1: %\documentclass[preprint,preprintnumbers,nofootinbib,aps]{revtex4}

2: \documentclass[twocolumn,showpacs,preprintnumbers,nofootinbib,aps]{revtex4}

3: \usepackage{graphicx,bm,latexsym,amssymb} 

4: \usepackage{amsmath,epsf,psfig}

5: 

6: 

7: \begin{document}

8: 

9: \newcommand{\hh}{\hspace{3mm}}

10: 

11: \newcommand{\sign}{\mbox{sign}}

12: \newcommand{\ssign}{\mbox{\scriptsize sign}}

13: \newcommand{\smin}{{\mbox{\scriptsize min}}}

14: \newcommand{\smax}{{\mbox{\scriptsize max}}}

15: \newcommand{\sch}{{\mbox{\scriptsize ch}}}

16: \newcommand{\tmin}{{\mbox{\tiny min}}}

17: 

18: \newcommand{\la}{\left\langle}

19: \newcommand{\ra}{\right\rangle}

20: 

21: \newcommand{\dla}{\la \!   \la}

22: \newcommand{\dra}{\ra \!   \ra}

23: 

24: \newcommand{\prtl}{\partial}

25: \newcommand{\we}{\widetilde}

26: 

27: \newcommand{\smfp}{{\mbox{\scriptsize mfp}}}

28: \newcommand{\smol}{{\mbox{\scriptsize mol}}}

29: \newcommand{\sph}{{\mbox{\scriptsize ph}}}

30: \newcommand{\sinhom}{{\mbox{\scriptsize inhom}}}

31: \newcommand{\sneigh}{{\mbox{\scriptsize neigh}}}

32: \newcommand{\srlxn}{{\mbox{\scriptsize rlxn}}}

33: \newcommand{\svibr}{{\mbox{\scriptsize vibr}}}

34: \newcommand{\smicro}{{\mbox{\scriptsize micro}}}

35: \newcommand{\selast}{{\mbox{\scriptsize elast}}}

36: \newcommand{\select}{{\mbox{\scriptsize elect}}}

37: \newcommand{\seq}{{\mbox{\scriptsize eq}}}

38: \newcommand{\scr}{{\mbox{\scriptsize cr}}}

39: \newcommand{\sT}{{\mbox{\scriptsize T}}}

40: \newcommand{\sTLS}{{\mbox{\scriptsize TLS}}}

41: \newcommand{\sd}{{\mbox{\scriptsize d}}}

42: \newcommand{\sext}{{\mbox{\scriptsize ext}}}

43: \newcommand{\scav}{{\mbox{\scriptsize cav}}}

44: \newcommand{\bmu}{\bm \mu}

45: \newcommand{\bE}{\bm E}

46: \newcommand{\bD}{\bm D}

47: \newcommand{\bd}{\bm d}

48: \newcommand{\br}{\bm r}

49: \newcommand{\bS}{\bm S}

50: \newcommand{\bg}{\bm g}

51: \newcommand{\wF}{\widetilde{F}}

52: 

53: \title{Electrodynamics of Amorphous Media at Low Temperatures}

54: 

55: \author{Vassiliy Lubchenko and Robert J. Silbey} 

56: \affiliation{Department of Chemistry, Massachusetts Institute of

57: Technology Cambridge, MA 02139}

58: 

59: \author{Peter G. Wolynes} \affiliation{Departments of Chemistry and

60: Biochemistry, and Physics, University of California, San Diego, CA

61: 92093-0371}

62: 

63: 

64: 

65: \date{April 29, 2005}

66: 

67: \begin{abstract}

68: 

69: Amorphous solids exhibit intrinsic, local structural transitions, that

70: give rise to the well known quantum-mechanical two-level systems at

71: low temperatures.  We explain the microscopic origin of the electric

72: dipole moment of these two-level systems: The dipole emerges as a

73: result of polarization fluctuations between near degenerate local

74: configurations, which have nearly frozen in at the glass transition.

75: An estimate of the dipole's magnitude, based on the random first order

76: transition theory, is obtained and is found to be consistent with

77: experiment. The interaction between the dipoles is estimated and is

78: shown to contribute significantly to the Gr\"{u}neisen parameter

79: anomaly in low $T$ glasses. In completely amorphous media, the dipole

80: moments are expected to be modest in size despite their collective

81: origin.  In partially crystalline materials, however, very large

82: dipoles may arise, possibly explaining the findings of Bauer and

83: Kador, J. Chem. Phys. {\bf 118}, 9069 (2003).

84: 

85: 

86: 

87: \end{abstract}

88: 

89: %\pacs{02.50.-r, 05.40.Fb, 05.30.Pr, 05.45.Df}

90: 

91: \maketitle

92: 

93: \section{Introduction}

94: 

95: Glasses are frozen liquids and thus lack long-range order, yet the

96: differences in material properties between amorphous materials and

97: crystals are often rather subtle. Crystalline samples themselves are

98: rarely flawless and thus contain a number of imperfections such as

99: point defects, dislocations or grain boundaries of various

100: sorts. These tend to further mask the difference. The size of defects

101: in crystals ranges over many scales, while in glasses, the static

102: heterogeneity in the atomic arrangement appears comparable to the

103: molecular size itself.  Simple molecular glasses thus seem perfect

104: candidates for description as isotropic continuum, at long enough wave

105: lengths. For instance at cryogenic temperatures, when the de Broiglie

106: wave-length of a thermal phonon at $\sim$ 1K exceeds the lattice

107: spacing by three orders of magnitude or so, continuum theory would be

108: thought to hold to high accuracy. Yet surprisingly, there clearly

109: exist degrees of freedom numbering in great excess of the Debye

110: density of states, leading to extra heat capacity and phonon

111: scattering in all amorphous materials \cite{ZellerPohl}. Here we

112: examine the electrodynamics of these degrees of freedom.

113: 

114: Since Rayleigh scattering is too weak to account for the observed

115: magnitude of sound attenuation in glasses, internal resonances must be

116: involved, in the form of {\em anharmonic} structural rearrangments, in

117: order to explain the data. The well known, empirical two-level system

118: (TLS) theory presumes such resonances exist

119: \cite{AHV,Phillips}. Simply postulating a flat energy spectrum and a

120: frequency independent coupling to the phonons accounts for all the

121: gross features of the low $T$ anomalies (for reviews, see

122: \cite{LowTProp,HunklingerRaychaudhuri,Esquinazi}).  Direct microscopic

123: evidence of the two-level nature of such entities comes both from the

124: phonon echo experiments \cite{GG} and, relatively recently, from

125: single-molecule experiments at cryogenic temperatures (see

126: e.g. \cite{BTLBO}). At these temperatures, the TLS picture is

127: internally consistent in so far as the structural transitions (ST) can

128: be defined as {\em local}, and thus, tautologically, sufficiently {\em

129: weakly} interacting. One may therefore speak of a multilevel system at

130: the location of each transition whose behavior reduces to a TLS

131: behavior at low enough $T$. We may call this a tunneling center

132: (TC). In 1986, Freeman and Anderson \cite{FreemanAnderson} showed that

133: the magnitude of the TLS density of states is apparently correlated

134: with the phonon coupling. This results in a universality of the ratio

135: of the phonon mean free path $l_\smfp$ to its wave-length $\lambda$:

136: $l_\smfp/\lambda \sim 150$, for all insulating glasses at $T \lesssim$

137: 1K. This universality seems hardly coincidental \cite{YuLeggett},

138: however understanding the origin of the universality requires a

139: microscopic picture of molecular motions in glasses. (The large size

140: of the factor $\sim$150, too, was a puzzle \cite{YuLeggett}.)

141: 

142: The Random First Order Transition (RFOT) Theory of the glass

143: transition \cite{KTW, MCT, MCT1, XW, LW_soft} provides an appropriate

144: microscopic picture of the motions in glass. Most commercial and

145: laboratory glasses are made by quenching supercooled melts. In the

146: deeply supercooled regime, most liquid motions are activated

147: transitions between distinct aperiodic states of comparable energy,

148: during which the current structural arrangement in a {\em local}

149: region is replaced by another, quite different arrangement that

150: nevertheless fits its environment. The size of the reconfigurable

151: region, $\xi$, grows with decreasing temperature, and reaches about

152: 5-6 molecular units across by $T_g$, i.e. the glass transition

153: temperature corresponding to the 1 hour time scale. This size is

154: predicted to be universal, within logarithmic accuracy, for {\em all}

155: substances. At any point in time, above $T_g$, the liquid can be

156: thought of as a mosaic of such cooperative regions \cite{XW}, most of

157: nearly the same size, but otherwise with distributed barrier heights

158: and transition energies. Upon freezing, a particular mosaic pattern

159: sets in and undergoes relatively slow changes, called aging. The aging

160: speed depends on the quench depth \cite{LW_aging}. A sufficient

161: fraction of the structural transitions have small enough energy change

162: and barriers so that when they occur, they can account for the

163: cryogenic anomalies, some of which were mentioned above: the density

164: of states and the universality of phonon scattering \cite{LW}, the

165: Boson Peak \cite{LW_BP}, but also the anomalous Gr\"{u}neisen

166: parameter, the so called ``fast'' TLS systems and more

167: \cite{LW_thesis,LW_RMP}.  According to the RFOT theory, the

168: universality of the $l_\smfp/\lambda$ ratio directly follows from the

169: universal cooperative region size $(\xi/a)^3 \sim 200$ at the glass

170: transition temperature $T_g$, where $a$ is the molecular length

171: scale. During a structural transition, a relatively large, $\sim 200$,

172: compact set of small units moves in a stage-wise fashion. This

173: corresponds to the motion of the domain wall, which separates the two

174: alternative arrangements, through the compact region.  At cryogenic

175: temperatures, these motions occur by tunneling. Despite their

176: collective nature, such tunneling events are possible because of the

177: enormous multiplicity of alternative structural states and of

178: low-barrier paths connecting pairs of states: an amorphous sample

179: actually resides in a high energy density state, well above its lowest

180: energy, perfect crystalline state. Consistent with the facility of

181: tunneling is the smallness of individual atomic displacements during

182: each transition. Their amplitude is roughly equal to the Lindemann

183: length $d_L$. This length is typically one tenth of the characteristic

184: lattice spacing $a$ and is nearly the same for all substances. The

185: precise identity of the ``molecular unit'', or ``bead'' depends on the

186: specific substance, but usually corresponds to a few atoms. (See

187: \cite{LW_soft} for a detailed discussion.)

188: 

189: In this article, we use the microscopic picture of the two-level

190: systems provided by the RFOT theory to estimate the coupling of the

191: transitions to external electric fields. Clearly, such a coupling must

192: be present because individual molecular bonds, that possess electric

193: dipoles, rotate during transitions.  Since these couplings directly

194: enter into spectral hole-burning experiments \cite{Maier} and can also

195: be directly probed in single molecule experiments \cite{BTLBO}, it is

196: important to know how much collective excitations interact

197: electrodynamically with probes and external electric fields.

198: 

199: \section{Interaction of a Single Tunneling Center with External Field}

200: 

201: \subsection{Many-Body Origin of the Transition-Induced Dipole Moment}

202: \label{Motivation}

203: 

204: To set the stage, let us briefly review the assumptions of the

205: traditional molecular models of dielectric response of insulating

206: media.  One often assigns an electric dipole value to an individual

207: molecule, or to a molecular bond connecting distinct atoms in a

208: condensed phase.  In a dilute liquid made up of polar molecules, the

209: medium polarizes in a field since the dipoles prefer to orient along

210: the field's direction at the cost of losing their rotational

211: freedom. Even without permanent electric dipoles, a dielectric

212: response occurs due to polarizability: An external field mixes in

213: higher energy molecular orbitals, which generally lack inversion

214: symmetry. Classically, this quantum mechanical response can be

215: imitated as two harmonically bound opposite charges that separate

216: after a field is turned on. In a polar substance, this polarizability

217: also changes the {\em length} or {\em orientation} of the permanent

218: dipoles.

219:   

220: While in a fluid the dipoles can freely reorient, during a structural

221: transition (ST) in glass, the dipole is restrained: Each individual

222: bead within a tunneling center, or ``domain'', moves only about the

223: Lindemann length $d_L$, as illustrated in

224: Fig.\ref{domain_dipole}. Suppose, for the sake of argument, one can

225: break up the set of all the beads within the domain into distinct

226: pairs. During a transition, each such pair - and hence the

227: corresponding ``bond'' - rotates about $d_L/a \sim 0.1$ radian. The

228: amorphous lattice generally exhibits no symmetry. There is therefore,

229: typically, excess charge, however small, on each atom.  Assume the

230: effective individual charges remain the same during such a

231: transition. One can thus unambiguously assign a permanent, {\em

232: point-like} electric dipole to each ``bond'' introduced above.  As

233: schematically shown in Fig.\ref{domain_dipole}, a total dipole moment,

234: $ \bmu_\sT = \sum_i \Delta \bmu_i$, may be generated during a

235: structural transition, that would couple to an external electric field

236: $\bE$ with energy $- \bmu_\sT \bE$.

237: \begin{figure}[htbp!]

238: \includegraphics[width=.95\columnwidth]{domain_dipole.eps}

239: %\includegraphics[width=.85\columnwidth]{domain_dipole.pdf}

240:  \caption{\label{domain_dipole} Shown on the left is a fragment of the

241: mosaic of cooperatively reconfiguring regions in the supercooled

242: liquid, with $a$ denoting the lattice spacing (more precisely ``bead''

243: spacing). $\xi$ is the cooperative region size; $d_L$ is the typical

244: bead displacement during a transition. (The shown magnitude of $\xi$

245: corresponds to a temperature near $T_g$ on 1 hour scale.) The two sets

246: of circles - solid and dashed ones - denote two alternative structural

247: states.  The expanded portion shows how rotation of a bond leads to

248: generating an elemental dipole change during a transition, where the

249: partial charges on the two beads are $\pm \zeta q$.}

250: \end{figure} 

251: 

252: In molecular glasses, the bond dipoles are fairly easy to assign.

253: Generally, the assignment of point-like dipole moments to individual

254: bonds is, strictly speaking, non-unique. It would be rather difficult

255: in the case of a highly networked, covalently bonded substance, such

256: as amorphous silica, but as we shall see, other arguments for such

257: systems give similar results. These arguments use the measurable

258: piezoelectric properties of corresponding crystals to unambiguously

259: extract the coupling to fields.  In weakly bonded molecular glasses

260: held together by Van der Waals forces, the point like dipole view is

261: already a good approximation.

262: 

263: With this in mind, an order of magnitude estimate of the dipole moment

264: of a tunneling center can be made: The Coulomb charge on a bead does

265: not exceed a fraction $\zeta < 1$ of an elementary charge $q$, which

266: is close in magnitude to the electron charge $e$: $q \sim e$. An

267: individual electric dipole {\em change} is therefore $\Delta \mu \sim

268: \zeta \mu_\smol (d_L/a)$, where $(d_L/a)$ is the rotation angle, as

269: already discussed, and $\mu_\smol \equiv \zeta q a$ is the elemental

270: dipole magnitude associated with each bond. The number of pairs that

271: reorient in a structural transition is $N_\sd = (\xi/a)^3/2$. The

272: $\bmu_T$ distribution is, of course, centered at the origin. Since the

273: dipole forces are a small part of the energetics of the glass

274: transition, the dipole motions are expected to be only weakly

275: connected with each other.  Therefore the individual dipoles $\Delta

276: \bmu_i$ make up a ``random walk'' of $N_\sd$ steps (in 3D). As a

277: result, the generic value of dipole change for the transition is given

278: by the width of the total displacements during such a walk:

279: \begin{equation}

280: \mu_T \simeq \zeta (q a) [(\xi/a)^3/2]^{1/2} (d_L/a),

281: \label{mu_Tqual}

282: \end{equation} 

283: If the elementary dipole rotations are correlated, one may introduce

284: an additional factor - like the Kirkwood $g$ factor of liquid theory

285: \cite{Kirkwood}.  $g$ is typically of order 2. At first we might

286: imagine a great deal of variability for the quantity

287: $[(\xi/a)^3/2]^{1/2} (d_L/a)$ but in fact it is nearly universally (!)

288: equal to unity for all substances. $a$ is typically a couple of

289: angstroms, implying $q a$ corresponds to $\sim 10$ Debye. ($e \AA

290: \leftrightarrow$ 4.8 Debye.)  $\zeta$ is expected to be well less than

291: unity, with $0.1$ or less being a reasonable generic estimate. We thus

292: obtain that $\mu_T$ is of the order 1 Debye or less, consistent with

293: experiment. Note the magnitude of $\mu_T$ is rather modest - only of

294: order the size of a typical individual dipole moment - despite the

295: large number of particles constituting a domain. The relative

296: smallness of the two-level system dipole moment is due to the small

297: deflection angle $\sim 0.1$, and the apparent smallness of the partial

298: charge $\zeta q$. There are deep reasons for both: The former stems

299: from the particular magnitude of atomic displacements during a

300: transition: it is equal to the typical thermal displacement at the

301: mechanical stability edge, i.e. the Lindemann length \cite{XW}.  The

302: latter is probably related to the intrinsic difficulty of making

303: ionically bonded aperiodic structures, which imposes an upper bound on

304: the value of $\zeta$; this will be discussed in due time.

305:  

306: The sublinear scaling of the dipole moment with the domain volume

307: $\xi^3$, in Eq.(\ref{mu_Tqual}), is worth noting: The tunneling

308: transition dipole moment is not a bulk response, and its relation to

309: the material's average bulk dielectric properties, as encoded e.g. in

310: the substance's dielectric susceptibility $\epsilon(\omega)$ in the

311: fluid phase, is not immediately obvious. While it is ultimately the

312: deflections of the same elemental dipoles that both give rise to

313: $\epsilon(\omega)$ and the dipole moment, the causes of the

314: deflections differ. In contrast with the bulk polarizability, where

315: the deflection magnitude is proportional to the field, the transition

316: induced deflections are intrinsic and correspond to distinct local

317: structural states. (To illustrate this distinction further, we point

318: out that a transition can be induced by things other than an AC (!)

319: electromagnetic field - a thermal phonon for instance.) Above $T_g$,

320: the distinct structural states, that evolve into TLS at cryogenic

321: temperatures, are transient metastable structures that live typically

322: as long as the $\alpha$-relaxation time \cite{XW}. The transient

323: structures are (transiently) frozen-in elastic fluctuations.

324: Analogously, the intrinsic generated dipole moment may be thought of

325: as due to frozen-in electric fields - at $T_g$. (A formal connection

326: between generated electric field and mechanical stress is discussed in

327: the following Subsection.) We can use this line of thought to relate

328: the dipole moment to the bulk dielectric properties of the material

329: near $T_g$. Suppose the frozen-in field is along $z$ direction. The

330: corresponding dipole moment is related to this field via an

331: appropriate (frequency dependent) dielectric constant

332: $\epsilon_\sTLS(\omega)$, but also through the fluctuation-dissipation

333: theorem. A self-consistent closure of this relationship gives for a

334: spherical region of {\em volume} $\xi^3$ \cite{TitulaerDeutch}:

335: \begin{widetext}

336: \begin{equation}

337: - \frac{1}{k_B T_g} \int_0^\infty e^{- i \omega t} \frac{d}{dt} \la

338:  \mu_{\sT,z}(0) \mu_{\sT,z}(t) \ra \, dt =

339:  \frac{[\epsilon_\sTLS(\omega)-1][2 \epsilon_\sext(\omega)+1]}{4\pi[2

340:  \epsilon_\sext(\omega) + \epsilon_\sTLS(\omega)]} \xi^3,

341: \end{equation}

342: \end{widetext}

343: where $\epsilon_\sext(\omega)$ is the dielectric susceptibility

344: outside the domain. The dielectric constant inside,

345: $\epsilon_\sTLS(\omega)$, contains both the usual (high frequency)

346: polarizability of the material and the polarization due to the dipolar

347: displacements accompanying the transition. Even though the interior

348: and exterior of the domain are chemically identical, it is necessary

349: to regard the two $\epsilon$'s as distinct, since we know a {\em

350: transition} occurs within the volume $\xi^3$. A similar relation can

351: be written for the volume occupied by a single elemental dipole, with

352: $\epsilon_\sext(\omega) = \epsilon_\sTLS(\omega)$:

353: \begin{widetext}

354: \begin{equation}

355: - \frac{1}{k_B T_g} \left(\frac{d_L}{a}\right)^2 \int_0^\infty e^{- i

356:  \omega t} \frac{d}{dt} \la \mu_{i,z}(0) \mu_{i,z}(t) \ra \, dt =

357:  \frac{[\epsilon_\sTLS(\omega)-1][2 \epsilon_\sTLS(\omega)+1]}{4\pi[3

358:  \epsilon_\sTLS(\omega)]} (2 a^3),

359: \end{equation}

360: \end{widetext}

361: where the factor $(d_L/a)^2$ on the left reflects that the elemental

362: change in polarization, $\Delta \mu_i$, is related to the full

363: elemental dipole $\mu_i$ by the rotation angle $(d_L/a)$, see

364: Fig.\ref{domain_dipole}. The volume $2 a^3$ on the right corresponds

365: to the volume occupied by a pair of beads, as before. One gets, as a

366: result, a frequency dependent generalization of Eq.(\ref{mu_Tqual}):

367: \begin{widetext}

368: \begin{equation}

369: \la \mu_\sT^{(2)}(\omega) \ra = \la \mu_i^{(2)}(\omega) \ra

370: [(\xi/a)^3/2] \left(\frac{d_L}{a}\right)^2 \frac{[2

371: \epsilon_\sext(\omega)+1] 3 \epsilon_\sTLS(\omega)}{[2

372: \epsilon_\sTLS(\omega)+1][2 \epsilon_\sext(\omega) +

373: \epsilon_\sTLS(\omega)]},

374: \label{freq_dep}

375: \end{equation}

376: \end{widetext}

377: where the two-point correlation functions are the $t$ integrals above.

378: Note $\la \mu_\sT^{(2)}(0) \ra = \la \mu_\sT^2 \ra$ and $\la

379: \mu_i^{(2)}(0) \ra = \la \mu_i^2 \ra$. Finally, $\mu_i \simeq \zeta (q

380: a) \equiv \mu_\smol$, as before.

381: 

382: Note that two adjacent regions are statistically unlikely to undergo a

383: structural transition at the same time. The physically preferable

384: choice for the external dielectric susceptibility

385: $\epsilon_\sext(\omega)$ is therefore the high frequency, electronic

386: component of the full dielectric response, which we call

387: $\epsilon_\infty$. With this, Eq.(\ref{freq_dep}) becomes

388: \begin{widetext}

389: \begin{equation}

390: \la \mu_\sT^{(2)}(\omega) \ra = \la \mu_i^{(2)}(\omega) \ra

391: [(\xi/a)^3/2] \left(\frac{d_L}{a}\right)^2 \frac{[2 \epsilon_\infty+1]

392: 3 \epsilon_\sTLS(\omega)}{[2 \epsilon_\sTLS(\omega)+1][2

393: \epsilon_\infty + \epsilon_\sTLS(\omega)]},

394: \label{freq_inf}

395: \end{equation}

396: \end{widetext}

397: Furthermore, the two-level systems that are active at low temperatures

398: correspond to the low barrier side of the barrier distribution. This

399: implies one should use the $\epsilon_\sTLS(\omega \rightarrow \infty)$

400: value with regard to the cryogenic phenomena and, therefore, no extra

401: frequency dependence appears in the coupling of the TLS to electric

402: field.  As a result, we find no significant reaction field correction

403: to our earlier argument and obtain

404: \begin{equation}

405: \la \mu_T^2 \ra \simeq \mu_\smol^2 [(\xi/a)^3/2] (d_L/a)^2,

406: \end{equation} 

407: c.f. Eq.(\ref{mu_Tqual}).

408: 

409: \subsection{Piezoelectric View}

410: 

411: To treat covalent network glasses, where the assignment of local

412: dipoles or changes is difficult, we now turn to a different way to

413: relate the dielectric response, within a domain, to frozen mechanical

414: stress. First consider a strictly periodic lattice that lacks parity

415: symmetry. Generally, the lattice strain induces internal electric

416: fields giving rise to piezoelectric behavior. In such a piezoeletric,

417: the energetics of the strain, in the lowest order, are described by

418: the free energy density (see e.g. \cite{LLcont})

419: \begin{equation}

420: \wF = \frac{1}{2} \lambda_{ik,lm} u_{ik} u_{lm} - \frac{1}{8 \pi}

421: \epsilon_{ik} E_i E_k + \beta_{i,kl} E_i u_{kl}.

422: \label{quadr1}

423: \end{equation}

424: Here $u_{ik}$ is the standard strain tensor \cite{LLelast},

425: $\lambda_{ik,lm}$ and $\epsilon_{ik}$ are the stiffness tensor and the

426: dielectric tensor respectively, and $\beta_{i,kl}$ is {\em a}

427: piezoelectric tensor. (Note, various sign conventions and free

428: energies have been used in the literature.) The double index summation

429: convention is implied throughout, with the exception of letters $x$,

430: $y$, and $z$, which will be obvious in the context.  Finally, the

431: elastic constant $\lambda$ is related to the material's mass density

432: $\rho$ and the speed of sound $c_s$:

433: \begin{equation}

434: \lambda \sim \rho c_s^2.

435: \label{stiffness}

436: \end{equation}

437: 

438: In the absence of external field: $\bD = - 4 \pi (\prtl \wF/\prtl \bE)

439: = 0$, - the internal electric field is simply proportional to the

440: strain itself:

441: \begin{equation}

442: E_i = 4 \pi \epsilon^{-1}_{ik} \beta_{k,lm} u_{lm}.

443: \label{E_spont}

444: \end{equation}

445: (Note, the $\beta$ tensor has the dimensions of electric field.)  If

446: expressed in terms of strain only, the free energy density reads, in

447: the absence of external electric field:

448: \begin{eqnarray}

449: \wF &=& \frac{1}{2} \lambda'_{ij,kl} u_{ij} u_{kl}, \label{quadr2} \\

450: \lambda'_{ij,kl} &\equiv& \lambda_{ij,kl} + 4 \pi \epsilon^{-1}_{mn}

451: \beta_{m,ij} \beta_{n,kl}. \label{lambda_pr}

452: \end{eqnarray}

453: The total, apparent stiffness, $\lambda'$, can be decomposed thereby

454: into a purely ``covalent'' and a ``coulomb'', i.e. electrostatic

455: component. The latter contribution is ordinarily quite small, owing to

456: the smallness of the charges induced by lattice distortions. Consider

457: $\alpha$-quartz, for example. Here, only $\beta_{x,xx} = 5.2 \cdot

458: 10^4$ esu, and $\beta_{x,yz} = \beta_{x,zy} = (1/2) 1.2 \cdot 10^4$

459: esu, are non-zero \cite{Cady}. As a result, the ``coulomb''

460: contribution to the $(xx,xx)$ component of the apparent stiffness

461: tensor is only about one percent of the covalent counterpart:

462: $\lambda_{xx,xx} = 8.8 \cdot 10^{11}$ dyne/cm$^2$ and $(4

463: \pi/\epsilon_1^T) \beta_{x,xx}^2 = 7.4 \cdot 10^9$ dyne/cm$^2$.

464: (Here, we used $\epsilon^T_1 = 4.58$ \cite{Mason}.)  The relative size

465: of the $\lambda$ vs. $\beta$ magnitudes can be understood as follows:

466: $\lambda$ reflects the energy (density) of the elastic restoring

467: force.  It is essentially the second derivative of an individual

468: atomic potential. Interatomic bonding, be it coulombic or covalent in

469: character, is ultimately of electrostatic origin. One may therefore

470: associate $\lambda$ with the quantity $\frac{1}{a^3}

471: \frac{\prtl^2}{\prtl^2 r} (q^2/r)|_{r=a} = (q/a^2)^2$, where $q$ is

472: the effective charge giving rise to the bond, and the $1/a^3$ factor

473: in front provides for energy density. (The total {\em first}

474: derivative of the (full quantum) potential energy is zero, of course.)

475: By virtue of being an electric field, $\beta$ roughly corresponds to

476: the quantity $q'/a^2$, where $q'$ would be the partial charge

477: introduced in the previous subsecton. The ratio $q'/q$ corresponds,

478: within the present framework, to the earlier introduced quantity

479: $\zeta$.  It follows that in $\alpha$-quartz, the partial atomic

480: charge is indeed about one tenth elementary charge, since the quantity

481: \begin{equation}

482: \zeta^2 = (q'/q)^2 \simeq \beta^2/\lambda 

483: \label{zeta2}

484: \end{equation}

485: is approximately equal to 1/100 in silica, as we just saw.

486: 

487: Now suppose for a moment that a relation similar to Eq.(\ref{E_spont})

488: exists between the bead displacements within a domain and the internal

489: electric field changes generated during a transition. (We stress, in

490: an amorphous sample such generated field changes are zero, upon

491: spatial average, but here we refer to {\em local} fields at a

492: particular, generally non-centrosymmetric site.) Since $d \wF = - \bD

493: d \bE/4\pi + \ldots$, the free energy change in the presence of a

494: (small) external field $\bD_\sext$ during the transition is given by

495: $\int_V dV \, \Delta \wF = - \bD_\sext \int_V dV \, \Delta \bE/4\pi$,

496: in the lowest order in $\bD_\sext$. (Here we have volume-integrated

497: over the reconfiguring domain that correspons with the two-level

498: dynamics; $\bD_\sext$ obviously varies sufficiently slowly within the

499: domain for realistic frequencies of light, and can be taken out of the

500: integral.)  The relation of the field $\bD_\sext$ to the external

501: field proper depends, of course, on the experiment's geometry.  We

502: will use an electric field $\bE_\sext (\br) = \epsilon^{-1}

503: \bD_\sext(\br)$ where $\epsilon$ is the average {\em bulk} dielectric

504: susceptibility (which, of course, is uniform and isotropic in an

505: amorphous material). The coupling to this field $\bE_\sext$ is

506: consequently given by:

507: \begin{equation}

508: \bmu_T = \frac{\epsilon}{4 \pi} \int_V dV \, \Delta \bE(\br),

509: \label{muT_formal1}

510: \end{equation}

511: i.e. the generated internal field difference during the transition,

512: integrated over the domain.  Eq.(\ref{muT_formal1}), among other

513: things, demonstrates that one may indeed unambiguously assign a

514: collection of point-like dipoles to the bead set within a TC, namely

515: by virtue of the relation $\bE(\br) = 4 \pi c_1 \bmu(\br) n(\br)$,

516: where $n(\br)$ is the (coordinate-dependent) dipole concentration and

517: the constant $c_1 \sim 10^0$ should be chosen depending on a specific

518: way to incorporate the already mentioned cavity effects. In what

519: follows, we will outline the microscopic picture of interaction of a

520: transition with elastic strain, which will naturally lead us to the

521: formula above and the ability to estimate the electric moment via

522: material's piezoelectric properties.

523: 

524: Since individual displacements $\bd_i$ during tunneling transitions

525: are only one-tenth of the lattice spacing, one can indeed describe the

526: corresponding additional elastic energy variations, due to the

527: presence of a phonon, by a quadratic form of the type from

528: Eq.(\ref{quadr1}) or (\ref{quadr2}).  Further, here one computes {\em

529: relative} displacements, not the absolute atomic coordinates which are

530: generally difficult to calculate.  Define $\phi_{ik}$ as the strain

531: tensor due to a (long-wave) lattice distortion of a stable lattice. In

532: addition, define $d_{ik}$ as the ``strain'' tensor corresponding to

533: the set of the tunneling displacements $\{\bd_i\}$.  The full elastic

534: energy within the domain, given a particular domain boundary

535: configuration (call it $\Omega_b$) can be written as:

536: \begin{equation}

537: \wF = \frac{1}{2}\lambda'_{ij, kl}

538: [(\phi_{ij}+d_{ij})(\phi_{kl}+d_{kl}) - d_{ij} d_{kl}] + {\cal

539: H}(\{d_{ij}\}, \Omega_b),

540: \end{equation}

541: where the energy functional ${\cal H}(\{d_{ij}\}, \Omega_b)$ includes

542: all the non-linear, many-body interactions giving rise to the

543: existence of the many metastable structural minima within the

544: domain. The construction of a library of the states corresponding to

545: these minima was described in Ref.\cite{LW_aging}. The full

546: multiplicity of the local states reveals itself directly in

547: calorimetric measurements above the glass transition. Here, we are

548: only concerned with the two lowest energy states of a region of the

549: otherwise undisturbed lattice. The two correspond to the two lowest

550: minima of ${\cal H}(\{d_{ij}\}, \Omega_b)$. The size of the domain is,

551: as we have seen, chosen so that one is guaranteed to have at least one

552: alternative structural state of nearly the same energy, and was found

553: to be only slightly larger that the cooperative region size at $T_g$

554: \cite{LW}. Note, by construction, the boundary state $\Omega_b$ is

555: independent of the phonon field $\phi_{ik}$. The effect of an external

556: {\em mechanical} stress on the internal displacements within a local,

557: compact region is passed on through the boundary, and so the

558: interaction of the region with the stress can be expressed through a

559: displacement integral over the region surface \cite{LLelast}. The

560: cross $\lambda_{ij,kl} \phi_{ij} d_{kl}$ term in Eq.(\ref{quadr2})

561: gives the amount by which the energy of a tunneling transition is

562: modified by the presence of a phonon. This therefore gives the

563: TLS-phonon coupling. The latter coupling was estimated in this way in

564: Ref.\cite{LW,LW_RMP}. A direct computation of the

565: $\phi_{ij}$-field-induced change of the transition energy gives:

566: $\Delta E(\phi_{ij}) = \lambda'_{ij,kl} \phi_{ij} \int_V dV \,

567: d_{kl}$, where we have integrated over the domain volume and taken

568: advantage of the elastic strain being $\phi_{ij}$ nearly constant

569: throughout the domain ($\lambda'$ in the latter equation is the

570: domain-averaged value of the atomic force-constant.) The coefficient

571: at the $\phi_{ij}$ gives a (tensorial) coupling of a transition to

572: strain according to:

573: \begin{equation}

574: {\cal H}_{\sTLS, \sph} = g_{ij} \phi_{ij} \sigma_z,

575: \label{H_int}

576: \end{equation}

577: where $\sigma_z = \pm 1$ is the usual Pauli matrix, and

578: \begin{equation}

579: g_{ij} = \frac{1}{2} \lambda'_{ij, kl} \int_V dV \, d_{kl}.

580: \end{equation}

581: The volume integral above indeed reduces to a surface integral of the

582: tunneling displacements. Consider for example, the term $\int_V dV \,

583: d_{xy} = \int_S (d_x dx + d_y dy) dz$ etc. The coupling to the {\em

584: longitudinal} phonons has the most vivid form, since $d_{ii}$ is the

585: divergence of a vector field, i.e. $\bd$. One gets $g_{ii} \simeq

586: \lambda \int_S \bd \, d\bS$. The atomic displacements at $T_g$ are

587: typically near the Lindemann length: $d \simeq d_L$, - but can also be

588: expressed in terms of the elastic constants, since the amount of

589: elastic energy contained in a unit cell is determined by the

590: temperature itself: $a^3 \lambda (d/a)^2 \sim T_g$. (This is by virtue

591: of the fluctuation-dissipation theorem.)  Estimating the surface

592: integral \cite{LW, LW_RMP} introduces additional numerical factors and

593: gives, within a factor of two or so:

594: \begin{equation}

595: g \simeq \sqrt{T_g \rho c_s^2 a^3},

596: \label{g}

597: \end{equation}

598: where we have used Eq.(\ref{stiffness}). The result above is easy to

599: rationalize on general grounds: Any atomic motions in a dense liquid

600: (at $T_g$ and otherwise) are either a vibration or an anharmonic

601: motion that is part of a structural transition. The two excitations

602: must coexist and thus be marginally stable against each other, in

603: order for both to be present. Such a marginal stability criterion

604: gives $\la \sigma_z g_{ij} \phi_{ij} \ra \simeq \la \lambda_{ij, kl}

605: \phi_{ij} \phi_{kl} \ra$, as follows from optimizing Eq.(\ref{H_int})

606: together with the elastic energy $\frac{1}{2} \lambda_{ij,kl}

607: \phi_{ij} \phi_{kl}$ with respect to $\phi_{ij}$, multiplying by

608: $\phi_{ij}$ and thermally averaging (at $T=T_g$). Owing to $\la

609: \sigma_z \phi_{ij} \ra \simeq \la |\phi_{ij}| \ra$, Eq.(\ref{g})

610: follows.

611: 

612: In the same way that the elastic fluctuations interact with the atomic

613: displacements, polarization waves and external electric sources will

614: interact with the internal electric fields generated in the domain

615: during a transition. As we have seen, the mechanical response

616: characteristics of a transition arise in response to stress

617: fluctuations at $T_g$. Analogously we can say, the electric moments of

618: the two-level system arise in response to the electric field

619: fluctuations at $T_g$. In full analogy with the

620: fluctuation-dissipation theorem context, the TLS dipole moment will

621: interact with an external field source just as it did with the

622: internal electric fields at the moment of freezing. We may thus

623: compute the dipole moment by substituting the generated electric field

624: from Eq.(\ref{E_spont}) into Eq.(\ref{muT_formal1}), bearing in mind

625: that the local lattice and the corresponding tensors are no longer

626: subject to any particular symmetry. Still, since the lattice locally

627: resembles a crystalline lattice, one may choose coordinates, again

628: locally, in such a way that the the $\beta$ tensor is maximally close

629: to a crystalline one. (Clearly, the $m$, $n$ sum in

630: Eq.(\ref{lambda_pr}) is independent of the coordinate choice.)

631: Therefore the latter sum will give a comparable result to that for a

632: crystal.  Summing over the displacement tensor $d_{ij}$ (contracted

633: with the $\beta$ tensor) is quite analogous to the argument leading to

634: Eq.(\ref{mu_Tqual}). As a result one obtains the following qualitative

635: estimate:

636: \begin{equation}

637: \mu_T \simeq (\bar{\beta} a^3) [(\xi/a)^3/2]^{1/2} (d_L/a),

638: \label{mu_Tpiezo}

639: \end{equation} 

640: where $\bar{\beta}$ is the local value of the piezoelectric

641: coupling. (We remind the reader that glasses are on average isotropic

642: and thus can not have bulk piezoelectric properties. It is only in the

643: frozen state that parity is locally broken.) The simple relation

644: \begin{equation}

645: \zeta q a \simeq \bar{\beta} a^3

646: \label{connection}

647: \end{equation}

648: establishes the connection between the ``molecular dipole'' view of

649: the previous Subsection and the ``piezoelectric'' analysis in this

650: Subsection, c.f. Eq.(\ref{zeta2}). Using the quartz parameters above,

651: one obtains that $\mu_T$ is, again, of the order Debye. Note that

652: formula (\ref{mu_Tpiezo}) uses quantities that can be measured

653: independently for substances which have a crystalline counterpart.

654: The bead size $a$ can be determined from the fusion entropy

655: \cite{LW_soft}. The piezoelectric constants are measurable too.  The

656: two views - one based on molecular moments, the other on local

657: piezoelectricity - are somewhat distinct but are highly overlapping:

658: Since the local hybridization pattern on individual atoms is

659: intrinsically asymmetric in amorphous lattices, partial atomic

660: charges, however small, are always expected to be present in glasses,

661: giving rise to both local permanent dipoles {\em and} local

662: piezoelectricity. Mixing in a dipolar species would enhance both

663: effects.  According to Ref.\cite{Schickfus_OH}, the two-level

664: systems's dipole magnitude is very correlated - nearly proportional -

665: to the OH$^-$ ion concentration, in amorphous silica with OH$^-$

666: impurities. Yet extrapolation to small ion concentrations shows TLS in

667: silica exhibit an {\em intrinsic} dipole moment, as was later

668: confirmed by an electric dipole echo study \cite{Golding_dipoleecho}.

669:  

670: 

671: 

672: \subsection{Electrodynamics and Electroacoustics: Connection with

673: Experiment}

674: 

675: In order to discuss experiments of two-level systems in glasses,

676: involving external fields, let us first recapitulate a few aspects of

677: the traditional phenomenological description, along with the

678: microscopic explanation.  The Hamiltonian of an isolated two-level

679: system, as usually written in the low $T$ glass context, is

680: \begin{equation}

681: {\cal H}_\sTLS = \frac{1}{2} \epsilon \sigma_z + \frac{1}{2} \Delta

682: \sigma_x + g_{ij} \phi_{ij} \sigma_z,

683: \label{H_TLS}

684: \end{equation}

685: where $\epsilon$ is the transition energy, $\Delta$ is the tunneling

686: matrix element. (The phonon part of the full Hamiltonian is given in

687: Eq.(\ref{quadr2}).) According to Refs.\cite{LW, LW_BP}, the TLS that

688: are thermally active at cryogenic temperatures have their splitting

689: $\epsilon$ distributed according to a simple Boltzmann-like law

690: $n(\epsilon) = \frac{1}{T_g} e^{-\epsilon/T_g}$. This roughly defines

691: the density of states of the tunneling transitions. The density of

692: states, as seen by calorimetry, is time dependent, because the

693: tunneling matrix elements $\Delta$ are widely distributed. According

694: to the semi-classical analysis in Refs.\cite{LW, LW_BP}, the

695: distribution is $P(\Delta) \propto 1/\Delta^{1+c}$, where $c \ll 1$ is

696: a small constant ($c \propto \hbar \omega_D/k_B T_g$). This

697: distribution is close to, but not precisely the same as the inverse

698: distribution $\propto 1/\Delta$ postulated by the phenomenological TLS

699: model. Including quantum effects reveals that the low splitting

700: two-level systems, i.e. those with $\epsilon \ll \Delta$, are special

701: in the following sense. In such regions, the excess strain energy of

702: the glass is concentrated in the domain wall itself, while the barrier

703: separating the two alternative structural states is not high enough to

704: keep the domain in any of the classical structural states as defined

705: in terms of the classical atomic coordinates. Such TLS, with depinned

706: domain walls, give rise to an extra piece in the combined $P(\epsilon,

707: \Delta)$ distribution \cite{LW_RMP}. They correspond to the so called

708: ``fast'' two-level systems, introduced early on phenomenologically

709: \cite{BlackHalperin} in order to rationalize certain quantitative

710: shortcomings of the original TLS model. The quantum depinning of the

711: domain wall has been also called ``quantum mixing'' \cite{LW_RMP}.

712: Finally, we note the Hamiltonian in Eq.(\ref{H_TLS}) leads to rich

713: relaxational behavior, due to both interaction with phonons and the

714: phonon-mediated interaction with other TLS's. This has been discussed

715: in detail previously \cite{relax}, \cite{HunklingerRaychaudhuri},

716: \cite{Phillips}.

717: 

718: The effects of the interaction of the dipole moment with a {\em

719: static} electric field are actually quite difficult to observe under

720: routine laboratory conditions (see e.g. \cite{Stephens_dipole}). The

721: upper limit for the field is given by the dielectric breakdown value

722: and is generically $10^6$ V/cm. For the typical dipole moment of 0.5

723: Debye, this implies an interaction energy of only 10$^{-3}$ eV $\sim$

724: 10 Kelvin. The constant field strength normally employed is actually

725: an order of magnitude weaker, or less. In a field $\bE_\sext$, the

726: transition energy is modified according to $\epsilon \rightarrow

727: (\epsilon - \bmu_\sT \bE)$. Typically, $|\bmu_\sT \bE| < 1$K. This is

728: clearly inferior to the characteristic energy scale of the TLS

729: spectrum, namely the glass transition temperature $T_g$ \cite{LW}. The

730: effect of a constant field thus turns out to be very generic because

731: of the intrinsic flatness of the energy distribution: The angular part

732: of $\bmu_\sT \bE_\sext$ is, obviously, uniformly distributed

733: resulting, again, in a flat distribution of the field-modified

734: transition energy. In order to discern such a small energy variation,

735: a {\em resonance} technique must be employed. Just such an experiment

736: was performed by Maier at el. \cite{Maier}, who took advantage of the

737: possibility to burn very narrow holes - only a few MHz - in the

738: chromophore's inhomogeneous spectrum. These authors turn on the field

739: immediately after burning the hole and observe the (time-dependent)

740: hole broadening, whose overall magnitude depends quadratically on the

741: field strength. Maier at el. report the value of $\mu_T = 0.4$ D for a

742: PMMA matrix.

743: 

744: It follows from our theory that light and sound couple to the

745: two-level system transitions in a very similar way, save the dipole

746: character of the TC-photon interaction distinct from the tensorial

747: coupling of the transitions to the phonons.  Indeed, the temperature

748: dependence of the speed of light in vitreous silica, as obtained early

749: on in by Schickfus at el. \cite{Schickfus}, nearly coincides with the

750: corresponding ultrasonic data \cite{HunklingerRaychaudhuri}. This

751: strongly suggested, at the time, that both the electromagnetic and

752: acoustic anomalies had the same origin. That the coincidence is not

753: purely circumstantial was shown soon afterwords in a number of elegant

754: electro-acoustic experiments: Increasing the AC electromagnetic field

755: leads to saturation of the structural transitions and a decrease in

756: ultrasonic attenuation \cite{Laermans, Doussineau}. In addition,

757: exposure to the AC field affects the {\em acoustic} impedance of a

758: glass \cite{Doussineau}. Again, the sufficient sensitivity of these

759: experiments is due to the interaction with the AC field being

760: resonant. Finally we mention yet another venue in investigating the

761: TLS coupling to electric fields, namely the electric dipolar echo, see

762: e.g. \cite{Bernard_dipoleecho, Golding_dipoleecho}.

763: 

764: The dipole moment magnitudes, reported in all these experiments on the

765: respective substances, are all of the order 1 Debye, although more

766: recent measurements seem to be converging on a fraction of a

767: Debye. Unfortunately, the extracted dipole values do not completely

768: agree between different experiments. So, for example, Kharlamov at

769: el. \cite{Kharlamov_dipole} give relatively low values of 0.2 D and

770: 0.1 D for PMMA and PS respectively, based on their field induced

771: spectral diffusion data.  The degree of the quantitative discrepancy

772: is, of course, subject to the detailed assumptions on the distribution

773: of the individual TLS parameters, various angular averagings etc.

774:  

775: 

776: \section{Dipole-Dipole Interaction}

777: 

778: The idea of local structural transitions is internally consistent in

779: that the transitions are indeed distinct, weakly interacting entities.

780: This is easy to understand by considering the moment of vitrification,

781: when a particular pattern of mobile regions sets in: A transition will

782: be found locally, upon freezing, if at $T_g$ it was of marginal

783: stability with respect to external mechanical perturbation, as

784: delivered by stress waves to the given local region. It does not

785: matter, of course, whether the source of these waves is thermal

786: elastic fluctuations or the other structural transitions. Now, upon

787: having estimated the energy spectrum of the TLS and their coupling to

788: the phonons \cite{LW}, one may check the magnitude of the resultant

789: TLS-TLS interaction, mediated by the acoustic waves in the frozen

790: lattice. Such interaction self-consistently turns out to be small

791: \cite{LW_RMP}. We are aware of several observable consequence of the

792: interaction. For one thing, this interaction tends to quench the

793: spontaneous echo generation \cite{BlackHalperin, GG}, by virtue of

794: dephasing each TLS's motion when the TLS precesses about its local

795: field. Another, remarkable effect from the interaction is that it

796: gives rise to a negative thermal expansion coefficient in some

797: glasses, at low enough temperatures \cite{LW_RMP}: The fluctuating

798: entities in the lattice attract in the Van der Waals fashion, via

799: exchanging phonons. This attraction, counterbalanced by the materials

800: stiffness, acts to partially contract the sample. The number of

801: thermally excited transitions increases with the temperature and

802: thereby enhances the degree of contraction.  The effect is small,

803: about $10^{-6}$, but nevertheless observable. The often employed

804: dimensionless parameter characterizing lattice non-linearity - the

805: Gr\"{u}neisen parameter - is usually positive and of the order unity

806: in crystals. It was found to be large and negative in many glasses at

807: cryogenic temperatures, see e.g. \cite{Ackerman}.

808: 

809: The direct phonon mediated TLS-TLS interaction goes as $1/r^3$, just

810: like the usual dipole-dipole interaction, but is somewhat complicated

811: by the tensorial form of the coupling (see e.g. \cite{BlackHalperin}

812: for a discussion). If, however, the transverse and longitudinal speeds

813: of sound were equal, the interaction would be strictly dipole-dipole.

814: It is therefore often convenient to assume the elastic interaction is

815: indeed purely dipole-dipole resulting in a small {\em quantitative}

816: error.  With this simplification, a ``scalar'' phononic Hamiltonian

817: can be used: $\frac{1}{2} \lambda_{ij, kl} \phi_{ij} \phi_{kl}

818: \rightarrow \frac{1}{2} \rho c_s^2 (\nabla \phi)^2$, where $\phi$ is a

819: scalar displacement field polarized in a single direction. The

820: coupling will become $g_{ij} \phi_{ij} \sigma_z \rightarrow (\bg

821: \nabla \phi) \sigma_z$. (The tensorial character of the interaction

822: may actually be important in the Gr\"{u}neisen parameter context, see

823: \cite{LW_RMP} and below). The presence of an electric component to

824: each transition dipole moment clearly leads to another contribution to

825: the total interaction.  Since a detailed discussion of the interaction

826: effects has already been given elsewhere \cite{LW_RMP}, here we simply

827: estimate the strength of the electric dipole-dipole coupling relative

828: to the purely elastic counterpart and the rest follows in a

829: straightforward fashion.

830: 

831: The elastic dipole-dipole interaction is given by a simple formula,

832: see e.g. \cite{LW_RMP}, \cite{YuLeggett}:

833: \begin{equation}

834: {\cal H}_{\selast} \simeq \frac{g^2}{\rho c_s^2} \frac{1}{r^3}

835: \simeq T_g \left( \frac{a}{r} \right)^3,

836: \label{elast}

837: \end{equation}

838: where, for the sake of clarity, we eschew some numerical constants

839: (these could be found in Ref.\cite{LW_RMP}) and have used

840: Eq.(\ref{g}). The electric dipole-dipole interaction, by

841: Eqs.(\ref{zeta2}) and (\ref{connection}), is, on the other hand:

842: \begin{equation}

843: {\cal H}_{\select} \simeq \zeta^2 \rho c_s^2 a^3 \left( \frac{a}{r}

844: \right)^3.

845: \label{elect}

846: \end{equation}

847: What is the relative value of the two interactions? A useful rule of

848: thumb is that $g \simeq \sqrt{T_g \rho c_s^2 a^3}$ is of the order eV

849: for all substances.  In silica, for instance, $\rho c_s^2 a^3$ is

850: several eV, the Rydberg scale being a convenient (and physically

851: justified) landmark. The $T_g$ of silica is 1500 K, i.e. slightly

852: larger than 0.1 eV. We therefore make an interesting observation that

853: the electric dipole-dipole interaction {\em can} be comparable in

854: magnitude to the elastic counterpart for polar enough substances.

855: This is despite the relatively weak contribution (1\% or less) of the

856: coulombic forces to the apparent mechanical stiffness. We will

857: speculate on the physical significance of this observation in the

858: final Section of the article, while for now, we limit ourselves to a

859: formal notion: The elastic dipole-dipole interaction is disadvantaged,

860: compared to the electric counterpart, due to the large $\rho c_s^2$

861: term in the denominator of Eq.(\ref{elast}): Phonons are not true

862: gauge particles.

863: 

864: Leggett has emphasized \cite{Leggett} that the dimensionless

865: Gr\"{u}neisen parameter varies ``wildly'' between different amorphous

866: substances, in contrast with the nearly universal $l_\smfp/\lambda$

867: ratio. Lubchenko and Wolynes have argued \cite{LW_RMP}, this stems

868: from the Van der Waals attraction between the tunneling centers, which

869: is strongly enhanced by ``Boson Peak'' excitations. The total

870: attractive interaction consists of several contributions, is

871: temperature dependent, and is expressed in terms of various

872: combinations of the temperature, $T_g$ and the Debye

873: temperature. While it may be argued that there is an intrinsic {\em

874: upper bound} on the value of $\zeta$ (see below), there otherwise

875: seems to be little intrinsic connection between the polar and

876: structural characteristics of glasses, in general. As a result, the

877: electrostatic interaction from Eq.(\ref{elect}) is expected also to

878: contribute to the non-universality of the Gr\"{u}neisen parameter.

879: This notion is consistent with the sensitivity of the Gr\"{u}neisen

880: parameter in silica to the concentration of polar impurities

881: \cite{Ackerman}. Further, the magnitude of the negative thermal

882: expansivity is indeed larger in more polar mixtures according to

883: Ref.\cite{Ackerman}.

884: 

885: 

886: {\em Symmetry vs. Transient Piezoelectricity.} We have so far focused

887: on the dipole moment of one of the structural states {\em relative} to

888: the other, namely $\sum_i \Delta \bmu_i = \mu_\sT$, since it is what

889: determines the coupling of the transition to external field. We next

890: inquire whether there is a correlation between the degree of polarity

891: of a state with its absolute energy. For example, suppose for a moment

892: that the lower energy state is completely non-polar, so that the

893: two-level system dipole moment is completely due to the excited state.

894: Such a situation could be exploited experimentally: One could

895: supercool a liquid just enough so that it does not crystallize readily

896: (the way glassblowers do), then expose the sample to strong electric

897: field for a sufficient while, and then quench the sample below its

898: $T_g$. After that, remove the field.  Clearly, the number of dipole

899: moments along and opposite to the field direction will differ. In

900: other words, a number of dipoles will be actually lined up in a

901: preferred direction, leading to a (weak) ferro-electric order. As a

902: consequence of this, removal of the field in the procedure above

903: should lead to sample's contraction (which is the sample's way of

904: minimizing the ferroelectric energy.) Some residual polarization will

905: appear as well.  If the field is removed sufficiently fast, the sample

906: will also heat up some. (This, in a sense, represents an antithesis to

907: adiabatic demagnetization.)  Such polarization will not take place if

908: the degree of polarity is uncorrelated with the energy of a structural

909: state, at least within the relevant energy range.

910: 

911: While we can not, at present, rule out {\it \`{a} priori} the

912: ``pyroelectric'' scenario above, it seems rather unlikely for the

913: following reason. An amorphous sample, at least at long enough times,

914: should have inversion symmetry, on average. With this symmetry

915: present, no piezoelectricity, let alone pyroelectricity, can take

916: place \cite{LLcont}.  The question therefore is whether an external

917: field breaks the inversion symmetry (locally) without inducing

918: crystallization. Perhaps it will, on short enough time scales, before

919: considerable aging takes place. An experimental study would settle

920: this issue. At any rate, even if present, ferroelectric order would

921: not affect the thermal expansion properties at low temperatures

922: (c.f. the discussion of the Gr\"{u}neisen parameter in

923: \cite{LW_RMP}). This attraction mechanism is temperature independent

924: and simply contributes to the effective molecular field at each TLS

925: site.

926: 

927: 

928: \section{Closing Remarks}

929: 

930: In this article, we have outlined the microscopic origin of the

931: coupling of the intrinsic structural transitions in amorphous solids

932: to electric fields. The coupling stems from rotation of the molecular

933: bonds, within the region encompassing the transition, which generates

934: a net electric dipole. The molecular constituents of a glass, even if

935: intrinsically non-polar, are strained due to disorder. Therefore small

936: partial charges on each atom are expected leading to the presence of

937: electric dipole moments associated with individual bonds.

938: 

939: A local structural transition occurs by moving a domain wall through

940: the region. The domain wall is a mechanically strained region

941: separating alternative structural states. Such strained regions are

942: frozen-in thermal fluctuations of the lattice. (Above $T_g$, the

943: life-time of such as frozen-in structure would not exceed the typical

944: $\alpha$-relaxation time in the liquid.) In an analogous fashion, the

945: dipole moment can be thought of as frozen-in polarization

946: fluctuations, or as the local piezoelectric response to local strains.

947: 

948: The glass transition is driven by steric, i.e. mechanical

949: interactions, not primarily by electrical ones, which is reflected in

950: the smallness of the partial charge, given in the theory by the

951: dimensionless quantity $\zeta$.  So, for example, the elastic modulus

952: can be represented as $\rho c_s^2 (1 + \zeta^2)$, where $\zeta^2

953: \lesssim 0.01$ is the contribution of the ionic forces to the overall

954: material stiffness. (Here, we discriminate between ionic (or hydrogen)

955: bonds, as in NaCl or H$_2$O, and covalent bonds, as in diamond.) In

956: spite of its apparent small contribution to the material's structural

957: integrity, the coulomb component could actually be comparable to the

958: elastic component of the interaction between distinct {\em

959: transitions}. Note such interactions are unimportant as far as the

960: identity of each transition is concerned, because the corresponding

961: regions are statistically far apart \cite{LW_RMP}. Nevertheless, this

962: weak interaction may be viewed, loosely, as the successful attempt of

963: the system to have avoided entropically costly local ferroelastic

964: order. With this in mind, and the relatively strong coulomb

965: interaction of structural fluctuations, the smallness of the partial

966: charges in many glasses may simply reflect the fact that purely ionic

967: compounds, such as sodium chloride or water, do not vitrify readily

968: but instead, form low density lattices.  (Note, ice does amorphize

969: under high pressure, see e.g. \cite{Strassle}.)  Conversely, this

970: imposes an upper bound on the $\zeta^2$ value.

971: 

972: Consider now Eq.(\ref{lambda_pr}). In order for the lattice to be

973: stable, the $\lambda'$ matrix (with an $ij$ pair considered a single

974: index) should be positive definite. This means, if a particular

975: $\lambda$ is negative, the corresponding $\beta$ could be, in

976: principle, quite large. This would imply large induced dipole moments.

977: As just argued, the coulomb component is likely to be small in a

978: glass, because the latter is rather homogeneous. Defects in crystals,

979: however, can be more extended and can be highly anisotropic. In this

980: regard we wish to mention the single chromophore studies of spectral

981: drift in a Shpolskii matrix, by Bauer and Kador \cite{BauerKador}.

982: These authors have seen a transition (presumably due to a structural

983: defect) that generated a remarkably large dipole moment of 8

984: Debye. Note, the Shpolskii matrix is polycrystalline, not strictly

985: amorphous.

986: 

987: Finally we remark there is more to electrodynamics of amorphous solids

988: than what could realistically be discussed in this communication. Here

989: we have analysed only the case of good electric insulators. Distinct,

990: interesting phenomena take place in semiconductors and metallic

991: glasses. As an example, let us mention an old experiment of Claytor

992: and Sladek \cite{ClaytorSladek}, who found enhanced ultrasonic

993: attenuation in As$_2$S$_3$ glass upon removal of electric field. This

994: extra attenuation was greatly reduced by infrared radiation, which

995: suggested that there is, we quote, ``atomic relaxation accompanying

996: electronic transition in gap states where injected carriers have been

997: trapped''. We leave this for future work.

998: 

999: {\em Acknowledgments:} P.G.W. was supported by the NSF grant CHE

1000: 0317017. V.L. and R.J.S. gratefully acknowledge the NSF grant CHE

1001: 0306287 and the Donors of the Petroleum Research Fund of the American

1002: Chemical Society. We are happy to dedicate this bagatelle to Bob

1003: Harris, whose elegant contributions are many, but more than scan the

1004: range from the electrodynamical properties of matter, parity, and

1005: tunneling systems, all of which enter the question discussed here.

1006: 

1007: \bibliography{/home/vas/tex/lowT/lowT}

1008: %\bibliography{lowT}

1009: 

1010: \end{document}

1011: