1: \documentstyle[epsfig,amsmath,prl,aps]{revtex}
2:
3: \begin{document}
4:
5: \bibliographystyle{prsty}
6:
7: \draft
8:
9: \title{Instantaneous Normal Mode analysis of liquid HF}
10:
11: \author{G. Garberoglio and R. Vallauri \cite{email}}
12: \address{Dipartimento di Fisica dell'Universit\`a di Trento and Istituto
13: Nazionale di Fisica della Materia \\ Via Sommarive 14, I-38050 Povo (TN),
14: Italy}
15:
16: \date{January 19, 2000. Revised March 21, 2000}
17:
18: \maketitle
19:
20: \begin{abstract}
21: We present an Instantaneous Normal Modes analysis of liquid HF aimed to
22: clarify the origin of peculiar dynamical properties which are supposed to stem
23: from the arrangement of molecules in linear hydrogen-bonded network. The
24: present study shows that this approach is a unique tool for the understanding
25: of the spectral features revealed in the analysis of both single molecule and
26: collective quantities. For the system under investigation we demonstrate the
27: relevance of hydrogen-bonding ``stretching'' and fast librational motion
28: in the interpretation of these features.
29: \end{abstract}
30:
31: \pacs{PACS numbers: 62.60.+v, 63.20.Pw}
32:
33: In recent years the analysis of Instantaneous Normal Modes (INM) of normal and
34: supercooled liquids has given a sound improvement to the understanding of the
35: microscopic processes underlying the dynamical properties of these
36: systems. Applications of the method are to be found in the calculation of
37: macroscopic quantities through the knowledge of the density of states
38: (e.g. the diffusion coefficient \cite{keyes_diff}) and in the interpretation
39: of atomic motion through the inspection of the eigenvectors
40: \cite{madden,sciortino,mazzacurati}.
41: This last method appears to be a unique tool for the interpretation of
42: dynamical features (single molecule or collective) of a disordered system in
43: terms of correlated motions of its constituents. In particular it is
44: interesting to explore how the presence of locally ordered units is reflected
45: in the time behaviour of, for example, the velocity autocorrelation function
46: (VACF). Attempts in this direction have to be found in the analyses of:
47: $i)$
48: the correlation function of the projection of the centre of mass (CoM)
49: velocity of water molecules along the directions of the normal coordinates of a
50: cluster of three molecules \cite{sv};
51: $ii)$
52: the projection of INM eigenvectors onto the totally symmetric displacement
53: coordinates of ZnCl$_4^{2-}$ tetrahedral units \cite{madden}.
54: Moreover the INM approach has been exploited to describe the dynamical
55: features of molecular liquids (e.g. diatomic Lennard--Jones \cite{buchner},
56: CS${}_2$
57: \cite{keyes_cs2}) including hydrogen bonded systems like water
58: \cite{water_inm,sciortino}. On the other hand recent molecular dynamics (MD)
59: simulations of HF have revealed peculiar dynamical features, e.g. a peak at
60: $\simeq$ 50 ps${}^{-1}$
61: in the spectra of both collective and single molecule correlation
62: functions \cite{bstv,bbstv}. In the collective longitudinal and transverse
63: current spectra this mode appears to have an optical-like character, since its
64: frequency is found to be independent of the wavevector. For the interpretation
65: of the CoM VACF spectrum this peak has been
66: related to the relative motion of two nearest neighbour molecules
67: \cite{bbstv}. Since nearest neighbours are also hydrogen bonded it is tempting
68: to assign this dynamical feature to hydrogen bonding ``stretching''. As a
69: matter of fact HF molecules have been demonstrated to form irregular zig-zag
70: chains of different size, being this peculiar clustering favoured by the
71: geometry of the molecule and the strong electrostatic interaction
72: \cite{jv_nonpol}. A dynamical characterisation of these ordered units has not
73: yet been given, so that any assignment of the spectral features to particular
74: dynamical process remains speculative.
75:
76: In the present Letter we report the results of an INM analysis of liquid HF
77: aimed to give an answer to the above questions. It will be shown to which
78: extent the presence of irregular chains is reflected in the INM spectra and
79: how the appearance of optical-like modes can be understood from the analysis
80: of the short time dynamics naturally expressed by the INM eigenvectors.
81:
82: The use of the INM analysis starts from the solution of the eigenvalue
83: problem
84: \begin{equation}
85: \omega^2 {\sf T} {\bf e} = {\sf K} {\bf e}
86: \label{eq:eigenproblem}
87: \end{equation}
88: where ${\sf T}$ is the mass matrix of the system and ${\sf K}$ is the matrix
89: of the second derivatives of the potential energy. Here ${\bf e}$ represents a
90: multidimensional vector which specifies the instantaneous configuration of the
91: system, i.e. in general the three CoM coordinates and Euler angles of the
92: assumed rigid molecule. An average over many independent configurations has to
93: be performed in order to obtain the distribution of eigenfrequencies and any
94: dynamical quantity derived from the knowledge of the eigenvector ${\bf e}$. In
95: order to avoid spurious effects originating in the fact that the mass matrix
96: ${\sf T}$, which depends on the sine of the polar angle, is not guaranteed to
97: be strictly positive definite at each time, we have used the orientational
98: coordinates described in Ref. \cite{keyes_irc} to evaluate the derivatives of
99: the potential energy.
100:
101: We have calculated the INM for liquid HF (at T = 203 K and $\rho$ = 1.178
102: g/cm${}^3$), using 60 configurations separated by 2 ps with N = 108
103: molecules. The potential model, reported in
104: Ref. \cite{jv_hfstruct}, accounts for the interaction of three fractional
105: charges plus a Lennard--Jones contribution and it has been shown to
106: satisfactorily reproduce the thermodynamics and structure of the liquid.
107: Due to the complexity of the interaction potential we have evaluated the
108: matrix of the second derivatives numerically by displacing each generalized
109: coordinate with an increment $h$ = 10${}^{-5}$ \AA\ since it gives a good
110: stability of the matrix elements and no unusually high eigenvalues. We also
111: checked the accuracy of the results by the presence of three eigenmodes with
112: zero frequency. Our value of the increment is in accordance with the one used,
113: for example, in Ref. \cite{keyes_tos}. The generalized eigenvalue problem was
114: resolved using standard methods \cite{nr}.
115:
116: %% Figure 1
117: \begin{figure}
118: \centerline{
119: \epsfig{file=inm_spectrum.eps,width=0.75\linewidth}
120: }
121: \caption{Spectrum of the INM, with imaginary frequencies reported in the
122: negative part, as usual. The thick line is the total spectrum, and has
123: been divided in translational and rotational components, the latter of which
124: appears only at high frequency. The dotted and dashed lines correspond to the
125: contribution parallel and perpendicular to the molecular axis respectively.}
126: \label{fig:inm_spectrum}
127: \end{figure}
128:
129: The INM spectrum is presented in Fig. \ref{fig:inm_spectrum}. A comparison
130: with the results for water at room temperature (reported in
131: Ref. \cite{water_inm}) points out some important features. The contribution of
132: imaginary modes (conventionally reported on the negative axis) is
133: substantially larger in HF than in H${}_2$O (16\% against 6\%), a result
134: consistent with the relatively higher diffusion coefficient found in HF
135: \cite{bbstv}. Moreover there is a much more definite separation between
136: translational and rotational contributions defined by:
137: \begin{eqnarray}
138: \rho_T(\omega) & = & \left\langle
139: \sum_i \sum_{\mu = 1,2,3} ({\bf e}^\alpha_{i \mu})^2
140: \delta(\omega - \omega_\alpha) \right\rangle\\
141: \rho_R(\omega) & = & \left\langle \sum_i \sum_{\mu = 4,5} ({\bf e}^\alpha_{i
142: \mu})^2 \delta(\omega - \omega_\alpha) \right\rangle
143: \end{eqnarray}
144: where ${\bf e}^\alpha_{i \mu}$ is the component of the eigenvector
145: corresponding to the frequency $\alpha$, referred to the $i$-th molecule and
146: to coordinate component $\mu$ ($\mu$ = 1,2,3 CoM, $\mu$ = 4,5
147: rotational coordinates).
148:
149: The rotational spectrum extends from 50 to 250 ps${}^{-1}$, whereas in water
150: it goes from 0 to 180 ps${}^{-1}$. The translational spectrum of HF shows a
151: clear second maximum at $\simeq$ 50 ps${}^{-1}$ absent in the spectrum of
152: water where one can only notice an asymmetry of the low frequency maximum with
153: a larger content at higher frequency. We believe (and demonstrate in the
154: later) that this secondary maximum is related to the ``stretching'' of
155: hydrogen bond between two HF molecules. A comparison with the spectrum of the
156: CoM VACF \cite{bbstv} points out that the peak at $\simeq$ 50
157: ps${}^{-1}$ is much less pronounced and separated in the translational
158: component of the INM.
159:
160: %% Figure 2
161:
162: \begin{figure}
163: \centerline{
164: \epsfig{file=perp_par_ratio.eps,width=0.75\linewidth}
165: }
166: \caption{The ratio between $\rho_\perp(\omega)$ and
167: $\rho_\parallel(\omega)$ (see Eq. \ref{eq:parproj}). It should be 2 for an
168: isotropic system.}
169: \label{fig:par_perp}
170: \end{figure}
171:
172: The strong anisotropy of the single molecule dynamics is shown by performing a
173: projection of the translational component of the eigenvector along the
174: directions parallel and perpendicular to the molecular axis:
175: \begin{equation}
176: \rho_\parallel(\omega) = \left\langle
177: \sum_i \sum_{\mu = 1,2,3} ({\bf e}^\alpha_{i \mu} {\bf u}_{i \mu})^2
178: \delta(\omega - \omega_\alpha) \right\rangle
179: \label{eq:parproj}
180: \end{equation}
181: where ${\bf u}_i$ is a unit vector along the symmetry axis of molecule $i$
182: and we have also defined $\rho_\perp(\omega) = \rho_T(\omega) -
183: \rho_\parallel(\omega)$.
184: If the molecular motion were isotropic, the ratio
185: between the perpendicular and parallel contributions would be equal to
186: two. Fig. \ref{fig:par_perp} shows that below 20 ps${}^{-1}$ this ratio
187: becomes larger than two going over four for modes at negative frequencies:
188: a result which points out that molecules can diffuse more freely in the
189: direction perpendicular to their axis rather than in the parallel one. Beyond
190: 50 ps${}^{-1}$ the ratio becomes equal or less than one, thus revealing that
191: the motion of the molecules in this frequency range occurs, on a large extent,
192: in the direction of the molecular axis. Such an observation will be relevant
193: in discussing the arrangement of molecule along irregular chains as revealed
194: by the subsequent analysis of the INM eigenvectors. As a final remark we wish
195: to point out that the clear separation between translational and rotational
196: components can be considered as a print of hydrogen bonded systems. In fact it
197: is present in water and HF but not in a linear Lennard--Jones diatomic which
198: has been investigated in \cite{buchner}.
199:
200: To analyse in more detail the structure of the INM we need to determine
201: which molecules participate in each given mode. We assume the following
202: criterion: the molecule $i$ said to belong to mode $\alpha$ if the condition
203: $
204: \sum_\mu ( {\bf e}_{i \mu}^\alpha )^2 > \frac{1}{N}
205: \label{eq:partmode}
206: $
207: is fulfilled. The distribution of the number of molecules per mode is reported
208: in Fig. \ref{fig:part_n_radius} (solid line) as a function of
209: frequency. Knowing the particles participating in a given mode, we can
210: determine the spatial localisation of the mode.
211:
212: %% Figure 3
213:
214: \begin{figure}
215: \centerline{
216: \epsfig{file=part_n_radius.eps,width=0.75\linewidth}
217: }
218: \caption{Solid line (left scale): Frequency distribution of the particles
219: participating in the modes. Dashed line (right scale): the distribution of the
220: mode radii, normalized with the simulation box length (14.504 \AA).}
221: \label{fig:part_n_radius}
222: \end{figure}
223:
224: If ${\bf r}_i$ is the CoM position of particle $i$ and ${\cal
225: S}_\alpha$ is the set of particles involved in the mode we can define a
226: ``radius'' for the mode as the maximum distance between two particles i.e.
227: $
228: R_\alpha = {\mathrm max}_{i,j \in {\cal S}_\alpha} | {\bf r}_i - {\bf r}_j |
229: \label{eq:mode_radius}
230: $
231: , the distribution of which is reported in Fig. \ref{fig:part_n_radius}
232: (dashed line). A comparison of the behaviour of the two quantities
233: indicates a clear correspondence between participation ratio and extension
234: of the modes. A lower number of participating molecules is accompanied by a
235: smaller radius of the mode. Surprisingly enough, however, in the range of
236: $\simeq$ 50 ps${}^{-1}$ the modes has an extension of about half of the box
237: length even if the number of molecules participating is smaller than ten.
238:
239: A better understanding of this localisation problem can be obtained by
240: performing a projection of a mode onto the hydrogen bonded chains present in
241: the system, where hydrogen bonding is defined by the same
242: energetic criterion adopted in Ref. \cite{jv_hfstruct}. If ${\cal C}_c$
243: denotes the set of molecules belonging to the chain $c$ we define the
244: projection of the mode $\alpha$ on the chain as
245: \begin{equation}
246: P^\alpha_{{\cal C}_c} =
247: \sum_{i \in {\cal C}_c, \mu} ( {\bf e}^\alpha_{i \mu} )^2
248: \end{equation}
249: which is one if the mode $\alpha$ is localised on the chain $c$, and is zero
250: if the mode $\alpha$ involves molecules not belonging to the chain. Clearly if
251: we consider the maximum of this projection taken on the set of all the chains,
252: we can see whether the modes are localised on some chain or not.
253:
254: The result presented in Fig. \ref{fig:maxproj_n_chain} (dashed line),
255: indicates that only in some frequency range the modes show a high degree of
256: localisation on a single chain, in particular in the same range where a
257: minimum partecipation ratio occurs and the radius of the mode is lower (see
258: Fig. \ref{fig:part_n_radius})
259:
260: % Figure 4
261:
262: \begin{figure}
263: \centerline{ \epsfig{file=maxproj_n_chain.eps,width=0.75\linewidth} }
264: \caption{Solid line (left scale): Frequency distribution for the number of
265: chains involved in the modes. Dashed line (right scale): Frequency dependence
266: of the mean value of the maximum projection of the eigenmodes onto a chain.}
267: \label{fig:maxproj_n_chain}
268: \end{figure}
269:
270: The idea that modes at particular frequencies are strongly correlated to the
271: presence of chains is confirmed by looking at the distribution of the number
272: of chains involved in a mode as reported in Fig. \ref{fig:maxproj_n_chain}
273: (solid line). In the range $\simeq$ 50 ps${}^{-1}$ (and in the region of
274: imaginary frequencies) we find the modes are spread over two distinct chains;
275: at the highest frequencies only one chain per mode is involved. Since in this
276: range only four molecules are participating to the mode (see
277: Fig. \ref{fig:part_n_radius}) we can conclude that these modes are confined on
278: the ring chains (tetramers) which are found to be particularly stable
279: \cite{jv_hfstruct}.
280:
281: Having said that, it is also evident that most of the modes are far from being
282: localized over a single chain, e.g. all the modes whose maximum projection is
283: less than 0.80. This result is in accordance with other INM studies of network
284: forming systems \cite{madden}, where it is shown that the modes do not
285: typically reproduce the behaviour that could be expected on the basis of the
286: group properties of the network, but generally have some sort of mixed
287: character.
288:
289: In order to characterize the spatial correlation of the molecular
290: displacements through the INM eigenvectors, we have examined the following quantity:
291: %%\begin{multline}
292: \begin{equation}
293: \psi(R,\omega) =
294: \left\langle
295: \frac{1}{3N} \sum_\alpha \frac{1}{n(R)}
296: \sum_{|{\bf r}_i - {\bf r}_j| \in R} \sum_{\mu = 1,2,3}
297: {\bf e}_{i \mu}^\alpha {\bf e}_{j \mu}^\alpha
298: \delta(\omega - \omega_\alpha)
299: \right\rangle
300: \label{eq:spat_corr}
301: \end{equation}
302: %%\end{multline}
303: where $R$ denotes a spatial range (e.g. first or second shell of nearest
304: neighbours, defined through the minima of the CoM pair correlation) and $n(R)$
305: the number of pairs present in that range for a given configuration. This
306: quantity is a sort of mean value of the scalar product of the displacement of
307: the particles being nearest or next-to-nearest neighbours in a given normal
308: mode and at least at short time characterises in an exact way the motion of
309: the molecules participating to the particular mode. It is of course positive
310: if the particles move in phase and negative for out-of-phase
311: displacements. The results are shown in Fig. \ref{fig:psi}.
312:
313: %% Figure 5
314:
315: \begin{figure}
316: \centerline{
317: \epsfig{file=spat_corr.eps,width=0.75\linewidth}
318: }
319: \caption{The value of the spatial correlation between particles in various
320: normal mode (see Eq. \ref{eq:spat_corr}). The solid line shows correlation
321: between the molecules being in the first shell, and the dashed line shows
322: correlation between molecules in the second shell (multiplied by a factor
323: of 5). The zero-frequency modes (corresponding to an overall translation
324: of the system) have been dropped from the calculation.}
325: \label{fig:psi}
326: \end{figure}
327:
328: We notice that the nearest neighbours have opposite phases in a broad
329: frequency range around 50 ps${}^{-1}$, a behaviour in accordance with the
330: presence of the ``stretching'' mode similar to the one observed in the
331: solid \cite{hf_solid} and consistent with the fact that, in this range of
332: frequency, the molecules move preferentially along the direction of the
333: molecular axis. We notice that the next-to-nearest neighbours are not very
334: much correlated in this range, a signature of the fact that those modes are
335: somehow localised over a chain (see Fig. \ref{fig:maxproj_n_chain}) and do
336: not involve many molecules, in agreement with the value of the
337: participation ratio (see Fig. \ref{fig:part_n_radius}). The present result
338: unambiguously confirms the optical-like character of the mode at 50
339: ps${}^{-1}$ present in the longitudinal and transverse current spectra
340: reported in Ref. \cite{bstv}.
341:
342: In a previous investigation of the collective properties of liquid HF
343: \cite{bstv} it has been shown that the longitudinal spectra have a peak,
344: $\omega_{max}$, at low frequency, which changes linearly with the
345: wavevector $k$. Its value is found to remain lower that 10 ps${}^{-1}$. The
346: corresponding phase velocity $v_{ph} = \omega_{max} / k$ turns out to be
347: somewhat higher than the ultrasonic (hydrodynamic) counterpart, but
348: compatible with the presence of a positive anomalous dispersion as in the
349: case of monatomic liquids (e.g. liquid metals). This feature is normally
350: interpreted in terms of overdamped acoustic modes propagating in the system
351: at wavevectors well beyond the hydrodynamic range. The results of the
352: present INM analysis are consistent with such an interpretation, since they
353: show that in the range below 10 ps${}^{-1}$: 1) the partecipation ratio is
354: large (see Fig. \ref{fig:part_n_radius}), 2) the short time displacements
355: of nearest and next-to-nearest neighbors are in phase as one would expect
356: from molecules participating to a collective (acoustic-like) motion (see
357: Fig. \ref{fig:psi}).
358:
359: In conclusion the INM analysis of HF has revealed how the dynamical features
360: are affected by the presence of topological chains of hydrogen-bonded
361: molecules. In certain ranges of frequencies there is a strict correspondence
362: between INM modes and chains, where these modes are found to be
363: localised. This result has allowed to give a sound interpretation to a
364: feature present both in the VACF and collective currents, namely a
365: peak at 50 ps${}^{-1}$ in the corresponding spectra. Such a
366: characteristic can in fact be assigned to the ``stretching'' of hydrogen
367: bonding of first neighbouring molecules. This finding reveals the
368: ``optical'' character of the collective mode present in the currents. Low
369: frequency modes are found to be spread over the whole system
370: involving several different chains, however they are
371: confined to frequencies not much higher than 10 ps${}^{-1}$ in agreement
372: with previous findings derived from the analysis of the longitudinal and
373: transverse currents.
374: The very high frequency (rotational) modes ($\simeq$ 200 ps${}^{-1}$)
375: are demonstrated to be confined over the highly stable tetramer chains.
376:
377: Finally, we wish to stress the fact that our present study has the
378: potential to pave the way for an unambiguous interpretation of the
379: dynamical feature of other hydrogen bonded systems (e.g. water) which are
380: still a matter of a large debate.
381:
382: %%\bibliography{inm}
383: \begin{thebibliography}{99}
384:
385: \bibitem[*]{email}
386: Author to whom correspondence should be addressed, e-mail:
387: \mbox{\tt vallauri@science.unitn.it}
388:
389: \bibitem{keyes_diff}
390: T. Keyes, J. Phys. Chem. {\bf 101}, 2921 (1997).
391:
392: \bibitem{madden}
393: M. Ribeiro, M. Wilson, and P. Madden, J. Chem. Phys. {\bf 109}, 9859 (1998).
394:
395: \bibitem{sciortino}
396: F. Sciortino and P. Tartaglia, Phys. Rev. Lett. {\bf 78}, 2385 (1997).
397:
398: \bibitem{mazzacurati}
399: V. Mazzacurati, G. Ruocco, and M. Sampoli, Europhys. Lett. {\bf 34}, 681
400: (1996).
401:
402: \bibitem{sv}
403: G. Sutmann and R. Vallauri, J. Phys.: Cond. Mat. {\bf 10}, 9231 (1998).
404:
405: \bibitem{buchner}
406: M. Buchner, B. Landanyi, and R. Stratt, J. Chem. Phys. {\bf 97}, 8522 (1992).
407:
408: \bibitem{keyes_cs2}
409: R.~L. Murry, J.~T. Fourkas, W.-X. Li, and T. Keyes, J. Chem. Phys. {\bf 110},
410: 10423 (1999).
411:
412: \bibitem{water_inm}
413: M. Cho {\it et~al.}, J. Chem. Phys. {\bf 100}, 6672 (1994).
414:
415: \bibitem{bstv}
416: D. Bertolini, G. Sutmann, A. Tani, and R. Vallauri, Phys. Rev. Lett. {\bf 81},
417: 2080 (1998).
418:
419: \bibitem{bbstv}
420: U. Balucani, D. Bertolini, G. Sutmann, and A. Tani, J. Chem. Phys. {\bf 111},
421: 4663 (1999).
422:
423: \bibitem{jv_nonpol}
424: P. Jedlovszky and R. Vallauri, Mol. Phys. {\bf 92}, 331 (1997).
425:
426: \bibitem{keyes_irc}
427: T. Keyes, J. Chem. Phys. {\bf 110}, 10410 (1999).
428:
429: \bibitem{jv_hfstruct}
430: P. Jedlovszky and R. Vallauri, Mol. Phys. {\bf 93}, 15 (1998).
431:
432: \bibitem{keyes_tos}
433: R. Murray, J. Fourkas, and T. Keyes, J. Chem. Phys. {\bf 109}, 2814 (1998).
434:
435: \bibitem{nr}
436: W. Press, S. Teutolsky, W. Vetterling, and B. Flannery, in {\em Numerical
437: recipes in C}, 2nd ed. (Cambridge University Press, Cambridge, 1992).
438:
439: \bibitem{hf_solid}
440: J.S. Kittelberger and D.F. Hornig, J. Chem. Phys. {\bf 46}, 3099 (1967).
441:
442: \end{thebibliography}
443:
444: %% Figures
445:
446:
447: \end{document}
448: