1: \documentclass[showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
2:
3: \usepackage{graphicx}% Include figure files
4: \usepackage{dcolumn}% Align table columns on decimal point
5: \usepackage{bm}% bold math
6: \bibliographystyle{apsrev}
7: \begin{document}
8:
9: \preprint{LA-UR 05-4348}
10:
11: \title{Application of Vibration-Transit Theory of Liquid Dynamics to the Brillouin Peak Dispersion Curve}
12:
13: \author{Duane C. Wallace}
14: \author{Giulia De Lorenzi-Venneri}
15: \author{Eric D. Chisolm}
16: \affiliation{Theoretical Division, Los Alamos National Laboratory,
17: Los Alamos, New Mexico 87545}
18:
19: \date{\today}
20:
21: \begin{abstract}
22: The Brillouin peak appears in the dynamic structure factor $S(q,\omega)$, and
23: the dispersion curve is the Brillouin peak frequency as function of $q$. The
24: theoretical function underlying $S(q,\omega)$ is the density autocorrelation function $F(q,t)$.
25: A broadly successful description of time correlation functions is provided by
26: mode coupling theory, which expresses $F(q,t)$ in terms of processes through
27: which the density fluctuations decay. In contrast, vibration-transit (V-T) theory is a Hamiltonian formulation of
28: monatomic liquid dynamics in which the motion consists of vibrations within
29: a many-particle random valley, interspersed with nearly instantaneous transits
30: between such valleys. Here, V-T theory is applied to $S(q,\omega)$. The theoretical
31: vibrational contribution to $S(q,\omega)$ is the sum of independent scattering cross
32: sections from the normal vibrational modes, and contains no explicit reference to
33: decay processes. For a theoretical model of liquid Na, we show that the
34: vibrational contribution with no adjustable parameters gives an excellent
35: account of the Brillouin peak dispersion curve, as compared to MD calculations
36: and to experimental data.
37: \end{abstract}
38:
39: \pacs{05.20.Jj, 63.50.+x, 61.20.Lc, 61.12.Bt}% PACS, the Physics and Astronomy
40: % Classification Scheme.
41: \keywords {Liquid Dynamics, Inelastic Neutron Scattering, Dispersion Relations, Mode Coupling Theory,
42: V-T Theory}
43: \maketitle
44:
45:
46: The purpose of this paper is to present a new theoretical explanation for
47: the Brillouin peak dispersion curve in a monatomic liquid. The Brillouin
48: peak represents inelastic neutron or photon scattering, and appears in
49: constant-$q$ graphs of the dynamic structure factor $S(q,\omega)$~\textit{vs}~$\omega$.
50: In monatomic liquids, the Brillouin peak is observable for $0 < q \lesssim q_{m}$, where
51: $q_{m}$ is at the maximum of the first peak in $S(q)$. At given $q$, the frequency
52: at the maximum of the Brillouin peak is $\Omega(q)$, and the dispersion
53: curve is $\Omega$~\textit{vs}~$q$. The new explanation results from the application of
54: vibration-transit (V-T) theory of liquid dynamics to the calculation
55: of $S(q,\omega)$. This theory was originally developed to explain the
56: equilibrium thermodynamic properties of monatomic liquids \cite{DWPRE56}. Now
57: that V-T theory has been shown to give a highly accurate account of
58: the thermodynamic properties of elemental liquids, with no adjustable
59: parameters \cite{DWPRE56,ChWJPCM01}, it is of interest to investigate its application to the
60: nonequilibrium problem of dynamic response.
61:
62: The development of microscopic theories of liquid dynamics is described in
63: the monographs of Boon and Yip \cite{BYbook}, Hansen and McDonald \cite{HMCDbook}, and
64: Balucani and Zoppi \cite{BZbook}. The theory most successful in accounting for time correlation
65: functions is mode coupling theory \cite{Golu75,GoSj92}. Here we shall limit our attention to
66: the density autocorrelation function $F(q,t)$ and its Fourier transform $S(q,\omega)$.
67: Mode coupling theory works with the generalized Langevin equation for $F(q,t)$,
68: and expresses the memory function in terms of processes through which
69: density fluctuations decay \cite{BYbook, HMCDbook, BZbook}. In the viscoelastic approximation, the
70: memory function decays with a $q$-dependent relaxation time \cite{BYbook, HMCDbook, BZbook}. This
71: approximation provides a good fit to the combined experimental data \cite{CRPRL74} and
72: MD data \cite{ARPRL74,ARPRA74} for the Brillouin peak dispersion curve in liquid Rb \cite{CLRPP38} (see also Fig.~9.2
73: of \cite{HMCDbook}). Going beyond the viscoelastic approximation,
74: Bosse et al \cite{BGLPRA17a, BGLPRA17b} constructed a self-consistent theory for the
75: longitudinal and transverse current fluctuation spectra, each expressed in
76: terms of relaxation kernels approximated by decay integrals which couple
77: the longitudinal and transverse excitations. This theory is in good overall
78: agreement with extensive neutron scattering data and MD calculations
79: for Ar near its triple point \cite{BGLPRA17b}. The theory was developed further by
80: Sj\"ogren \cite{Sjog80, Sjog80b}, who separated the memory function into a binary
81: collision part, approximated with a Gaussian ansatz, and a more collective
82: tail represented by a mode coupling term. For liquid Rb, this theory
83: gives an ``almost quantitative'' agreement with results from neutron scattering experiments
84: \cite{CRPRL74} and MD calculations \cite{ARPRL74,ARPRA74}. More recently, inelastic x-ray scattering
85: measurements have been done for the
86: light alkali metals Li \cite{R&c002} and Na \cite{SBRSb,Naexp}. In analyzing these data,
87: Scopigno et al \cite{SBRSb,Naexp,SRSVPRE02,SRSVPM02,SBRS,SBRSa}, include two or three decay
88: channels in the memory function, with each decay process represented by
89: adjustable $q$-dependent relaxation times and decay strengths. The resulting
90: fits to $S(q,\omega)$ are excellent, both for the experimental data and for MD
91: calculations \cite{SBRSb,Naexp,SRSVPRE02,SRSVPM02,SBRS,SBRSa}.
92:
93: We shall now present a summary of V-T theory. From the outset, this
94: theory was based on experimental data for the constant-volume entropy of
95: melting of the elements, and the constant-volume atomic-motion specific
96: heat of the elemental liquids at melt \cite{DWPRE56}. These data led us to the
97: conjecture that the potential energy surface consists entirely of intersecting
98: nearly-harmonic many-particle potential energy valleys; that these
99: valleys belong to two classes, the symmetric (microcrystalline) and
100: random classes; that the random valleys all have the same potential energy
101: surface in the thermodynamic limit; and that the random valleys are
102: by far the most numerous, and therefore dominate the statistical mechanics
103: of the liquid state. This conjecture has since been verified by computer
104: studies \cite{WCPRE59, CWPRE59}. The corresponding atomic motion consists of vibrations
105: within a single random valley, interspersed with motions across the
106: intersections between valleys, called transits. The vibrational contribution
107: dominates the thermodynamic properties. This contribution is modeled
108: by the motion of the system in a single harmonic random valley
109: extended to infinity, and having no intersection with another valley. The
110: Hamiltonian then consists of a set of analytically tractable
111: extended random valley Hamiltonians, plus corrections for transits and
112: for anharmonicity \cite{DWPRE56, DWbook2}. The role of transits in equilibrium is merely
113: to allow the system to move among all the random valleys, and hence
114: to exhibit the correct liquid entropy. The fact that transits contribute
115: little else to equilibrium thermodynamic properties is interpreted to indicate
116: that transits are nearly instantaneous \cite{DWPRE56, WChCPRE64}. Transits have a more
117: direct role in nonequilibrium properties, since transits are responsible
118: for diffusive motion.
119:
120: Since the vibrational contribution dominates the thermodynamic properties of the liquid,
121: we expect it to be important in the nonequilibrium properties as well.
122: This is the hypothesis to be tested. In fact, in the glass state,
123: where the system is frozen into a single random valley, the vibrational
124: contribution should be the \textit{only} contribution. We shall first check this
125: expectation in the glass, as its confirmation is required before we can
126: apply V-T theory to the liquid.
127:
128: We have previously discussed V-T theory for a general time correlation
129: function, and have derived the vibrational contribution to the dynamic response
130: functions \cite{WDCh05}. We work with a system containing $N$ atoms in a cubic volume,
131: with the motion governed by classical mechanics and by periodic boundary
132: conditions. For motion in an extended random valley, the atomic positions
133: $\bm{r}_{K}(t)$ are written
134: \begin{equation} \label{eq1}
135: \bm{r}_{K}(t)=\bm{R}_{K}+\bm{u}_{K}(t),
136: \end{equation}
137: for $K=1, \dots,N$,where $\bm{R}_{K}$ is the equilibrium position of atom $K$,
138: and $\bm{u}_{K}(t)$ is its displacement. The vibrational contribution to $F(q,t)$ is $F_{vib}(q,t)$, and
139: in the one-mode approximation this is denoted $F_{1}(q,t)$. It is convenient to
140: separate the constant $F_{vib}(q,\infty)$, and write
141: \begin{equation} \label{eq2}
142: F_{1}(q,t)=F_{vib}(q,\infty)+\left[F_{1}(q,t)-F_{vib}(q,\infty) \right],
143: \end{equation}
144: where
145: \begin{equation} \label{eq3}
146: F_{vib}(q,\infty) = \frac{1}{N} \left < \sum_{KL} e^{-i \bm{q}\cdot \bm{R}_{KL}} e^{-W_{K}(\bm{q})} e^{-W_{L}(\bm{q})}
147: \right >_{\bm{q}^{\ast}},
148: \end{equation}
149: \begin{equation}\label{eq4}
150: F_{1}(q,t)-F_{vib}(q,\infty)=
151: \frac {1}{N} \Bigg < \sum_{KL}e^{-i \bm{q}\cdot \bm{R}_{KL}}
152: \;e^{-W_{K}(\bm{q})} e^{-W_{L}(\bm{q})}
153: \;\left < \bm{q} \cdot \bm{u}_{K}(t)\;\bm{q} \cdot \bm{u}_{L}(0)\right >_{h}
154: \Bigg >_{\bm{q}^{\ast}} .
155: \end{equation}
156: Here $\bm{R}_{KL}=\bm{R}_{K}-\bm{R}_{L}$, $W_{K}(\bm{q})$ is the Debye-Waller factor for atom $K$,
157: $<\dots>_{h}$ is the harmonic average over the vibrational motion, and $<\dots>_{\bm{q}^{\ast}}$ is
158: the average over the star of $\bm{q}$. The atomic displacements are analyzed into
159: normal vibrational modes labeled $\lambda$, for $\lambda=1, \dots,3N$, each mode having frequency
160: $\omega_{\lambda}$ and eigenvector components $\bm{w}_{K,\lambda}$. Then $S_{vib}(q,\omega)$,
161: the Fourier transform of $F_{1}(q,t)$, is
162: \begin{equation} \label{eq5}
163: S_{vib}(q,\omega) = F_{vib}(q,\infty)\delta(\omega)+S_{1}(q,\omega),
164: \end{equation}
165: where
166: \begin{equation} \label{eq6}
167: S_{1}(q,\omega)=\frac{3kT}{2M}\frac{1}{3N}\sum_{\lambda}f_{\lambda}(q)[\delta(\omega+\omega_{\lambda})+
168: \delta(\omega-\omega_{\lambda})], \\
169: \end{equation}
170: \begin{equation} \label{eq7}
171: f_{\lambda}(q)=\frac{1}{\omega_{\lambda}^{2}}\left < \left|\sum_{K}e^{-i\bm{q}\cdot\bm{R}_{K}}e^{-W_{K}(\bm{q})}
172: \bm{q}\cdot\bm{w}_{K\lambda}\right|^{2}\right >_{\bm{q}^{\ast}}.\\
173: \end{equation}
174: As a sum over normal modes, the Debye-Waller factors are
175: \begin{equation} \label{eq8}
176: W_{K}(\bm{q})=\sum_{\lambda}\frac{kT(\bm{q}\cdot\bm{w}_{K\lambda})^{2}}{2M\omega_{\lambda}^{2}}.
177: \end{equation}
178:
179:
180: Eq.~(\ref{eq6}) for $S_{1}(q,\omega)$ has an obvious physical meaning: it is the sum over all vibrational
181: modes of single-mode scattering at momentum transfer $\hbar q$, with creation
182: or annihilation of energy $\hbar \omega_{\lambda}$ in mode $\lambda$. If crystal phonon eigenvectors
183: are used in Eq.~(\ref{eq7}), and if the average over $\bm{q}^{\ast}$ is omitted, one recovers the crystal
184: results of Lovesey \cite{Lovbook}, or of Glyde \cite{Glybook}, for $S_{1}(\bm{q},\omega)$.
185:
186:
187: \begin{figure}[t]
188: \includegraphics[height=3.0in,width=3.0in]{fig_1.eps}
189: \caption{\label{Sqw_T3} $S(q,\omega)$ for motion in a single random valley at $q=0.29711$ a$_{0}^{-1}$ and 24~K:
190: full line is MD,
191: points are theory, Eq.~(\ref{eq6}).}
192: \end{figure}
193:
194:
195:
196: The system we study is a model representing metallic Na, derived from
197: electronic structure theory (see e.g. \cite{DWbook2}). This model gives an excellent account
198: of the experimental phonon dispersion curves of bcc Na, and of the corresponding
199: thermal properties of the crystal (\cite{DWbook2}, Secs.~14, 17, 19). Through the spectrum
200: of the random valley vibrational mode frequencies $\omega_{\lambda}$, the model also accounts
201: for the experimental entropy of liquid Na (\cite{DWbook2}, Sec.~23 and Eq.~(24.2)).
202: We shall use this model to evaluate the equations of V-T theory, and we shall
203: compare the results with MD calculations for the same model and with
204: experimental data on liquid Na.
205: The same approach of comparing theory with MD was
206: successfully used by Glyde, Hansen, and Klein \cite{GHKPRB16} to evaluate the
207: contribution of anharmonicity to $S(\bm{q},\omega)$ for crystalline K.
208:
209:
210: Our first comparison is for the vibrational contribution alone, at a
211: representative $q$. For this purpose, the MD system was equilibrated at
212: 24~K, where transits are frozen out and the system moves in a single random
213: valley. The MD result for $S(q,\omega)$ is shown by the solid line in Fig.~1.
214: Theory was evaluated from Eqs.~(\ref{eq6})-(\ref{eq8}), and is shown by the points in
215: Fig.~1. The $\delta(\omega)$ term in Eq.~(\ref{eq5}) expresses elastic scattering, and is not
216: indicated in the figure. The agreement between V-T theory and MD in Fig.~1 is excellent to the smallest
217: significant detail. Hence the same comparison for the liquid can be seen to reveal
218: the effect of transits.
219:
220: \begin{figure}[t]
221: \includegraphics[height=3.0in,width=3.0in]{fig_2.eps}
222: \caption {\label{Sqw_T4} $S(q,\omega)$ at $q=0.29711$ a$_{0}^{-1}$ and 395~K: line is MD for the liquid,
223: points are the vibrational contribution alone, Eq.~(\ref{eq6}). What the vibrational contribution
224: gets right is the location of the Brillouin peak.}
225: \end{figure}
226:
227:
228:
229: In previous work, Mazzacurati et al.\cite{Mazz96} (see also \cite{R&cPRL00}) have shown
230: that a vibrational analysis is in excellent agreement with MD calculations of $S(q,\omega)$
231: for a Lennard-Jones glass. We have done this comparison again, in Fig.~1, because
232: we want to study the \textit{liquid}, and V-T theory requires confirmation
233: that the valley studied is random, and not some other symmetry.
234:
235: We shall now make the same comparison for the liquid state, at 395~K.
236: For our system, this temperature is some 24~K above melting. The vibrational
237: contribution to $S(q,\omega)$, again calculated from Eqs.~(\ref{eq6})-(\ref{eq8}), is shown by
238: the points in Fig.~2. For this contribution, the temperature dependence is
239: explicit in Eqs.~(\ref{eq6}) and (\ref{eq8}), and is dominated in Fig.~2 by the linear
240: dependence in Eq.~(\ref{eq6}). On the other hand, the MD system at 395~K is
241: transiting rapidly among the random valleys. The MD result, the solid line
242: in Fig.~2, displays the well-known Rayleigh-Brillouin spectrum \cite{BYbook, HMCDbook, BZbook},
243: where the Rayleigh peak arises from the elastic term in Eq.~(\ref{eq5}).
244: What is missing from the V-T curve is the transit contribution. From the
245: comparison with the MD curve, which includes transits, we can see that the
246: effect of transits is to broaden both the Rayleigh and the Brillouin peaks,
247: but not to shift them. That these peaks are not shifted results from the
248: nearly instantaneous character of transits.
249:
250:
251: Three additional properties related to Fig.~2 are of interest. First, the
252: vibrational contribution alone gives an accurate account of the MD results for
253: the area of the Rayleigh peak, and for the area of the Brillouin peak. This
254: property of V-T theory is also understood from the nearly instantaneous
255: character of transits. Second, the vibrational contribution alone also gives
256: around half of the total width of the Brillouin peak. And third, a
257: parameterized model for $S(q,\omega)$, in the spirit of Zwanzig's model for
258: the velocity autocorrelation function \cite{ZJCP79}, is obtained by multiplying the two
259: terms on the right side of Eq.~(\ref{eq2}) by the relaxation functions $e^{-\alpha_{1} t}$
260: and $e^{-\alpha_{2} t}$, respectively. Since they represent the transit-induced decay
261: of $F(q,t)$, the relaxation rates $\alpha_{1}$ and $\alpha_{2}$ are expected to be approximately
262: the mean transit rate in the liquid \cite{DW05}. For the example shown in Fig.~2, and also
263: for the other $q$ we have studied, a good fit to the MD $S(q,\omega)$ is obtained
264: from this relaxation model \cite{DW05}, with $\alpha_{1}$ and $\alpha_{2}$ close to the known
265: mean transit rate \cite{DWPRE58,ChCWPRE63}.
266:
267: \begin{figure}[h]
268: \includegraphics[height=3.0in,width=3.0in]{fig_3.eps}
269: \caption {\label{wq} Brillouin peak dispersion curve for liquid sodium: circles are MD at 395~K, points
270: are the vibrational contribution from V-T theory at 395~K, and triangles are
271: x-ray scattering data at 390~K \cite{SBRSb,Naexp}. }
272: \end{figure}
273:
274:
275: We shall now present the main result of this paper. Since transits
276: broaden the Brillouin peak, without shifting it, the Brillouin peak
277: dispersion curve is determined by the vibrational contribution alone.
278: In Fig.~3, $\Omega(q)$ is
279: graphed for the entire range of $q$ for which the Brillouin peak is observable in our system, comparing
280: results from MD in the liquid at
281: 395~K with the vibrational contribution alone at the same temperature. The
282: relative rms difference between the $\Omega_{MD}$ and $\Omega_{vib}$ points is 2$\%$, a difference on the
283: order of numerical errors. Experimental data for liquid Na at 390~K \cite{SBRSb,Naexp} are
284: also graphed, and our theoretical $\Omega_{vib}$ points fall within the experimental
285: error bars. The relative rms difference of our seven $\Omega_{vib}$ points from a smooth curve
286: fitting the experimental points is 5$\%$.
287:
288:
289:
290:
291:
292: In summary: (a) the vibrational contribution to $S(q,\omega)$ for a monatomic
293: liquid is the sum of independent inelastic scattering cross sections from
294: each of the random valley normal vibrational modes; (b) in contrast to
295: mode coupling theory, this formulation of $S(q,\omega)$ is not based on
296: explicit decay processes; and (c) for liquid Na, comparison with MD
297: calculations and with experimental data shows that the vibrational
298: contribution accurately accounts for the location of the Brillouin peak,
299: and accounts for half the width of the Brillouin peak as well.
300: Our results are consistent with the findings of Sciortino and Tartaglia \cite{ST97}
301: for water, that the vibrational motion is quite harmonic, and that the fast
302: decay of correlation functions is due to mode dephasing. Here, the dephasing
303: gives rise to the natural width of the vibrational contribution to
304: the Brillouin peak, as shown in Figs.~1 and 2.
305:
306:
307:
308:
309: The authors gratefully acknowledge Tullio Scopigno and R. M. Yulmetyev for providing
310: the original experimental data, Brad Clements for collaboration, and
311: Francesco Venneri for technical
312: advice. This work was supported by the US DOE through contract W-7405-ENG-36.
313:
314: \bibliography{VTTref}
315:
316: \end{document}
317:
318:
319:
320: