraction range. Thus, in low-valence particle 
93: systems, a window of packing fraction $\phi$ values opens up in which it is 
94: possible to reach very low $T$ (and hence states with extremely long bond 
95: lifetimes) without encountering phase separation. This favors the 
96: establishment of a spanning network of long-living bonds, which in the 
97: colloidal community provides indication of gel formation but which, in the 
98: field of network forming liquids, would be rather classified as glass 
99: formation~\cite{silica}. 
100: 
101: As previously mentioned, we present here an accurate estimate of the location 
102: of the critical point for the PMW (for which $M=4$), based on finite-size 
103: scaling, and of the associated gas-liquid coexistence phase diagram. Comparing 
104: with the known fluid-crystal phase coexistence loci~\cite{Veg98a} we are able 
105: to show that the gas-liquid phase separation is metastable as compared to the 
106: crystal state. Despite its metastability, we have never observed 
107: crystallization, not even close to the critical point, an observation which 
108: could be of relevance in the case of the crystallization of patchy proteins or 
109: colloids. We also show that the number density $\rho$ of the liquid phase at 
110: coexistence is comparable to the crystal (diamond) density, much smaller than 
111: the characteristic liquid density observed in systems of particles interacting 
112: through spherically symmetric potentials. 
113: 
114: \begin{figure}[tb]
115: \includegraphics[width=8cm]{figure1.eps}
116: \caption{Pictorial representation of the model studied. Each particle is 
117: modeled as an hard-core sphere (grey large sphere of diameter $\sigma$). 
118: The four interaction sites are located in a tetrahedral arrangement. Two of 
119: the sites (the H-sites, in green) are located on the surface whereas the 
120: remaining two sites (LP, in brown) are located inside the sphere, at distance 
121: $0.45 \sigma$ from the center. Only H-sites and L-sites on different particles 
122: interact with a square well interaction of depth $u_0$ and width 
123: $\delta=0.15\sigma$.}
124: \label{fig:m2}
125: \end{figure}
126: 
127: \section{The model}
128: In the PMW, each particle is composed by a hard sphere of diameter $\sigma$ 
129: and four additional sites arranged according to a tetrahedral geometry. 
130: Two of these sites, the proton sites H, are located on the surface of the hard 
131: sphere, i.e. at distance $0.5\sigma$ from the center of the particle, while 
132: the two remaining sites, the lone-pair sites LP, are placed within the hard 
133: sphere at a distance $0.45\sigma$ from its center. Besides the hard-sphere 
134: interaction, preventing different particles to sample relative distances 
135: smaller than $\sigma$, only the H and LP sites of distinct particles interact 
136: via a square well (SW) potential $u_{SW}$ of width $\delta=0.15\sigma$ and 
137: depth $u_0$, i.e. 
138: \begin{equation}
139: u_{SW}(r)\,=\,\left\{
140: 		\begin{array}{cc}
141: 			-u_0 & r\,<\,\delta \\
142: 			0 & r\,\ge\,\delta
143: 		\end{array}
144: 	\right.
145: \end{equation}
146: where $r$ is here the distance between H and LP sites. We remark that the 
147: choice $\delta=0.15 \sigma$ guarantees that multiple bonding can not take 
148: place at the same site.
149: 
150: \section{Simulations}
151: Three types of Monte Carlo simulations (Metropolis Grand Canonical, GCMC, 
152: Umbrella Sampling Grand Canonical, US-GCMC and Gibbs Ensemble GEMC) have been 
153: performed, with the aim of locating the critical point of the model, assessing 
154: the consistency with the Ising universality class and evaluating the 
155: gas-liquid coexistence curve. Throughout the remaining sections, we use $u_0$ 
156: as energy scale, $\sigma$ as length scale and reduced units in which 
157: $k_{\rm B}=1$, thus measuring $T$ and the chemical potential $\mu$ in units of 
158: $u_0/k_{\rm B}$ and $u_0$ respectively. 
159: 
160: We define a MC step as an average of $N_{\Delta}$ attempts to translate and 
161: rotate a randomly chosen particle and an average $N_{N}$ attempts to insert or 
162: remove a particle. The translation in each direction is uniformly chosen 
163: between $\pm0.05$ and the rotations are performed around a random axis of an 
164: angle uniformly distributed between $\pm 0.1$\,rad. Unless otherwise stated, 
165: $N_{\Delta}/N_N=500$. The choice of such a large ratio between 
166: translation/rotation and insertion/deletion attempts is dictated by the 
167: necessity of ensuring a proper equilibration at fixed $N$. This is 
168: particularly important in the case of particles with short-range and highly 
169: directional interactions, since the probability of inserting a particle with 
170: the correct orientation and position for bonding is significantly reduced as 
171: compared to the case of spherical interactions. 
172: 
173: \subsection{Critical Point Estimate and Finite Size Scaling Analysis}
174: \label{sec:cp}
175: 
176: The location of the critical point has been performed through the 
177: comparison of the fluctuation distribution of the ordering operator 
178: $\mathcal{M}$ at the critical point with the universal distribution 
179: characterizing the Ising class~\cite{Wilding_96}. The ordering operator 
180: $\mathcal{M}$ of the gas-liquid transition is a linear combination 
181: $\mathcal{M}\sim \rho +s u$, where $\rho$ is the number density, 
182: $u$ is the energy density of the system, and $s$ is the field mixing parameter. 
183: Exactly at the critical point, fluctuations of $\mathcal{M}$ are found to 
184: follow a known universal distribution, i.e. apart from a scaling factor, the 
185: same that characterizes the fluctuations of the magnetization in the Ising 
186: model~\cite{Wilding_96}. 
187: Finite Size Scaling (FSS) analysis has been performed to test the appartenance 
188: of the PMW model to the three dimensional Ising class. Recent applications of 
189: this method to soft matter can be found in Ref.~\cite{Miller_03,puertas,horbach}.
190: 
191: To locate the critical point we perform, at each size, GCMC simulations, i.e. 
192: at fixed $T$, $\mu$ and volume $V$. We follow the standard procedure of tuning 
193: $T$ and $\mu$ until the simulated system shows ample density fluctuations, 
194: signaling the proximity of the critical point. Once a reasonable guess of the 
195: critical point in the $(T,\mu)$-plane has been reached, we start at least 10 
196: independent GCMC simulations to improve the statistics of the fluctuations in 
197: the number of particles $N$ in the box and of the potential energy $E$. 
198: 
199: The precise evaluation of the critical temperature and chemical potential 
200: at all system sizes is performed by a fitting procedure associated to 
201: histogram reweighting. We briefly recall~\cite{histrew} that this technique 
202: allows us to predict, from the joint distribution $P(N,E;T,\mu)$ obtained from 
203: a simulation, the distribution at a different $T'$ and $\mu'$ through the 
204: ratio of the Boltzmann factors as follows:
205: \begin{equation}
206: \frac{P(N,E;\mu',T')}{P(N,E;\mu,T)}\, \sim \, 
207:  \frac{e^{-\beta'E}}{e^{-\beta E}}
208:   \frac{e^{\beta'\mu' N}}{e^{\beta\mu N}} \, 
209:    \sim \, e^{(\beta-\beta')E}e^{(\beta'\mu'-\beta\mu)N}\:. 
210: \end{equation}
211: where the proportionality constant can be calculated imposing the 
212: normalization of $P(N,E;\mu',T')$. The re-weighting procedure offers reliable 
213: results provided $T'$ and $\mu'$ are within few percents of $T$ and $\mu$.
214: 
215: We implement a least-square fit procedure to evaluate the values of $T$, 
216: $\mu$ and $s$ for which the reweighted distribution of $\mathcal{M}$ 
217: is closest to the known form for Ising-like systems~\cite{russi}.
218: The result of the fit provides the best estimate for the critical temperature 
219: $T_{\rm c}$, the critical chemical potential $\mu_{\rm c}$ and $s$. The 
220: location of the critical point has been performed for boxes of sizes $6$ 
221: through $9$, corresponding to an average number of particles ranging from 
222: about $60$ to over $400$. The simulations for larger boxes have been started 
223: at the critical parameters calculated for the closer smaller box. The $L=9$
224: simulation required about one month of CPU time for each of the 10 studied 
225: replicas on a 2.8\,GHz Intel Xeon.
226: 
227: \subsection{Phase coexistence}
228: \label{sec:phase}
229: We have calculated the phase coexistence curve using US-GCMC and GEMC simulations. 
230: The US-GCMC method is a natural extension of the standard GCMC 
231: method used to locate the critical point discussed above. Once the density 
232: fluctuations at the critical point have been evaluated, one can again apply the 
233: histogram reweigthing method to predict the shape of the density fluctuations 
234: at $T<T_{\rm c}$ for any fixed $\mu$ value. The value of $\mu$ for which 
235: coexistence between a gas and a liquid phase is present is chosen by selecting 
236: the value $\mu_{\rm x}$ for which the areas underneath the two peaks of 
237: $P({\cal M})$, the ordering operator fluctuation distribution, are equal.
238: 
239: We stress that performing a standard (Metropolis) GCMC simulation at a 
240: temperature even a few percent lower than $T_{\rm c}$ is not feasible, due to 
241: the large free-energy barrier separating the two phases which would prevent 
242: the simulated system to sample both liquid and gas configurations, thus 
243: yielding physically meaningless results. Several techniques have been developed 
244: to overcome this problem, among which the Umbrella Sampling Monte 
245: Carlo~\cite{umbrella} which has been used to perform the calculations 
246: presented herein. The US is a biased sampling technique which aims to flatten 
247: the free energy barrier between the two phases modifying the standard GCMC 
248: insertion/removal~\cite{frenkelsmith} probabilities as follows:
249: \begin{eqnarray}
250: \label{umbprob}
251: P_{{\rm ins}}^{{\rm US-GCMC}}\,&=\,P_{{\rm ins}}^{{\rm GCMC}}\frac{w(N)}{w(N+1)}\: \\ 
252: \nonumber
253: P_{{\rm rem}}^{{\rm US-GCMC}}\,&=\,P_{{\rm rem}}^{{\rm GCMC}}\frac{w(N)}{w(N-1)}\:.
254: \end{eqnarray}
255: In these last equations $w(N)$ is a forecast on the real $P(N)$ which 
256: we have repeatedly extracted from previous higher $T$ simulations through the 
257: histogram reweighting technique. If the predicted fluctuation distribution $w(N)$ 
258: is a good approximation to the real $P(N)$, the resulting biased fluctuation will 
259: result flat in $N$ and the system will thus not experience any difficulty in 
260: crossing the barrier between the liquid and gas phase. As shown in~\cite{umbrella}, 
261: the unbiased fluctuation distribution can be easily recovered adding a reverse 
262: bias to the results obtained with the insertion/removal probability in 
263: Eq.~\ref{umbprob}.
264: 
265: Starting from the critical point, we evaluate the phase diagram iterating the 
266: above procedure, progressively lowering $T$, down to the point where 
267: equilibration was not achieved any longer. Indeed, on cooling, due to the 
268: formation of a well-connected tetrahedral network, the dynamics in the liquid 
269: side slows down considerably~\cite{pwmnoi}. We have been very careful in 
270: progressively increasing the ratio $N_{\Delta}/N_N$ to compensate for the 
271: slowing down of the dynamics and the extremely slow equilibration times of the 
272: liquid phase. At the lower $T$, $N_{\Delta}/N_N=10000$. This sets a bound to 
273: the smallest $T$ which can be investigated.
274: 
275: We have also implemented a Gibbs Ensemble evaluation of the phase diagram. 
276: The GEMC method was designed~\cite{GEMC} to study 
277: coexistence in the region where the gas-liquid free-energy barrier is 
278: sufficiently high to avoid crossing between the two phases. 
279: Since nowadays this is a standard method in computational physics, we do 
280: not discuss it here and limit ourselves to noting that also in this case it is 
281: important to progressively increase, on cooling, the ratio $N_{\Delta}/N_N$ to 
282: account for the slow dynamics characterizing 
283: the liquid state. We have studied a system of (total) 350 particles which 
284: partition themselves into two boxes whose total volume is $2868\sigma^3$, 
285: corresponding to an average density of $\rho=0.122$. At the lowest $T$ this 
286: corresponds to roughly $320$ particles in the liquid box (of side $\approx 8\sigma$) 
287: and about 30 particles in the gas box (of side $\approx 13\sigma$). Equilibration 
288: at the lowest reported $T$ required about three months of computer time.
289: 
290: \begin{figure}[tb]
291: \includegraphics[width=8cm,clip=true]{figure2.eps}
292: \caption{Distributions of the fluctuations of the ordering operator for all studied sizes at the apparent critical point. The 
293: factor $a_{\mathcal{M}}^{-1}=0.34$ has been chosen to scale $P(x)$ to unit variance. 
294: The theoretical curve for the Ising model (full line) is from~\cite{russi}.}
295: \label{fig:pdix}
296: \end{figure}
297: 
298: \section{Results}
299: We start showing the distributions of the ordering operator fluctuations at 
300: the (apparent) critical points for all studied sizes. Implementing the fitting 
301: procedure described in Sec.~\ref{sec:cp}, using histogram reweighting, we 
302: first evaluate the values of the critical parameters and the shape of the 
303: density fluctuations at the critical point. The resulting distributions, for 
304: all investigated $L$, are shown as a function of the scaled variable 
305: $x \equiv a_M^{-1} L^{\beta/\nu} ({\cal M-M}_{\rm c})$ in Fig.~\ref{fig:pdix}. 
306: Here $\nu$ is the critical exponent of the correlation length and $\beta$ is 
307: critical exponent of the order parameter. Within the $d=3$ Ising universality 
308: class $\nu=0.629$ and $\beta=0.326$. From the fits we find that the 
309: non-universal amplitude is $a_M^{-1}=0.34$ (independent on $L$). The size 
310: dependence of $T_{\rm c}$ and $\mu_{\rm c}$ for $L=6$ to $L=9$ is shown in 
311: Fig.~\ref{fig:pc}. Finite size scaling predicts $T_{\rm c} \sim 
312: L^{-(\theta+1)/\nu}$ and $\mu_{\rm c} \sim L^{-(\theta+1)/\nu}$, where 
313: $\theta=0.54$ (Ising) is the universal correction to the scaling 
314: exponent~\cite{Wilding_96}. Fig.~\ref{fig:pc}(a-b) shows that the size 
315: dependence of the critical parameters is consistent with the 
316: expected universality class. Extrapolating the observed size dependence to 
317: $L \rightarrow \infty$ it is possible to provide an estimate for the bulk 
318: behavior of the PMW potential. We find $T_{\rm c}^{bulk}=0.1083(3)$, 
319: $\mu_{\rm c}^{bulk}=-1.265(1)$.
320: Fig.~\ref{fig:pc}(c) shows the $L$ dependence of $\rho_{\rm c}^{*}$, the 
321: latter being the system density (evaluated via histogram reweighting for 
322: each size) at $T_{\rm c}^{bulk}$ and $\mu_{\rm c}^{bulk}$. Finite size 
323: scaling predicts $\rho_{\rm c}^{*}\sim L^{-(d-1/\nu)}$. The resulting $L$ 
324: dependence of $\rho_{\rm c}^*$ is consistent with the theoretical 
325: prediction, despite the fact that $\rho_{\rm c}^*$ data are the noisiest, 
326: since the estimate of density is the most delicate and the error is at 
327: least of the order of one particle over the volume. The field mixing parameter 
328: $s$ has been difficult to infer precisely due to its small value and to the 
329: discrete nature of the square well interactions which makes the distribution 
330: of $N$ and $E$ quantized over integer numbers. A value of $s \approx 0.07$, 
331: consistent with the results reported in~\cite{bianchiprl} for a similar model, 
332: allows a simultaneous fit of all distribution functions, independently from $L$. 
333: 
334: \begin{figure}[tb]
335: \includegraphics[width=8cm,clip=true]{figure3.eps}
336: \caption{Size dependence of the apparent critical temperature 
337: $T_{\rm c}(L)$ (a), the apparent critical chemical potential $\mu_{\rm c}(L)$ (b) 
338: and the density at the true critical point $\rho_{\rm c}^{*}(L)$ (c).}
339: \label{fig:pc}
340: \end{figure}
341: 
342: \begin{figure}[tb]
343: \includegraphics[width=8cm,clip=true]{figure4.eps}
344: \caption{{\it (a)}: Logarithm of the distribution of density fluctuations 
345: $P(\rho)$ at $T~=~0.1082$, $0.1060$, $0.1049$, $0.1037$, $0.1026$, 
346: $0.1009$, $0.0999$, providing the negative of the density dependence of the 
347: free energy. The difference between the value of $P(\rho)$ at the valley and 
348: at the top is a measure of the free energy barrier separating the gas and the 
349: liquid state {\it (b)}: Free energy barrier (in units of $k_{\rm B}T$) between 
350: gas and liquid versus $T$ at coexistence.}
351: \label{fig:logPrho}
352: \end{figure}
353: 
354: \begin{figure}[tb]
355: \begin{center}
356: \includegraphics[width=8cm,clip=true]{figure5.eps}
357: \end{center}
358: \caption{{\it (a)}: Calculated gas-liquid coexistence for the PMW. Both GEMC (diamonds) 
359: and US-GCMC (circles) simulations results are shown. The star indicates the 
360: bulk ($L \rightarrow \infty$) critical point estimate. {\it (b)}: Extended phase diagram 
361: for the PMW, including: (i) the theoretical (Wertheim) gas-liquid coexistence line (dashed), 
362: (ii) the field of stability of the fluid, of the high-density crystal (HDS) and of the 
363: low-density crystal (LDS) phases from Ref.~\cite{Veg98a}; (iii) an iso-diffusivity line 
364: (for the smallest investigated value of $D$) from Ref.~\cite{pwmnoi}; (iv) percolation 
365: line from Ref.~\cite{pwmnoi}.}
366: \label{fig:coex}
367: \end{figure}
368: 
369: Next we discuss the gas-liquid coexistence curve for the PMW. Since only 
370: close to the critical point size effects are relevant, we have performed the 
371: calculations at $L=6$. Fig.~\ref{fig:logPrho} shows the behavior of the 
372: density fluctuations $P(\rho)$ (in log scale), evaluated with US-GCMC 
373: simulations (see Sec.~\ref{sec:phase}) along the coexistence curve. The 
374: difference between the peak and the valley in $\ln[P(\rho)]$ is a measure, 
375: in unit of $k_{\rm B}T$, of the activation free-energy 
376: $F_{barrier}/k_{\rm B}T$ needed to cross from the gas to the liquid phase 
377: and vice-versa. At the lowest studied temperature, the barrier reaches a 
378: value of about $20k_{\rm B}T$, which would clearly be impossible to 
379: overcome without the use of a biased sampling technique. Fig.~\ref{fig:logPrho} 
380: also shows the $T$ dependence of the barrier height, which indeed becomes of 
381: the order of the thermal energy close to the critical point.
382: 
383: The phase diagram resulting from our calculations is reported in 
384: Fig.~\ref{fig:coex}-(a). Both US-GCMC and GEMC data are reported, showing a 
385: perfect agreement for all studied $T$ proving that, despite the long 
386: equilibration times required, an accurate determination of the phase diagram 
387: for this model of patchy particles can nowadays be achieved. 
388: Fig.~\ref{fig:coex}-(b) shows the same data, together with the fluid-crystal 
389: coexistence lines calculated by Vega and Monson~\cite{Veg98a} and complemented 
390: with the bond percolation line from Ref.~\cite{pwmnoi}. We also report in the graph 
391: a so-called iso-diffusivity line, i.e. the set of points in the $T-\rho$ plane
392: in which the diffusion coefficient $D$ is constant. By selecting the smallest value
393: of $D$ which can be calculated with the present time computational facilities,
394: the iso-diffusivity line provides an estimate of the shape of the glass line.
395: Fig.~\ref{fig:coex} also shows the coexistence curve evaluated from the Wertheim theory, from Ref.~\cite{Veg98a}. 
396: 
397: Several observations arise from the data in Fig.~\ref{fig:coex}. 
398: First of all, we observe that the critical point, as well as the 
399: liquid branch of the coexistence curve, are characterized by 
400: a percolating (but transient) structure of bonds between pairs of H and LP. 
401: This is not unexpected, since the propagation of infinite range 
402: correlations, characteristic of a critical point, does require the 
403: presence of a spanning cluster~\cite{coniglio}. 
404: Particle diffusion is still significant, despite the presence of the transient 
405: percolating network of bonds, as shown by the comparison between the
406: phase-coexistence and the small $D$ iso-diffusivity line. In the region of $T$ 
407: where dynamics becomes so slow that equilibration can not be achieved on the 
408: present computational time scale, it becomes impossible also to evaluate the 
409: gas-liquid coexistence.
410: 
411: The gas-liquid coexistence is found to be metastable respect to fluid-crystal 
412: coexistence, in analogy with the case of particles interacting through 
413: spherical short-range potentials. This notwithstanding and differently from 
414: the case of spherical interacting particles, we never observe crystallization 
415: close to the critical point. This suggests that the increase in local density 
416: brought in by the critical fluctuations does not sufficiently couple with the 
417: orientational ordering required for the formation of the open diamond crystal 
418: structure.
419: 
420: The comparison between the GEMC and US-GCMC simulation phase diagram with the 
421: theoretical predictions of the Wertheim theory suggests
422: that the latter provides a quite accurate estimate for $T_{\rm c}$, whereas the 
423: $\rho$ dependence is only approximate. Indeed, the Wertheim theory predicts 
424: a vapor-liquid critical point at $T_{\rm c}=0.1031$ and 
425: $\rho_{\rm c}=0.279$~\cite{Veg98a}. 
426: These differences between theoretical predictions and numerical data 
427: confirm the conclusions that have been previously reached for models 
428: of patchy particles with different number of sticky spots~\cite{bianchiprl}. 
429: 
430: Finally, we note that in a very limited $T$-interval (less than 10 per 
431: cent of $T_{\rm c}$), the liquid density approximatively reaches a 
432: value comparable to the diamond crystal density, which for the present model has 
433: been calculated~\cite{Veg98a} $0.4880<\rho<0.6495$. Thus, as for the crystal state, 
434: the density of the liquid phase at coexistence is rather small, 
435: approximatively a factor of two smaller as compared to the case of spherically 
436: interacting particles. 
437: 
438: \begin{figure}[tb]
439: \includegraphics[width=8cm,clip=true]{figure6.eps}
440: \caption{Chemical potential (circles) and fugacity $z \equiv e^{\beta\mu}$ 
441: (diamonds) versus temperature $T$ at coexistence for $L=6$.}
442: \label{fig:muT}
443: \end{figure}
444: 
445: Finally, for the sake of completeness we show in Fig.~\ref{fig:muT} the 
446: location of the gas-liquid coexistence in the $\mu-T$ plane. 
447: 
448: \section{Conclusions}
449: In this article we have presented an accurate determination of the 
450: critical point and of the gas-liquid phase coexistence curve for a 
451: primitive model for water, introduced by Kolafa and Nezbeda~\cite{Kol87a}. 
452: Despite its original motivation, the PMW can also be studied as 
453: a model for patchy colloidal particles and, perhaps, as an elementary 
454: model for describing patchy interactions in proteins. To this extent, 
455: it is particularly important to understand the qualitative features of 
456: the phase diagram, the stability or metastability of the gas-liquid line and 
457: the propensity to crystallize.
458: 
459: We have shown that the critical fluctuations are consistent with the Ising 
460: universality class, both via the analysis of the shape of the density 
461: fluctuation distribution and via the size dependence of the critical 
462: parameters. The determination of the critical point allows us to prove that, for 
463: the present model, the liquid phase is not present in equilibrium, and it is 
464: only observed in a metastable condition. In the case of spherical interaction 
465: potentials, the thermodynamic stability of the gas-liquid critical point 
466: is achieved when the range of the interaction potential becomes of the order 
467: of one tenth of the particle diameter~\cite{Vli00a}. Data reported in this 
468: article (see Fig.~\ref{fig:coex}) confirm that the property of short-range 
469: potentials of missing a proper equilibrium liquid phase is retained in 
470: short-range patchy particles. It would be of interest to study for which 
471: critical value of the interaction range a proper liquid phase appears in 
472: such patchy particles.
473: 
474: The evaluation of the gas-liquid coexistence has been performed using two 
475: distinct methods, the US-GCMC and the GEMC, and a very good agreement has been 
476: recorded despite the difficulties in equilibrating the liquid phase at low $T$.
477: The present results, together with previous studies of the crystal phase and of 
478: the slow low-$T$ dynamics, offer a coherent picture of the behavior of the 
479: model. It is shown that phase separation is intervening only at low $T$ for 
480: $\rho \lesssim 0.6$, a value significantly smaller than the corresponding 
481: value for spherical interaction potentials. This is consistent with the fact 
482: that the number of attractive interactions in which a particle can be engaged 
483: (four in the present model) controls the minimum density of the liquid phase. 
484: These results elucidate the different nature of bonded liquids (the so-called 
485: network forming) with respect to the category of spherically interacting 
486: liquids. For bonded systems, there is an intermediate region, between the high 
487: density packed structure and the unstable region, in which gas-liquid phase 
488: separation is observed, which is not accessible to spherically interacting 
489: particles. In this intermediate density region, the system is structured in a 
490: percolating network and both static and dynamic quantities are controlled by 
491: the presence of bonds. This picture, which has been developing in a progression 
492: of studies~\cite{Zac05a,pwmnoi,silica,valency}, is confirmed by the present 
493: results.
494: 
495: \section*{Acknowledgments}
496: 
497: We acknowledge support from MIUR-Prin and
498: MCRTN-CT-2003-504712.
499: 
500: \section*{References}
501: \bibliographystyle{apsrev}
502: \bibliography{pmw}
503: 
504: \end{document}
505: