1: \documentclass{cpcauth}	

2: \usepackage{epsfig}

3: \begin{document}	

4: \begin{frontmatter}

5: \title{Classical and ab initio molecular dynamic simulation of an amorphous silica surface}

6: \author{C. Mischler$^1$, W. Kob$^2$ and K. Binder$^1$}

7: \address{$^1$Institut f\"ur Physik, Johannes Gutenberg-Universit\"at

8: \\55099 Mainz; Germany\\

9: $^2$Laboratoire des Verres, Universit\'e Montpellier II \\

10: 34095 Montpellier; France}

11: 

12: \begin{abstract}

13: We present the results of a classical molecular dynamic simulation

14: as well as of an {\it ab initio} molecular dynamic simulation of an

15: amorphous silica surface. In the case of the classical simulation we use

16: the potential proposed by van Beest {\it et al.} (BKS) whereas the {\it

17: ab initio} simulation is done with a Car-Parrinello method (CPMD). We

18: find that the surfaces generated by BKS have a higher concentration of

19: defects (e.g. concentration of two-membered rings) than those generated

20: with CPMD. In addition also the distribution functions of the angles and

21: of the distances are different for the short rings. Hence

22: we conclude that whereas the BKS potential is able to reproduce correctly

23: the surface on the length scale beyond $\approx$5~\AA, it is necessary

24: to use an {\it ab initio} method to predict reliably the structure at

25: small scales.

26: 

27: \end{abstract}

28: \begin{keyword} glass surface; silica; molecular dynamics simulations; 

29: ab initio simulations

30: \end{keyword}

31: \end{frontmatter}

32: 

33: \section{Introduction}

34: Obtaining a good understanding of the structural and dynamical

35: properties of the surface of amorphous silica is very important for

36: the manufacture of glass as well as the construction of electronic

37: devices~\cite{legrand98}. This is the reason why in the past a large

38: number of experiments have been done to investigate this type of

39: surface. Since in real experiments it is rather difficult to obtain

40: reliably details on the structure also quite a few computer simulations

41: have been done in order to study this system (see~\cite{roder01} and

42: references therein). Most of these studies have, however, been done by

43: using effective classical potentials, such as, e.g., the one proposed

44: some years ago by van Beest, Kramer, and van Santen (BKS)~\cite{beest90}.

45: Although it has been shown that these type of potentials can reproduce

46: quite reliably the structure and dynamics of silica in the {\it bulk}~(

47: see, e.g., \cite{horbach99} and references therein), it is much less

48: obvious to what extent they are also able to give a correct description

49: of the properties of silica close to a surface, since the parameters for

50: these potentials, effective charges, etc., have often been optimized

51: to reproduce only experimental data for the bulk. One possibility to

52: avoid this problem with the classical effective potential

53: is to use {\it ab initio} simulations such as the scheme proposed

54: by Car and Parrinello~\cite{car85} since in this type of approach an

55: effective potential between the ions is calculated self consistently on

56: the fly, i.e. the instantaneous geometry of the ions is always taken

57: into account. The drawback of this approach is that due to the huge

58: computational burden only relatively short time scales, a few ps, 

59: as well as small systems, a few hundred particles, can be

60: simulated, whereas classical simulations allow to simulate thousands of

61: particles over several ns.

62: 

63: In the present work we compare the results of a classical simulation of an

64: amorphous silica surface with the BKS potential with the results obtained

65: by the Car-Parrinello Molecular Dynamics method (CPMD). The goal is to

66: check which quantities are reproduced correctly by the BKS potential,

67: using the results of the CPMD simulation as the reference system.

68: 

69: \section{The BKS potential and the setup of the geometry}

70: 

71: We have first prepared the system using the BKS potential. In this

72: two-body potential the atoms interact also by means of a Coulomb

73: potential where the effective charges of a silicon and oxygen atom is

74: 2.4 and $-1.2$, respectively. More details on this potential can be

75: found in Ref.~\cite{beest90}.

76: 

77: In order to minimize finite size effects as well as surface effects

78: it is customary to use periodic boundary

79: conditions (PBC) in all three directions. If one wants to investigate a

80: free surface, the most straightforward idea is to use a film geometry,

81: i.e. to have PBC in two directions and to have an infinite free space

82: above and below the system. Unfortunately it turns out, however, that

83: from a computational point of view this setup is not very good for

84: systems with Coulombic interactions (i.e.  such as the one studied

85: here), since it prevents to make an {\it efficient} use of the Ewald

86: summation method. Therefore we have adopted the following strategy

87: which is explained also in Fig.~\ref{fig1} (see also \cite{feuston89,ceresoli00}):

88: \begin{figure}[ht]

89: \begin{center}

90: \unitlength1mm

91: \begin{picture}(0,100)

92: \put(-50,-7){

93: \includegraphics[width=100mm]{bild.konf.ps}

94: }

95: \end{picture}

96: \caption{Procedure to create the used sandwich geometry}

97: \label{fig1}

98: \end{center}

99: \end{figure}

100: i) We start with a relatively large (bulk) silica system at

101: $T=3400$~K with PBC in three dimension (with box size $L_x=L_y=11.51$\AA\,

102: and $L_z'=23$\AA.  ii) We cut the system perpendicular to the

103: $z$-direction into two pieces. Without loss of generality we can assume

104: that the mean position of this cut is at $z=L_z'$. At this point it is

105: very important, that we cut only oxygen-silicon-bonds, such that we get

106: only free oxygen atoms at this interface (thus the interface will have

107: a bit of roughness). iii) These free oxygen atoms are now saturated

108: by hydrogen atoms. The place of these hydrogen atoms are chosen such

109: that each of the new oxygen-hydrogen bonds is in the same direction as

110: the oxygen-silicon bond which was cut and has a length of approximately

111: 1 \AA. The interaction between the hydrogen atoms and the oxygen atoms

112: as well as the silicon atoms are described only by a Coloumbic term. The

113: value of the effective charge of the hydrogen atoms is set to $0.6$, which

114: ensures that the system is still (charge) neutral. iv) We make atoms which have a distance from this interface that is less than 4.5 \AA\

115: completely immobile, whereas the atoms that have a larger distance can

116: propagate subject to the force field. v) We add in $z$-direction an empty

117: space of $\Delta z=6.0$~\AA\, and thus generate a free surface at around

118: 14.5~\AA. With this sandwich geometry we now can use periodic boundary

119: conditions in all three directions.  We have made sure that the value

120: of $\Delta z$ is sufficiently large that the results do not depend on it

121: anymore~\cite{mischler01}. Note that it is not advisable to choose $\Delta

122: z$ too large, since this would increase the cost of the CPMD simulation.

123: At the end of this procedure we have a system of 91 oxygen, 43 silicon

124: and 10 hydrogen atoms in a simulation box with L$_x$=L$_y$=11.51 \AA\/

125: and $L_z\approx 25$~\AA.

126: 

127: 

128: 

129: \section{CPMD-Simulation}

130: Since the time scale which is accessible to the CPMD-method is very

131: restricted, we have to combine the {\it ab initio} with classical

132: calculations. For this we first prepared a classical system as

133: described in the previous section, equilibrated it for about 1~ns which

134: is sufficient to equilibrate it completely~\cite{horbach99}. From

135: a subsequent production run with the same duration we picked 100

136: statistical independent configurations and used them to characterize the

137: static properties of the system with a high accuracy. Using a subset

138: of these configurations as starting points, we subsequently started

139: CPMD-simulation using the CPMD code developed in Stuttgart~\cite{cpmd}. For

140: the CPMD we used conventional pseudopotentials for silicon and oxygen and

141: the BLYP exchange-functions~\cite{trouiller91,lee88}. The electronic wave-functions

142: were expanded in a plane wave basis set with an energy cutoff of 60 Ry

143: and the equations of motion were integrated with a time step of 0.085 fs

144: for 0.2 ps. In the analysis of the CMPD data only those configurations

145: were taken into account that were produced later than 5 fs after the

146: start of the CPMD run in order to allow the system to equilibrate at

147: least locally~\cite{benoit00}.

148: 

149: In the analysis of the classical configurations we noted that typically

150: one of the three following situations is present on the surface:

151: 

152: \begin{itemize}

153: \item systems with no defects (i.e. all Si and O atoms are four and two-fold

154: coordinated, respectively)

155: \item systems with an undercoordinated oxygen atom and an undercoordinated silicon atom

156: \item systems with an overcoordinated oxygen atom and an undercoordinated oxygen atom

157: \end{itemize}

158: 

159: Therefore we picked for each case two BKS configurations and started

160: the CPMD runs.

161: 

162: The largest differences between the results of the classical and of

163: the CPMD simulation are found for the short rings (n$<$5). (A ring

164: is a closed loop of $n$ consecutive Si-O segments~\cite{wright78}.) Figure~\ref{fig2}

165: shows the probability to find a ring of size $n$ for the case of BKS

166: and CPMD. We see that the BKS potential overestimates the frequency

167: with which a ring of size two occurrs by about a factor of two. Related

168: to this is the observation that the overshoot that is observed in the

169: $z-$dependent mass density profile is less pronounced in the case of

170: the CPMD than the one for the BKS (inset of Fig.~\ref{fig2}), since

171: two-membered rings are relatively dense.

172: 

173: \begin{figure}[ht]

174: \begin{center}

175: \unitlength1mm

176: \includegraphics[width=55mm]{ring.dichte.cp.bks.eps}

177: \caption{Probability to find a ring of size $n$. Inset: $z-$dependence of the mass

178: density.}

179: \label{fig2}

180: \end{center}

181: \end{figure}

182: 

183: From this inset we also see that the density profile for the CPMD extends

184: to larger $z$ values that the one for the BKS (by about 0.4~\AA). This

185: is because the BKS potential is not able to reproduce correctly the

186: density of silica at zero pressure.

187: 

188: Another interesting result is the dependence of the distribution of angles

189: O-Si-O on the ring size (Fig.~\ref{fig3}). For large $n$, $n>$4, i.e. the

190: sizes which are normally found in the bulk \cite{rino93}, the results

191: of the two different methods are in good agreement \cite{benoit00}. For

192: smaller $n$, however, the mean O-Si-O-angle from CPMD is shifted to

193: larger values in comparison to the classical one. This shift becomes

194: more pronounced with decreasing $n$. Furthermore also the shape of the

195: distributions starts to become different if $n$ is small.

196: 

197: \begin{figure}[ht]

198: \begin{center}

199: \unitlength1mm

200: \includegraphics[width=55mm]{angle.osio.n2.n6.cp.bks.eps}

201: \caption{Distribution of O-Si-O-angles for different ring sizes. BKS$=$filled symbols; CPMD$=$open symbols.}

202: \label{fig3}

203: \end{center}

204: \end{figure}  

205: 

206: This effect can be understood better by analyzing the partial radial

207: distribution functions $g(r)$ which are shown in Fig.\ref{fig4}. We

208: see that for the Si-O pair the curves from CPMD are shifted to larger

209: distances by about 0.04~\AA\, and that this shift is independent of

210: $n$. Also the $g(r)$ for the O-O pairs are shifted to larger $r$, but

211: this time the amount does depend on $n$. In particular we note that the

212: O-O distance is nearly independent of $n$ for the case of CPMD, whereas

213: it increases with $n$ for the case of BKS. These effects results in the

214: difference in the distribution of the O-Si-O angles if $n$ is small.

215: 

216: \begin{figure}[ht]

217: \begin{center}

218: \unitlength1mm

219: \includegraphics[width=55mm]{paar.oo.sio.cp.bks.eps}

220: \caption{Radial distribution function for different ring sizes. Left: O-O. Right Si-O. BKS$=$filled symbols; CPMD$=$open symbols.}

221: \label{fig4}

222: \end{center}

223: \end{figure} 

224: 

225: \section{Conclusion}

226: 

227: In this work we have investigated some structural properties of an

228: amorphous silica surface. In particular we have studied how these

229: properties depend on the simulation method: A classical simulation

230: with the potential proposed by van Beest {\it et al.} (BKS) and a

231: Car-Parrinello simulation (CPMD). We find that the structure on larger

232: length scales are independent of the method used, whereas the details of

233: structural elements on short scales (short rings, distribution function

234: for angles, etc.) differ. Thus this shows that it is probably necessary

235: to use {\it ab initio} methods if one wants to understand these systems

236: at short length scales in a quantitative way.

237: 

238: Acknowledgement: We gratefully acknowledge the financial support by the

239: SCHOTT Glaswerke Fond and the DFG under SFB 262 and the BMBF under grant N$^o$ 03N6015. We thank the NIC J\"ulich for a generous grant of computing time.

240: 

241: \begin{thebibliography}{}

242: \bibitem{legrand98}

243: A. P. Legrand, 

244: {\it The Surface Properties of Silica}, (Wiley, New York, 1998).

245: 

246: \bibitem{roder01}

247: A.Roder, W.Kob and K.Binder 

248: J.Chem.Phys. {\bf 114}, (2001) 7602.

249: 

250: \bibitem{beest90}

251: B. W. H. van Beest, G. J. Kramer, and R. A. van Santen 

252: Phys. Rev. Lett. {\bf 64}, (1990) 1955.

253: 

254: \bibitem{horbach99}

255: J. Horbach and W. Kob  

256: Phys. Rev. B {\bf 60} (1999) 3169.

257: 

258: \bibitem{car85}

259: R. Car, M. Parrinello 

260: Phys. Rev. Lett. {\bf 55}, (1985) 2471.

261: 

262: \bibitem{feuston89}

263: B. P. Feuston and S. H. Garofalini  

264: J. Chem. Phys. {\bf 91}, (1989) 564.

265: 

266: \bibitem{ceresoli00}

267: D. Ceresoli, M. Bernasconi, S. Iarlori, M. Parrinello, and E. Tosatti 

268: Phys. Rev. Lett. {\bf 84}, (2000) 3887.

269: 

270: \bibitem{mischler01}

271: C. Mischler, W. Kob, and K. Binder 

272: (in preparation).

273: 

274: \bibitem{cpmd}

275: CPMD Version 3.3a, J. Hutter, P. Ballone, M. Bernasconu, P. Focher, E. Fois, 

276: St. Goedecker, M. Parrinello, and M. Tuckermann, MPI f\"ur Festk\"orperphysik and 

277: IBM Research 1990-2000

278: 

279: \bibitem{trouiller91}

280: N. Trouiller, J. L. Martins 

281: Phys. Rev. B {\bf 43}, (1991) 1993.

282: 

283: \bibitem{lee88}

284: C.Lee, W.Yang and R.G.Parr 

285: Phys.Rev.B {\bf 37}, (1988) 785.

286: 

287: \bibitem{benoit00}

288: M.Benoit, S.Ispas, P.Jund and R.Jullien 

289: Eur. Phys. J. B {\bf 13}, (2000) 631.

290: 

291: \bibitem{wright78}

292: A.C.Wright and J.A.E.Desa 

293: Phys. Chem. Glasses {\bf 19}, (1978) 140.

294: 

295: \bibitem{rino93}

296: J.P.Rino, I.Ebbsj\"o and R.K.Kalia et al.

297: Phys. Rev. B {\bf 47}, (1993) 3053.

298: 

299: \end{thebibliography}

300: \end{document}

301: 

302: 

303: 

304: 

305: 

306: 

307: 

308: 

309: