Third edition

Lectures on Fr ohlich Polarons from 3D to 0D including detailed theoretical derivations

J. T. Devreese

arXiv:1012.4576v4 [cond-mat.other] 19 Nov 2013

Theory of Quantum and Complex Systems (TQC), Universiteit Antwerpen, CDE, Universiteitsplein, 1, B-2610 Antwerpen, Belgium
(Dated: November 20, 2013)

Abstract
Based on a course presented at the International School of Physics Enrico Fermi, CLXI Course,.Polarons in Bulk Materials and Systems with Reduced Dimensionality, Varenna, Italy, 21.6. - 1.7.2005.

In the present course, an overview is presented of the fundamentals of continuum-polaron physics, which provide the basis of the analysis of polaron eects in ionic crystals and polar semiconductors. These Lecture Notes deal with large, or continuum, polarons, as described by the Fr ohlich Hamiltonian. The emphasis is on the polaron optical absorption, with detailed mathematical derivations.

Electronic address: jozef.devreese@ua.ac.be

Contents

Single polaron
I. Introduction. The standard theories A. The polaron concept B. Intuitive concepts C. The Fr ohlich Hamiltonian D. Innite mass model [shiftoperators] E. The standard theories 1. Weak coupling via a perturbation theory 2. Weak coupling via a canonical transformation [shift-operators] 3. Strong coupling via a canonical transformation [shift-operators] 4. All-coupling theory. The Feynman path integral 5. On Monte Carlo calculations of the polaron free energy 6. On the contributions of the N -phonon states to the polaron ground state F. Polaron mobility II. Optical Absorption. Weak coupling A. Optical absorption at weak coupling [within the perturbation theory] B. Optical absorption at weak coupling [within the canonical-transformation method [71] (DHL)] III. Optical absorption. Strong coupling IV. Arbitrary coupling A. Impedance function of large polarons: An alternative derivation of FHIP [74] B. Calculation of the memory function (Devreese et. al. [48]) C. Discussion of optical absorption of polarons at arbitrary coupling 1. Sum rules for the optical conductivity spectra D. Scaling relations 1. Derivation of the scaling relations 35 42 44 44 66 69 73 77 77 6 6 7 11 12 13 13 14 16 25 27 29 32 34 34

2. Check of the scaling relation for the path integral Monte Carlo result for the polaron free energy Appendix 1. Weak coupling: LLP approach Appendix 2. Expansion in Stieltjes continuous fractions [52] 84 85 98

II

Many polarons
V. Optical conductivity of an interacting many-polaron gas A. Kubo formula for the optical conductivity of the many-polaron gas B. Force-force correlation function C. Canonical transformation D. Dynamic structure factor 1. Calculation of the dynamic structure factor using the retarded Greens functions 2. Plasmon-phonon contribution E. Comparison to the infrared spectrum of Nd2x Cex CuO2y F. Experimental data on the optical absorption in manganites: interpretation in terms of a many-polaron response 120 124 124 130 131 135 139 140 114 118 119 102 102 105 107 110

VI. Interacting polarons in a quantum dot A. The partition function and the free energy of a many-polaron system B. Model system 1. Analytical calculation of the model partition function C. Variational functional D. Two-point correlation functions 1. The correlation function g (q, | {N } , ) E. Many-polaron ground state in a quantum dot: extrapolation to the

homogeneous limit and comparison to the results for a polaron gas in bulk [88] 150 F. Optical conductivity 154 1. Selected results: the manifestations of the shell lling in optical conductivity 160 VII. Variational path-integral treatment of a translation invariant N -polaron

system A. The many-polaron system B. Variational principle C. Results VIII. Ripplonic polarons in multielectron bubbles A. Ripplon-phonon modes of a MEB B. Electron-ripplon interaction in the MEB C. Locally at approximation D. Ripplopolaron in a Wigner lattice: the mean-eld approach E. Ripplopolaron Wigner lattice at nite temperature F. Melting of the ripplopolaron Wigner lattice Acknowledgments A. Optical conductivity of a strong-coupling polaron (S. N. Klimin and J. T. Devreese to be published in Physical Review B, 2013. Figures are published in Ref. [123].) 1. Introduction 2. Optical conductivity 3. Results and discussion 4. Conclusions Appendix 1. Correlation function 5. Appendix 2. Eective phonon modes a. Exact averaging b. Averaging neglecting the Jahn-Teller eect B. Feynmans path-integral polaron treatment approached using time-ordered operator calculus [S. N. Klimin and J. T. Devreese, Solid State Communications 151, 144 (2011)] C. Notes on the polaron mobility D. Publications on the polarons in Physical Review Letters (since 2005)

163 163 164 166 169 169 171 172 172 174 176 180

181 181 182 190 193 194 196 198 200

205 215 217

References

223

Part I

Single polaron
I. INTRODUCTION. THE STANDARD THEORIES A. The polaron concept

A charge placed in a polarizable medium is screened. Dielectric theory describes the phenomenon by the induction of a polarization around the charge carrier. The induced polarization will follow the charge carrier when it is moving through the medium. The carrier together with the induced polarization is considered as one entity (see Fig. 1). It was called a polaron by L. D. Landau [1]. The physical properties of a polaron dier from those of a band-carrier. A polaron is characterized by its binding (or self-) energy E0 , an eective mass m and by its characteristic response to external electric and magnetic elds (e. g. dc mobility and optical absorption coecient).

FIG. 1: Artist view of a polaron. A conduction electron in an ionic crystal or a polar semiconductor repels the negative ions and attracts the positive ions. A self-induced potential arises, which acts back on the electron and modies its physical properties. (From [2].)

If the spatial extension of a polaron is large compared to the lattice parameter of the solid, the latter can be treated as a polarizable continuum. This is the case of a large (Fr ohlich) 6

polaron. When the self-induced polarization caused by an electron or hole becomes of the order of the lattice parameter, a small (Holstein) polaron can arise [134]. As distinct from large polarons, small polarons are governed by short-range interactions.

B.

Intuitive concepts

a. The polaron radius. Large polarons vs small polarons Consider the LO phonon eld with frequency LO interacting with an electron. Denote by the quadratic mean square deviation of the electron velocity. In the electron-phonon interaction is weak, the electron can travel a distance x LO (1.1)

1 during a time LO ,characteristic for the lattice period,because it is the distance within

which the electron can be localized using the phonon eld as measuring device. From the uncertainty relations it follows

px =

m ( )2 LO LO , m mLO . (1.2)

At weak coupling x is a measure of the polaron radius rp . To be consistent, the polaron radius rp must be considerably larger than the lattice parameter a.(this is a criterion of a large polaron). Experimental evaluation of the polaron radius leads to the follwing typical for alkali halides, rp 20A for silver halides, rp 100A for II-VI, II-V values: rp 10A semiconductors. The continuum approximation is not satised for transition metal oxides (NiO, CaO, MnO), in other oxides (UO2 ,NbO2 ...). For those solids the small polaron concept is used. In some substances (e.g. perovskites) some intermediate region between large and small polarons is realized. b. The coupling constant [3] Consider the case of strong electron-phonon interaction in a polar crystal. The electron of mass m is then localized and can - to a rst approximation

- be considered as a static charge distribution within a sphere with radius l1 . The medium is characterized by an average dielectric constant ,which will be dened below. The potential energy of a sphere of radius l1 uniformly charged with the charge e in a vacuum is (see Eq. (8.6) of Ref. [4]) Uvac = 3 e2 . 5 l1 (1.3)

The potential energy of a uniformly charged sphere in a medium with the high-frequency dielectric constant is 3 e2 U1 = . 5 l1

(1.4)

This is the potential energy of the self-interaction of the charge e uniformly spread over the sphere of radius l1 in a medium with the dielectric constant . In a medium with an inertial polarization eld (due to LO phonons), the potential energy of the uniformly charged sphere is U2 = 3 e2 , 5 0 l1 (1.5)

where 0 is the static dielectric constant. The polaron eect is then related to the change of the potential energy of the interaction of the charged sphere due to the inertial polarization eld. This change is the potential energy U2 of the uniformly charged sphere in the presence of the inertial polarization eld minus the potential energy of the self-interaction U1 of the charge e uniformly spread over the sphere in a medium without the inertial polarization: Upol U2 U1 = with 1 1 1 . = 0 The electron distribution in a sphere may be non-uniform, what may inuence the numerical coecient in Eqs. (1.4) to (1.6). In this connection one can use the estimate [3] Upol e2 . l1 (1.7) 3 e2 5 l1 1 1 0 = 3 e2 , 5 l1 (1.6)

The restriction of the electron in space requires its de Broglie wave length to be of the
2 order l1 ,so that its kinetic energy is of the order 4 2 2 /2ml1 .Minimizing the total energy

with respect to l1 leads to

 l1

e2 4 2 2 + 2 l1 2ml1

= 0 =

1 e2 m = 2 2 , l1 4

wherefrom the binding energy is e4 m U1 = 2 2 2 . 8 radius according to (1.2), rp = 2 /mLO , the binding energy is U2 = We note that e4 m 1 U1 = 2 3 2 = 2 LO 8 LO 4 U2 LO
2

(1.8)

For weak coupling, one can neglect the kinetic energy of the electron. Taking the polaron

e2 e2 = rp

mLO . 2

(1.9)

(1.10)

Following the conventions of the eld theory, the self energy at weak coupling is written as U2 = LO . Therefore the so-called Fr ohlich polaron coupling constant is

e2 e2 c

m 2
3 LO

mc2 1 . 2 LO

(1.11)

For the average dielectric constant one shows that 1 1 1 = , 0 where and 0 are, respectively, the electronic and the static dielectric constant of the polar crystal. The dierence 1/ 1/0 arises because the ionic vibrations occur in the infrared spectrum and the electrons in the shells can follow the conduction electron adiabatically.

c. Polaron mobility Here we give a simple derivation leading to the gross features of the mobility behaviuor, especially its temperature dependence. The key idea is that the mobility will change because the number of phonons in the lattice, with which the polaron interacts, is changing with temperature. The phonon density is given by 1

. 1 e The mobility for large polaron is proportional to the inverse of the number of phonons:
LO kT

n=

and for low temperatures kT LO

1 =e n

LO kT

LO kT

(1.12)

The mobility of continuum poarons decreases with increasing temperature following an exponential law. The slope of the straight line in ln vs 1/T is characterized by the LO phonon frequesncy. Systematic study performed, in particular, by Fr ohlich and Kadano, gives e e 2mLO
LO kT

(1.13)

The small polaron will jump from ion to ion under the inuence of optical phonons. The lerger the numver of phonons, the lerger the mobility. The behaviuor of the small polaron is the opposite of that of the large polaron. One expects: n= For low temperatures kT LO 1
LO kT

e one has: e

LO kT

(1.14)

the mobility of small polaron is thermally activated. Systematic analysis within the smallpolaron theory shows that

10

 e with 5.
C. The Fr ohlich Hamiltonian

LO kT

(1.15)

Fr ohlich proposed a model Hamiltonian for the large polaron through which its dynamics is treated quantum mechanically (Fr ohlich Hamiltonian). The polarization, carried by the longitudinal optical (LO) phonons, is represented by a set of quantum oscillators with frequency LO , the long-wavelength LO-phonon frequency, and the interaction between the charge and the polarization eld is linear in the eld [3]: H= p2 + 2mb LO a+ k ak +
k k ik r (Vk ak eikr + Vk a ), ke

(1.16)

where r is the position coordinate operator of the electron with band mass mb , p is its canonically conjugate momentum operator; a k and ak are the creation (and annihilation) operators for longitudinal optical phonons of wave vector k and energy LO . The Vk are Fourier components of the electron-phonon interaction LO Vk = i k 4 V
1 2 1 4

2mb LO

(1.17)

The strength of the electronphonon interaction is expressed by a dimensionless coupling constant , which is dened as: = e2 mb 2 LO 1 1 0 . (1.18)

In this denition, and 0 are, respectively, the electronic and the static dielectric constant of the polar crystal. In Table I the Fr ohlich coupling constant is given for a few solids1 . In deriving the form of Vk , expressions (1.17) and (1.18), it was assumed that (i) the spatial extension of the polaron is large compared to the lattice parameter of the solid (continuum
1

In some cases, due to lack of reliable experimental data to determine the electron band mass, the values of are not well established.

11

approximation), (ii) spin and relativistic eects can be neglected, (iii) the band-electron has parabolic dispersion, (iv) in line with the rst approximation it is also assumed that the LO-phonons of interest for the interaction, are the long-wavelength phonons with constant frequency LO . The model, represented by the Hamiltonian (1.16) (which up to now could not been solved exactly) has been the subject of extensive investigations, see, e. g., Refs. [2128]. In what follows the key approaches of the Fr ohlich-polaron theory are briey reviewed with indication of their relevance for the polaron problems in nanostructures.

D.

Innite mass model [shiftoperators]

Here some insight will be given in the type of transformation that might be useful to study the Fr ohlich Hamiltonian (1.16). For this purpose the Hamiltonian will be treated for a particle with innite mass mb , (which is at r = 0) : H =
k

LO a+ k ak +
k

(Vk ak + Vk a k ),

(1.19)

TABLE I: Electron-phonon coupling constants (After Ref. [2])

Material InSb InAs GaAs GaP CdTe ZnSe CdS

Ref.

Material AgCl KI TlBr KBr

Ref.

0.023 [5] 0.052 [5] 0.068 [5] 0.20 [5] 0.29 [6] 0.43 [5] 0.53 [5]

1.84 [11] 2.5 [5] 2.55 [5] 3.05 [5]

Bi12 SiO20 3.18 [19] CdF2 KCl CsI SrTiO3 RbCl 3.2 [5] 3.44 [5] 3.67 [5] 3.77 [20] 3.81 [5]

-Al2 O3 1.25 [7] AgBr -SiO2 1.53 [11] 1.59 [12]

12

which can be transformed into the following expression with shifted phonon operators: H =
k

LO a k +

Vk LO

ak +

Vk LO

|Vk |2 . LO

(1.20)

To determine the eigenstates of this Hamiltonian, one can perform a unitary transformation which produces the following shift of the phonon operators: ak bk = ak + The transformation S = exp is canonical: S = exp and has the desired property: S 1 ak S = ak The transformed Hamiltonian is now: S
1

Vk Vk , a . k bk = ak + LO LO a k
k

Vk + LO

Vk ak LO

(1.21)

ak
k

Vk + LO

Vk ak = S 1 LO

Vk Vk , S 1 a . k S = ak LO LO

H S=
k

LO a k ak

|Vk |2 . LO

The eigenstates of the Hamiltonian contain an integer number of phonons (|nk ) .The eigenenergies are evidently: E=
k

nk LO

|Vk |2 . LO

This expression is divergent at it is often the case in eld theory of point charges are considered. A transformation of the type S has been of great interest in developing weak coupling theory as shown below.

E. 1.

The standard theories Weak coupling via a perturbation theory

For actual crystals -values typically range from = 0.02 (InSb) up to 3 to 4 (alkali halides, some oxides), see Table 1. A weak-coupling theory of the polaron was developed 13

originally by Fr ohlich [3]. He derived the rst weak-coupling perturbation-theory results: E0 = LO and m = mb . 1 /6 (1.23) (1.22)

Expressions (1.22) and (1.23) are rigorous to order .

2.

Weak coupling via a canonical transformation [shift-operators]

Inspired by the work of Tomonaga on quantum electrodynamics (Q. E. D.), Lee, Low and Pines (LLP) [29] analyzed the properties of a weak-coupling polaron starting from a formulation based on canonical transformations (cp. the results of the subsection I D).As hown by them, the unitary transformation U = exp i P ka k ak
k

r ,

(1.24)

where P is a c-number representing the total system momentum allows to eliminate the electron co-ordinates from the system. Intuitively one might guess this transformation by writing the exact wave function in the form total H = exp i p r | .

It is plausible that the Bloch factor exp (i/ p r) attaches the system to the electron as origin of the co-cordinates. After this transformation the Hamiltonian (1.16) becomes: P
k

H = U 1 HU =

ka k ak

2mb

+
k

LO a k ak +
k

(Vk ak + Vk a k ).

(1.25)

If, for the sake of simplicity, the case of total momentum equal to zero is considered, this expression becomes:
2

H=

k, k

k k a k ak ak ak + 2mb

LO +
k

k 2mb

2 2

a k ak +
k

(Vk ak + Vk a k ).

(1.26)

The rst term of this Hamiltonian is the correlation energy term involving dierent values for k and k .If one diagonalizes the second and the trird term of the Hamiltonian (1.26) 14

(this can be done exactly by means of the shifted-oscillator canonical transformation S (1.21)), the result of LLP is found. The expectation value of the rst term is zero for the wave function S |0 . Therefore one is sure to obtain a variational result. It is remarkable that merely extracting the k = k term from the expression
2 k, k

k k a k ak ak ak 2mb

(1.27)

eliminates the divergency from the problem (cp. with the case mb ) and is equivalent to the sophisticated theory by Lee, Low and Pines (LLP), which corresponds thus to neglect of the term (1.27). The details of the LLP theory are given in Appendix 1. The explicit form for the energy is now E= | Vk | 2 2 k 2 = . LO + 2m b

This self energy is no longer divergent. The divergence is elmininated by the quantum cut-o occurring at k = 2mb LO / . For the self energy the LLP result is equivalent to the perturbation result. The eective mass however is now given by , 6 a result, which follows if one considers the case P =0 and which is also exact for 0. m = mb 1 + However, the LLP eective mass is dierent from the perturbation result if insreases. The LLP approximation has often been called intermediate-coupling approximation. However its range of validity is the same as that of perturbation theory to order . The signicance of the LLP approximation consists of the exibility of the canonical transformations together with the fact that it puts the Fr ohlich result on a variational basis. To order 2 , the analytical expressions for the coecients are 2 : 2 ln( 2 + 1) 3 ln 2 2 7 2 0.01591962 for the energy and 4 ln( 2 + 1) 2 ln 2 5 8 2 + 36 0.02362763 for the 2 3 3 polaron mass [30]. At present the following weak-coupling expansions are known: for the energy [31, 32] E0 = 0.01591962202 0.0008060700483 . . . , LO and for the polaron mass [30] m = 1 + + 0.023627632 + . . . mb 6 15 (1.29) (1.28)

3.

Strong coupling via a canonical transformation [shift-operators]

Historically, the strong coupling limit was studied before all other treatments (Landau, Pekar [21, 33]). Although it is only a formal case because the actual crystals seems to have values smaller than 5, it is very interesting because it contains some indication of the intermediate coupling too: approach the excitations from the strong coupling limit and extrapolate to intermediate coupling is interesting because it is expected that some specic strong coupling properties survive at intermediate coupling. In what follows, a treatment, equivalent to that of Pekar, but in second quantization and written with as much analogy to the LLP treatment as possible is given. We start from the Fr ohlich Hamiltonian (1.16). At strong coupling one makes the assumption (a Produkt-Ansatz) for the polaron wave-function | = | |f where | is the electron-component of the wave function ( | component of the wave function |f ( f |f (1.30) = 1).The eld-

= 1) parametrically depends on | . The

Produkt-Ansatz (1.30) or Born-Oppenheimer approximation implies that the electron adiabatically follows the motion of the atoms, while the eld cannot follow the instantaneous motion of the electron. Fr ohlich showed that the approximation (1.30) leads to results, which are only valid for suciently large , i. e. in the strong-coupling regime. A more systematic analysis of strong-coupling polarons based on canonical transformations applied to the Hamiltonian (1.16) was performed in Refs. [3436]. The expectation value for the energy is now: H = | with k = | eikr | . We wish to minimize H , but also f| LO a+ k ak +
k k (Vk ak k eikr + Vk a k k ) |f

p2 | + f | 2mb

LO a+ k ak +
k k

(Vk ak k eikr + Vk a k k ) |f

has to be minimized. This expression will be minimized if |f is the ground state wave function of the shifted oscullator-type Hamiltonian. As we can diagonalize this Hamiltonian 16

exactly: LO a+ k ak +
k k (Vk ak k eikr + Vk a k k )

(1.31) |Vk |2 |k |2 , LO (1.32)

=
k

LO

a k

Vk k + LO

V ak + k k LO

we can apply a canonical transformation similar to (1.21): S = exp

k

Vk k V ak k k a LO LO k

(1.33)

which has the property: S 1 ak S = ak The transformed Hamiltonian is now: Vk Vk k 1 k , S ak S = a . k LO LO

S 1
k

LO a+ k ak +
k

(Vk ak k eikr + Vk a k k ) S

=
k

LO a k ak

|Vk |2 |k |2 . LO

The phonon vacuum |0 provides a minimum: 0| S

1 k

LO a+ k ak

+
k

(Vk ak k e

ik r

Vk a k k )

S |0 =

|Vk |2 |k |2 . LO

Hence, the Hamiltonian (1.31) is minimized by the ground state wave function S |0 = exp It gives the ground state energy p2 E0 = | | 2mb |Vk |2 |k |2 , LO (1.35) V Vk k ak k k a LO LO k |0 . (1.34)

which is still a functional of | . The functionals k are dierent for diererent excitations.

17

a. Ground state of strong-coupling polarons For the ground state one considers a Gaussian wave function: |1s = C exp with a variational parameter 0 . 1s |1s = C 2 =C
2

mb 0 2 r 2

d3 r exp
3 mb 0

mb 0

r2

= C2
3/2

dx exp
2

mb 0

x2

=C

mb 0
3/4

=1C = mb 0 2 r . 2
mb 0 1/2

mb 0

3/2

|1s =

mb 0

exp

2 For the further use, we introduce a notation C1 =

. Such a wave function is

consistent with the localization of the electron, which we expect for large . The kinetic energy in (1.35) for this function is calculated using the representation of the operator p2 = 2 2 = 1s |
2 2 p2 |1s = C2 2mb 2mb 2 2 (2 x + x + x ):

mb 0 2 r 2 mb 0 2 2 2 r 2 x + x + x exp 2 2 mb 0 2 mb 0 2 = 3 C2 dx exp x 2 x x exp 2mb 2 2 mb 0 2 mb 0 2 dz exp dy exp y z d3 r exp

2 C1

mb 0 2 x 2mb 2 mb 0 mb 0 2 x x exp x 2 2 mb 0 2 2 mb 0 dx 1 C1 x exp =3 2mb 2 mb 0 2 2 mb 0 == 3 C1 dx 1 x exp mb 0 2 mb 0 2 mb 0 = 3 C1 mb mb 0 2 mb 0 4 = 3 dx exp =3 0 2 C 4 1 0 =3 mb 0 4 mb 0

mb 0

x2 x2

mb 0

3 = 0 . mb 0 4

18

The functional k1s = 1s | eikr |1s = C 2 = C2

2

d3 r exp r2 + i mb 0

mb 0

r 2 + ik r
2 2

d3 r exp mb 0

mb 0
2 2

k r

2 2 k k + 2 2 2 2 4mb 0 4mb 0 2

= C exp k1s = exp

k 2 2 4mb 0 .

d r exp

mb 0

r+i

2mb 0

(1.36)

k2 4mb 0

The second term in (1.35) is then V |Vk |2 |k1s |2 = LO (2 )3 LO 4 d3 k 2 k V

1 2 1 2

2mb LO
0

exp

k2 2mb 0

LO = .4 2 2 = 2 LO

2mb LO
1 2

dk exp 2mb 0

k2 2mb 0 0 LO . (1.37)

2mb LO

The variational energy (1.35) thus becomes E0 = Putting E0 = 0, 0 one obtains 3 = 4 2 LO = 0 0 2 4 2 0 = = = = LO 3 LO 9 LO 3 0 4 0 . LO (1.38)

4 2 LO . (1.39) 0 = 9 Substituting (1.39) in (1.38), we nd the ground state energy of the polaron E0 (calculated with the energy of the uncoupled electron-phonon system as zero energy): E0 = 3 4 2 LO 2 = LO 4 9 3 E0 = 1 2 3 3 2 LO =

1 2 LO = 0.1062 LO . (1.40) 3 The strong-coupling mass of the polaron, resulting again from the approximation (1.30), is given [37] as:
4 m 0 = 0.0200 mb .

(1.41)

19

More rigorous strong-coupling expansions for E0 and m have been presented in the literature [38]: E0 = 0.1085132 2.836, LO m 0 = 1 + 0.02270194. mb for > 10. b. The excited states of the polaron: SS, FC, RES In principle, excited states of the polaron exist at all coupling. In the general case, and for simplicity for P = 0, a continuum of states starts at LO above the ground state of the polaron. This continuum physically corresponds to the scattering of free phonons on the polaron. Those scattering.states (SS) were studied in [39] anf for the rst time more generally in [37] are not the only excitations of the polaron. There are also internal excitation states corresponding to the excitations of the electron in the potential it created itself. By analogy with the excited states of colour centers, the following terminology is used. (i) The states where the electron is excited in the potential belonging to the ground state conguration of the lattice are called Franck-Condon (FC) states (ii) Excitations of the electron in which the lattice polarization is adapted to the electronic conguration of the excited electron (which itself then adapts its wave function to the new potential, etc. . . . leading to a self-consistent nal state), are called relaxed excited state (RES) [21]. c. Calculation of the lowest FC state The formalism used until now is well adapted to treat the polaron excitations at strong coupling. The eld dependence of the wave function is (1.34). For the FC state the k are the same as for the ground state (1.36). Physically k tells us, to what electronic distribution the eld is adapted. The electronic part of the excited wave function is 2p-like: |2p = C2p z exp mb p 2 r 2 (1.44) (1.42) (1.43)

The strong-coupling ground state energy (1.40) is lower than the LLP ground state energy

20

with a parameter p , which is equal to 0 : 2p |2p = =

2 C2 p

dzz exp
2

mb p

dx exp mb p
3/2

mb p

2 C2 p

3/2

mb p

mb p

2 C2 p

2mb p

=1

2 C2 p =

mb p

3/2

2mb p
3/4 1/2

|2p = We introduce still a notation

mb 0

2mb p

z exp
3/2

mb 0 2 r . 2

2 C2

2 =

mb p

The FC state energy is, similarly to (1.35), EF C = 2p | p2 |2p 2mb |Vk |2 |k1s |2 . LO (1.45)

21

The kinetic energy term is 2p

2 p2 2p = C2 2mb 2mb 2p

1 2 1 2

mb p 2 r 2 mb p 2 2 2 r 2 z exp x + x + x 2 2 mb p 2 mb p 2 2 = C2 z 2 z dzz exp z z exp p 2mb 2 2 2 mb p 2 mb p 2 y z + p dz exp dy exp 4 2 mb p 2 2 = C2 z dzz exp 2mb 2 mb p 2 1 mb p 2 + p z exp z z 1 2 2 2 mb p 2 2 C2 z dzz exp = 2mb 2 mb p 2 mb p mb p 2 2mb p z+ 1 z z exp z 2 2 mb p 2 mb p 2 mb p 2 2 p = C2 z z exp z dz 3 2mb mb p p 2 3 3 1 p = C2 + p 3 / 2 5 / 2 2 2 2 mb p 4 mb p p 2 1 3 = + p C2 3 / 2 2 2 4 mb p 3/2 3 p 2 1 mb p = + p 3 / 2 2 2 4 mb p d3 rz exp = 3 1 5 p + p = p . 4 2 4 (1.46)

For the FC state, p = 0 . The second term in (1.45) is precisely (1.37), |Vk |2 |k1s |2 = LO 0 LO ,

and the FC energy (1.45) becomes EF C = LO 0 5 0 4 LO 2 2 5 4 2 = LO LO = 4 9 3 22

5 2 9 3

2 LO =

The energy of the lowest FC state is, within the Produkt-Ansatz [40]: EFC = 2 LO = 0.03542 LO . 9 (1.47)

The fact that this energy is positive, is presumably due to the choice of a harmonic potential. The real potential the electron sees is anharmonic, and a bound state may be expected. d. Calculation of RES The electronic part of the excited wave function is (1.44) with a variational parameter p ,which is determined below. The variational RES energy is, similarly to (1.35), ERES = 2p | p2 |2p 2mb |Vk |2 |k2p |2 . LO (1.48)

Here the kinetic energy term is given by Eq. (1.46). The functional, which is now needed, is
2 k2p = 2p | eikr |2p = C2 p 2 = C2 p

d3 rz 2 exp r2 + i mb p

mb p

r 2 + ik r
2 2

d3 rz 2 exp mb p

mb p
2 2

k r

2 2 k k + 2 2 2 2 4mb p 4mb p 2

2 = C2 p exp

k 2 2 4mb p

d3 rz 2 exp

mb p

r+i

2mb p
2

k2p = exp

k2 4mb p

2 C2

dzz 2 exp dz z i
2

mb p

z+i
2

2mb p

kz z2 z2

k2 = . exp 4mb p k2 = exp 4mb p = exp = k2 4mb p

2 C2

2mb p

kz kz

exp
2 2

mb p

2 C2

dz z
3/2

2mb p

exp
2

mb p

2 C2

2 kz 1 2mb p

k2 exp 4mb p

mb p

2mb p

kz

mb p

1/2

(1.49)

23

Further, we substitute (1.49) in the second term in the r.h.s. of Eq. (1.48): |Vk |2 |k1s |2 V = LO (2 )3 1 LO 4 dk 2 k V
3 2
1 2

2mb LO k2 2mb p
1

2 kz 2mb p

exp
1 2

dx 2mb LO 1 2 4 4 k2 k x k 2 x2 exp + dk 1 2 mb p 4m2 2mb p 0 b p LO =

0
1 2

LO = 2 2

2mb LO 2 2k4 2 k2 + dk 2 2 3 mb p 5 4m2 b p

1 2

exp

k2 2mb p

LO =
2mb p

2mb LO 1 3 mb p 2
3/2 2mb p

1/2

2 1 2 5 4m2 b p

3 4
2mb p

5/2

= =

1 3 + 3 20 49 60 20 + 9 = LO p 60 60 LO p 1

LO p

The variational energy (1.48) becomes ERES = Putting ERES = 0, p one obtains 5 49 = 4 120 LO = p
2

49 5 p 4 60

LO p .

p 49 p = = = LO 150 LO 5 5 4 2

49 150

2 =
2

ERES

5 = 4 = 5 4

49 150 49 150

2 LO 49 49 = LO 60 150

49 150

2 LO

2 492 2 LO = LO = 120 150 24

The energy of the RES is (see Refs. [37, 39]): ERES = 0.0422 LO . The eective mass of the polaron in the RES is given [37] as: m RES 4 = 0.621 mb = 0.02004mb . 2 81 (1.51) (1.50)

The structure of the energy spectrum of the strong-coupling polaron is shown in Fig. 2.

FIG. 2: Structure of the energy spectrum of a polaron at strong coupling: E0 the ground state, ERES the (rst) relaxed excited state; the Franck-Condon states (EFC ). In fact, both the Franck-Condon states and the relaxed excited states lie in the continuum and, strictly speaking, are resonances.

The signicance of the strong-coupling large polaron theory is formal only: it allows to test all-coupling theories in the limit . Remarkably, the eective electron-phonon coupling strength signicantly increases in systems of low dimension and low dimensionality.

4.

All-coupling theory. The Feynman path integral

Feynman developed a superior all-coupling polaron theory using his path-integral formalism [41]. He studied rst the self-energy E0 and the eective mass m of polarons [41]. 25

Feynman got the idea to formulate the polaron problem into the Lagrangian form of quantum mechanics and then eliminate the eld oscillators, . . . in exact analogy to Q. E. D. . . . (resulting in) . . . a sum over all trajectories . . . . The resulting path integral (here limited to the ground-state properties) is of the form (Ref. [41]): 0, | 0, 0 = D r( ) exp 1 2

r 2 d +
0

22
3

0 0

e| | d d , | r ( ) r ( )|

(1.52)

where = 1/(kB T ). (1.52) gives the amplitude that an electron found at a point in space at time zero will appear at the same point at the (imaginary) time . This path integral (1.52) has a great intuitive appeal: it shows the polaron problem as an equivalent oneparticle problem in which the interaction, non-local in time or retarded, occurs between the electron and itself. Subsequently Feynman showed how the variational principle of quantum mechanics could be adapted to the path-integral formalism and he introduced a quadratic trial action (non-local in time) to simulate (1.52). Applying the variational principle for path integrals then results in an upper bound for the polaron self-energy at all , which at weak and strong coupling gives accurate expressions. Feynman obtained smooth interpolation between a weak and strong coupling (for the ground state energy). The weak-coupling expansions of Feynman for the ground-state energy and the eective mass of the polaron are: E0 = 0.01232 0.000643 . . . ( 0), LO m = 1 + + 0.0252 + . . . ( 0). mb 6 In the strong-coupling limit Feynman found for the ground-state energy energy: E0 E3D () = 0.1062 2.83 . . . ( ) LO LO and for the polaron mass: m m () 3D = 0.02024 + . . . ( ). mb mb (1.56) (1.55) (1.53) (1.54)

Over the years the Feynman model for the polaron has remained the most successful approach to this problem. The analysis of an exactly solvable (symmetrical) 1D-polaron model [39, 42], Monte Carlo schemes [28, 43] and other numerical schemes [44] demonstrate the remarkable accuracy of Feynmans path-integral approach to the polaron ground-state 26

energy. Experimentally more directly accessible properties of the polaron, such as its mobility and optical absorption, have been investigated subsequently. Within the path-integral approach, Feynman et al. studied later the mobility of polarons [45, 46]. Subsequently the path-integral approach to the polaron problem was generalized and developed to become a powerful tool to study optical absorption, magnetophonon resonance and cyclotron resonance [4751]. In Ref. [52], a self-consistent treatment for the polaron problem at all was presented, which is based on the Heisenberg equations of motion starting from a trial expression for the electron position. It was used to derive the eective mass and the optical properties of the polaron at arbitrary coupling. A variational justication of the approximation used in Ref. [52] (through a Stiltjes continuous fraction) is reproduced in Appendix 2.

5.

On Monte Carlo calculations of the polaron free energy

In Ref. [53], using a Monte Carlo calculation, the ground-state energy of a polaron was
0 with F the free energy per polaron derived as E0 = lim F, where F = F F 0 and F = [3/ (2 )] ln (2 ) the free energy per electron. The value LO = 25, used for

the actual computation in Ref. [53], corresponds to T /TD = 0.04 (TD = LO /kB ; LO is the LO phonon energy). So, as pointed out in Ref. [54], the authors of Ref. [53] actually calculated the free energy F , rather than the polaron ground-state energy. To investigate the importance of temperature eects on F, the authors of Ref. [54] considered the polaron energy as obtained by Osaka [55], who generalized the Feynman [41] polaron theory to nonzero temperatures: 3 F = ln
0v w sinh 2 w v sinh 0 2

3 v2 w2 4 v du eu D (u )

coth ,

0 v 2 2 0 v (1.57)

[1 + n (LO )] 2 where 0 = LO , n ( ) = 1/ e D (u ) =

0 0

1 , and + v2 w2 vu . 1 evu 4n (v ) sinh2 3 2v 2 (1.58)

u w2 u 1 2 v 2 0

This result is variational, with variational parameters v and w, and gives an upper bound to the exact polaron free energy. 27

The results of a numerical-variational calculation of Eq. (1.57) are shown in Fig. 3, where the free energy F is plotted (in units of LO ) as a function of for dierent values of the lattice temperature. As seen from Fig. 3, (i) F increases with increasing temperature and (ii) the eect of temperature on F increases with increasing .

FIG. 3: Contribution of the electron-phonon interaction to the free energy of the Feynman polaron as a function of the electron-phonon coupling constant for dierent values of the lattice temperature. Inset: temperature dependence of the free energy for = 3. (From Ref. [54].)

In Table II, the Monte Carlo results [53], (F )MC , are compared with the free energy of the Feynman polaron, (F )F , calculated in [54]. The values for the free energy obtained from the Feynman polaron model are lower than the MC results for 2 and 4 (but 2 and 4 the lie within the 1% error of the Monte Carlo results). Since the Feynman result for the polaron free energy is an upper bound to the exact result, we conclude that for

results of the Feynman model are closer to the exact result than the MC results of [53].

28

TABLE II: Comparison between the free energy of the Feynman polaron theory, (F )F , and the Monte Carlo results of Ref. [53], (F )MC , for T /TD = 0.04. The relative dierence is dened as = 100 [(F )F (F )MC ]/(F )MC . (From Ref. [54]) (F )F (F )MC (%) 0.5 0.50860 1.0 1.02429 1.5 1.54776 2.0 2.07979 2.5 2.62137 3.0 3.17365 3.5 3.73814 4.0 4.31670 0.505 1.020 1.545 2.080 2.627 3.184 3.747 4.314 0.71 0.42 0.18 0.010 0.21 0.32 0.24 0.063

6.

On the contributions of the N -phonon states to the polaron ground state

The analysis of an exactly solvable (symmetric) 1D-polaron model was performed in Refs. [39, 56, 57]. The model consists of an electron interacting with two oscillators possess also a k and ak ), commutes with the Hamiltonian of the system. Hence, the polaron states

ing the opposite wave vectors: k and -k.The parity operator, which changes ak and ak (and

are classied into the even and odd ones with the eigenvalues of the parity operator +1 and 1, respectively. For the lowest even and odd states, the phonon distribution functions WN are plotted in Fig. 1, upper panel, at some values of the eective coupling constant of the symmetric model. The value of the parameter = ( k )2 mb LO

for these graphs was taken 1, while the total polaron momentum P = 0. In the weakcoupling case ( 0.6) WN is a decaying function of N . When increasing , WN acquires a maximum, e.g. at N = 8 for the lowest even state at 5.1. The phonon distribution function WN has the same character for the lowest even and the lowest odd states at all values of the number of the virtual phonons in the ground state. (as distinct from the 29

higher states). This led to the conclusion that the lowest odd state is an internal excited state of the polaron. In Ref [28], the structure of the polaron cloud was investigated using the diagrammatic quantum Monte Carlo (DQMC) method. In particular, partial contributions of N -phonon states to the polaron ground state were found as a function of N for a few values of the coupling constant , see Fig. 1, lower panel. It was shown to gradually evolve from the weak-coupling case ( = 1) into the strong-coupling regime ( = 17). Comparion of the lower panel to the upper panel in Fig. 4 clearly shows that the evolution of the shape and the scale of the distribution of the N -phonon states with increasing as derived for a large polaron within DQMC method [28] is in remarkable agreement with the results obtained within the symmetric 1D polaron model [39, 56, 57].

30

FIG. 4: Upper panel : The phonon distribution functions WN in the symmetric polaron model for various values of the eective coupling constant at = 1, P = 0 (from [56], Fig. 23). Lower panel : Distribution of multiphonon states in the polaron cloud within DQMC method for various values of (from [28], Fig. 7).

31

F.

Polaron mobility

The mobility of large polarons was studied within various theoretical approaches(see Ref. [58] for the detailed references). Fr ohlich [59] pointed out the typical behavior of the largepolaron mobility exp( LO ), (1.59)

which is characteristic for weak coupling. Here, = 1/kB T , T is the temperature. Within the weak-coupling regime, the mobility of the polaron was then derived, e. g., using the Boltzmann equation in Refs. [60, 61] and starting from the LLP-transformation in Ref. [62]. A nonperturbative analysis was embodied in the Feynman polaron theory, where the mobility of the polaron using the path-integral formalism was derived by Feynman et al. (usually referred to as FHIP) as a static limit starting from a frequency-dependent impedance function. For suciently low temperature T the mobility then takes the form [45] = w v
3

3e e 2 4mb LO

LO

exp{(v 2 w 2 )/w 2v } ,

(1.60)

where v and w are (variational) functions of obtained from the Feynman polaron model. Using the Boltzmann equation for the Feynman polaron model, Kadano [63] found the mobility, which for low temperatures can be represented as follows: = w v
3

e e 2mb LO

LO

exp{(v 2 w 2 )/w 2 v } ,

(1.61)

The weak-coupling perturbation expansion of the low-temperature polaron mobility as found using the Greens function technique [64] has conrmed that the mobility derived from the Boltzmann equation is exceedingly exact for weak coupling (/6 1) and at low temperatures (kB T LO ). As shown in Ref. [63], the mobility of Eq. (1.60) diers by the factor of 3/(2 LO ) from that derived using the polaron Boltzmann equation as given by Eq. (1.61). In the limit of weak electron-phonon coupling and low temperature, the FHIP polaron mobility of Eq. (1.60) diers by the factor of 3/(2 LO ) from the previous result [6062], which, as pointed out in Ref. [45] and in later publications (see, e.g., Refs.[58, 63, 64]), is correct for 1. As follows from this comparison, the result of Ref. [45] is not valid when T 0. As argued in Ref. [45] and later conrmed, in particular, in Ref. [65] the above 32

discrepancy can be attributed to an interchange of two limits in calculating the impedance. In FHIP, for weak electron-phonon coupling, one takes lim0 lim0 , whereas the correct order is lim0 lim0 ( is the frequency of the applied electric eld). It turns out that for the correct result the mobility at low temperatures is predominantly limited by the absorption of phonons, while in the theory of FHIP it is the emission of phonons which gives the dominant contribution as T goes to zero [65]. The analysis based on the Boltzmann equation takes into account the phonon emission processes whenever the energy of the polaron is above the emission threshold. The independent-collision model, which underlies the Boltzmann-equation approach, however, fails in the strong coupling regime of the large polaron, when the thermal mean free path becomes less than the de Broglie wavelength; in this case, the Boltzmann equation cannot be expected to be adequate [45, 66]. For the polaron mobility in the weak- and strong-coupling regimes, see also Appendix C. In fact, the expression (1.60) for the polaron mobility was reported to adequately describe the experimental data in several polar materials (see, e.g., Refs. [6668]). Experimental work on alkali halides and silver halides indicates that the mobility obtained from Eq. (1.60) describes the experimental results quite accurately [67]. Measurements of mobility as a function of temperature for photoexcited electrons in cubic n-type Bi12 SiO20 are explained well in terms of large polarons within the Feynman approach [66]. The experimental ndings on electron transport in crystalline TiO2 (rutile phase) probed by THz time-domain spectroscopy are quantitatively interpreted within the Feynman model [68]. One of the reasons for the agreement between theory based on Eq. (1.60) and experiment is that in the path-integral approximation to the polaron mobility, a Maxwellian distribution for the electron velocities is assumed, when applying the adiabatic switching on of the Fr ohlich interaction. Although such a distribution is not inherent in the Fr ohlich interaction, its incorporation tends to favor agreement with experiment because other mechanisms (interaction with acoustic phonons etc.) cause a Gaussian distribution.

33

II.

OPTICAL ABSORPTION. WEAK COUPLING A. Optical absorption at weak coupling [within the perturbation theory]

At zero temperature and in the weak-coupling limit, the optical absorption is due to the elementary polaron scattering process, schematically shown in Fig. 5.

FIG. 5: Elementary polaron scattering process describing the absorption of an incoming photon and the generation of an outgoing phonon.

In the weak-coupling limit ( 1) the polaron absorption coecient was rst obtained by V. Gurevich, I. Lang and Yu. Firsov [69], who started from the Kubo formula. Their opticalabsorption coecient is equivalent to a particular case of the result of J. Tempere and J. T. Devreese (Ref. [51]), with the dynamic srtucture factor S (q, ) corresponding to the HartreeFock approximation (see also [70], p. 585). At zero temperature, the absorption coecient for absorption of light with frequency can be expressed in terms of elementary functions in two limiting cases: in the region of comparatively high polaron densities ( ( LO )/ 1) ( ) = e2 1 1 21/2 N 2/3 ( 1), 0 nc (3 2 )1/3 ( mb LO )1/2 3 1 2Ne2 ( 1)1/2 ( 1), 0 nc 3mb LO 3 (2.1)

and in the low-concentration region ( ( LO )/ 1) ( ) =

(2.2)

where = /LO , 0 is the dielectric permittivity of the vacuum, n is the refractive index of the medium, N is the concentration of polarons and is the Fermi level for the electrons. A step function 1 if > 1, ( 1) = 0 if < 1 34

reects the fact that at zero temperature the absorption of light accompanied by the emission of a phonon can occur only if the energy of the incident photon is larger than that of a phonon ( > 1). In the weak-coupling limit, according to Eqs. (2.1), (2.2), the absorption spectrum consists of a one-phonon line. At nonzero temperature, the absorption of a photon can be accompanied not only by emission, but also by absorption of one or more phonons. A simple derivation in Ref. [71] using a canonical transformation method gives the absorption coecient of free polarons, which coincides with the result (2.2) of Ref. [69].

B.

Optical absorption at weak coupling [within the canonical-transformation

method [71] (DHL)]

The optical absorption of large polarons as a function of the frequency of the incident light is calculated using the canonical-transformation formalism by Devreese, Huybrechts and Lemmens (DHL) Ref. [71]. A simple calculation, which is developed below in full detail, gives a result for the absorption coecient, which is exact to order . We start from the Hamiltonian of the electron-phonon system interacting with light is written down using the vector potential of an electromagnetic eld A (t): H= 1 e p + A (t) 2mb c
2

+
k

LO a+ k ak +
k

+ ik r Vk ak eikr + Vk ak e .

(2.3)

The electric eld is related to the vector potential as E (t) = 1 A (t) . c t (2.4)

Within the electric dipole interaction the electric eld with frequency is E (t) = E cos (t) c A = E sin (t) .

(2.5) (2.6)

When expanding

1 2mb

A (t) p+ e c

in the Hamiltonian, we nd
2

1 e p + A (t) 2mb c

e e2 p2 + A (t) p+ A2 (t) 2mb mb c 2mb c2

(2.7)

35

where the rst term is the kinetic energy of the electron, and the second term describes the interaction of the electron-phonon system with light Vt = e e A (t) p = E p sin t mb c mb Vt V sin t, e V E p. mb Since A (t) does not depend on the electron coordinates, the term
e2 A2 2mb c2

(2.8)

(2.9) (2.10) (t) in (2.7) does

not play a role in our description of the optical absorption. The total Hamiltonian for the system of a continuum polaron interacting with light is thus Htot = Hpol + Vt , where Hpol is Fr ohlichs Hamiltonian (1.16). The absorption coecient for absorption of light with frequency by free polarons is proportional to the probability P () that a photon is absorbed by these polarons in their ground state, p () = N P (). 0 cn2E 2 (2.11)

Here N is number of polarons, which are considered as independent from each other, 0 is the permittivity of vacuum, c is the velocity of light, n is the refractive index of the medium in which the polarons move, E is the modulus of the electric eld vector of the incident photon. If the incident light can be treated as a perturbation, the transition probability P () is given by the Golden Rule of Fermi: P () = 2 0 |V | f f |V | 0 (E0 + Ef ). (2.12)

V is the amplitude of the time-dependent perturbation given by (2.10). The ground state wave function of a free polaron is |0 and its energy is E0 . The wave functions of all possible nal states are |f and the corresponding energies are Ef . The possible nal states are all the excited states of the polaron. The main idea of the present calculation is to avoid the explicit summation over the nal polaron states, which are poorly known, by eliminating all the excited state wave functions |f from the expression (2.12). 36

With this aim, the representation of the -function is used: (x) = This leads to:
0

1 Re

dt exp [i (x + i) t] .

P () = 2 Re = 2 Re Using the fact that

dt 0 |V | f

f |V | 0 exp [i( + i + E0 Ef )t] f eiHt V eiHt 0 .

dt exp [i( + i)t] 0 |V | f |f f| = 1

and the notation eiHt V (0)eiHt = V (t), dV (t) = i [H, V (t)] dt we nd P () = 2 Re Dening R() = one has
0 0

dt exp [i( + i)t] 0 |V (0)V (t)| 0 .

(2.13)

dt exp [i( + i)t] 0 |V (0)V (t)| 0 , P () = 2 Re R().

(2.14)

(2.15)

Substituting (2.10) to (2.14), we nd that R () = and hence e P () = 2 mb

2 0

e mb

dtei(+i)t 0 |(E p (0)) (E p (t))| 0

(2.16)

Re

dtei(+i)t 0 |(E p (0)) (E p (t))| 0 .

(2.17)

It is convenient to apply the rst LLP transformation S1 (4.187), which eliminates the electron operators from the polaron Hamiltonian:
1 H H = S1 Hpol S1 = H0 + H1 :

H0 =

P2 + 2mb

LO +
k

k2 kP a k ak + 2mb mb

(Vk ak + Vk a k ),
k

1 H1 = 2mb

k k a k ak ak ak ,

37

where H0 can be diagonalized exactly and gives rise to the self-energy E = LO ,and H1 contains the correlation eects between the phonons. The optical absorption will be calculated here for the total momentum of the system P = 0. In the LLP approximation the explicit form of the matrix element in (2.16) is
1 1 1 0 |(E p (0)) (E p (t))| 0 = 0 S2 S1 E pS1 S1 E p(t)S1 S2 0 ,

(2.18)

where S1 and S2 are the rst (4.187) and the second (4.196) LLP transformations. The application of S1 gives:
1 1 iHt 1 iHt 1 1 iHt S1 p(t)S1 = S1 e peiHt S1 = S1 e S1 S1 pS1 S1 e S1 . 1 1 iHt 1 Using H = S1 HS1 , we arrive at S1 e S1 = eiHt .Further we recall S1 pS1 = P k

ka k ak + p, where P = 0 and p is set 0 (see Appendix 1). This results in

1 S1 p(t)S1 = eiHt peiHt = iHt keiHt a = k ak e

ka k (t)ak (t).
k

Then (2.18) takes the form 0 |(E p (0)) (E p (t))| 0 =

1 0 S2 k

E ka k ak

E ka k (t)ak (t) S2 0 . (2.19)

Here the second LLP transformation is given by (4.196) with fk = Vk 2 LO + 2k mb (2.20)

and the vacuum is dened by ak |0 = 0. The calculation of the matrix element (2.19) proceeds as follows:
1 0 S2 k

E ka k ak E ka k ak E ka k ak

E ka k (t)ak (t) S2 0 E ka k ak E ka k ak
1 iHt S2 S2 e S2 0

= =

1 0 S2 k 1 0 S2 k

1 iHt 1 S2 S2 e S2 S2 k

S2 eiS2

HS2 t

1 S2 k

S2 eiS2

HS2 t

0 .

(2.21)

Further on, we calculate

1 S2 H S2 = H 0 + H 1

38

where
1 1 H 0 = S2 H0 S2 = S2

LO +
k

k2 2mb

a k ak +
k

(Vk ak + Vk a k ) S2

1 Further we use S2 ak S2 = ak + fk :

H0 =
k

LO +

k2 2mb

a k ak +
k

LO +

k2 2mb

|fk |2

+
k

k2 LO + 2mb LO + k 2mb
2

a k fk + ak fk +

Vk (ak + fk ) + Vk a k + fk k 2 k2 2mb

=
k

a k ak +
k

| Vk | LO +

2 .

| Vk | 2
k

LO +

k2 2mb

=
k

k2 LO + 2mb

a k ak

| Vk | 2
k

LO +

k2 2mb

The last term can be calculated analytically: V |Vk |2 = 2 LO + 2k (2 )3 mb LO = 4 2 2 = 2LO

0

LO dk k
3 0

4 V

1 2mb LO
1 2

1 2

1 LO +

k2 2mb

dk

1 2mb LO

1 1+
k2 2mb LO

1 2LO = arctan | 0 = LO , 1 + 2 k2 2mb

H0 = LO + The term

LO +
k

a k ak .

1 H 1 = S2 H1 S2

will be neglected: eiS2

1

HS2 t

eiH0 t .

Neglecting H1 , consistent with the LLP description, introduces no error in order . Therefore (2.21) becomes 0
k iH0 t E k a k ak + fk ak + fk ak + fk fk e

iH0 t k E k ak ak + fk ak + fk ak + fk fk e

0 .

(2.22)

39

For P = 0 there is no privileged direction and 0

k E kfk ak eiH0 t

E kfk fk = 0, (2.22) reduces to:

iH0 t 0 . E kfk a ke

From the equation of motion for a k : da k2 k (t) = i + = i H0 , a LO k dt 2mb it is easy now to calculate
iH0 t eiH0 t a ke

a k,

a k

k2 exp i LO + 2mb

t .

The matrix element (2.19) now becomes 0 |(E p (0)) (E p (t))| 0 =

(E k)2 fk fk exp i LO +

k2 2mb

t + O ( 2 ).

The transition probability (2.13) is then given by the expression P () = 2 Re = 2 e2 2 m2 b

(E k)2 fk fk 0

dt exp i + i LO k2 2mb .

k2 2mb

e2 2 m2 b

(E k)2 |fk |2 LO

40

Using (2.20), we obtain 2e2 P () = 2 2 mb (E k)2

k 2

LO +
3

k2 2mb

|Vk |2 LO
2

k2 2mb
1 2

2e2 V = 2 2 mb (2 )3 (E k)2 LO +
k2 2mb

LO dk k

4 V

1 2mb LO

k2 2 LO 2mb
1

1 2 e2 E 2 1 = 2 2 2 dk dxx2 mb 2mb LO 0 1 k2 k2 1 1 2 2 LO LO 2mb LO 1 + 2mk b LO

8e2 E 2 = 3mb 2 = 4e2 E 2 3mb 2

2 1 2 2 2 LO (1 + ) 0 1 d 2 LO (1 + ) 0 d
LO LO

4e2 E 2 = 1 2 3mb LO where

1
2

2 4e2 E 2 LO 3mb 4

1 LO

1 , LO

1 LO

The absorption coecient (2.11) for absorption by free polarons for 0 nally takes the form
2 1 2Ne2 LO p () = 0 cn 3mb 3

1 if = 0 if

LO LO

>1 <1

1 LO

1 . LO

(2.23)

The behaviuor of p () (2.23) as a function of is as follows. For < LO there is no , p absorption. The threshold for absorption is at = LO .From = LO up to = 6 5 LO
6 LO the absorption coecient decreases slowly with increases to a maximum and for > 5

increasing . Experimentally, this one-phonon line has been observed for free polarons in the infrared absorption spectra of CdO-lms, see Fig. 6. In CdO, which is a weakly polar material with 0.74, the polaron absorption band is observed in the spectral region between 6 and 20 m (above the LO phonon frequency). The dierence between theory and experiment in the

41

wavelength region where polaron absorption dominates the spectrum is due to many-polaron eects.
7
-1

Experiment (Finkenrath et al., 1968) Gurevich, Lang and Firsov, 1962 Tempere and Devreese, 2001

Absorption coefficient (10

cm

10

20 (

30 m)

40

50

FIG. 6: Optical absorption spectrum of a CdO-lm with the carrier concentration n = 5.9 1019 cm3 at T = 300 K. The experimental data (solid dots) of Ref. [72] are compared to dierent theoretical results: with (solid curve) and without (dashed line) the one-polaron contribution of Ref. [69] and for many polarons (dash-dotted curve) of Ref. [51].

III.

OPTICAL ABSORPTION. STRONG COUPLING

The absorption of light by free large polarons was treated in Ref. [47] using the polaron states obtained wihtin the adiabatic strong-coupling approximation, which was considered above in subsection I E 3. It was argued in Ref. [47], that for suciently large ( > 3), the (rst) RES of a polaron is a relatively stable state, which can participate in optical absorption transitions. This idea was necessary to understand the polaron optical absorption spectrum in the strong-coupling regime. The following scenario of a transition, which leads to a zero-phonon peak in the absorption by a strong-coupling polaron, can then be suggested. If the frequency of the incoming photon is equal to RES ERES -E0 = 0.0652LO ,

42

then the electron jumps from the ground state (which, at large coupling, is well-characterized by s-symmetry for the electron) to an excited state (2p), while the lattice polarization in the nal state is adapted to the 2p electronic state of the polaron. In Ref. [47] considering the decay of the RES with emission of one real phonon it is demonstrated, that the zero-phonon peak can be described using the Wigner-Weisskopf formula valid when the linewidth of that peak is much smaller than LO . For photon energies larger than RES + LO , a transition of the polaron towards the rst scattering state, belonging to the RES, becomes possible. The nal state of the optical absorption process then consists of a polaron in its lowest RES plus a free phonon. A one-phonon sideband then appears in the polaron absorption spectrum. This process is called one-phonon sideband absorption. The one-, two-, ... K -, ... phonon sidebands of the zero-phonon peak give rise to a broad structure in the absorption spectrum. It turns that the rst moment of the phonon sidebands corresponds to the FC frequency FC : FC EFC E0 = 0.1412 LO .

To summarize, the polaron optical absorption spectrum at strong coupling is characterized by the following features (T = 0): a) An intense absorption peak (zero-phonon line) appears, which corresponds to a transition from the ground state to the rst RES at RES . b) For > RES + LO , a phonon sideband structure arises. This sideband structure peaks around FC . The qualitative behaviour predicted in Ref. [47], namely, an intense zero-phonon (RES) line with a broader sideband at the high-frequency side, was conrmed after an all-coupling expression for the polaron optical absorption coecient at = 5, 6, 7 had been studied [48]. In what precedes, the low-frequency end of the polaron absorption spectrum was discussed; at higher frequencies, transitions to higher RES and their scattering states can appear. The two-phonon sidebands in the optical absorption of free polarons in the strongcoupling limit were numerically studied in Ref. [73]. 43

The study of the optical absorption of polarons at large coupling is mainly of formal interest because all reported coupling constants of polar semiconductors and ionic crystals are smaller than 5 (see Table 1).

IV. A.

ARBITRARY COUPLING Impedance function of large polarons: An alternative derivation of FHIP [74]

a. Denitions We derive here the linear response of the Fr ohlich polaron, described by the Hamiltonian H= p2 + 2mb k a+ k ak +
k k ik r (Vk ak eikr + Vk a ), ke

(4.1)

to a spatially uniform, time-varying electric eld E (t) = E0 exp (it) ex . This eld induces a current in the x-direction j (t) = 1 E (t). Z () (4.3) (4.2)

The complex function Z () is called the impedance function. The frequency-dependent mobility is dened by () = Re 1 . Z () (4.4)

For nonzero frequencies (in the case of polarons the frequencies of interest are in the infrared) one denes the absorption coecient [48] () = 1 1 Re , n0 c Z () (4.5)

where 0 is the dielectric constant of the vacuum, n the refractive index of the crystal, and c the velocity of light. In the following the amplitude of the electric eld E0 is taken suciently small so that linear-response theory can be applied. The impedance function can be expressed via a frequency-dependent conductivity of a single polaron in a unit volume 1 = () Z () 44 (4.6)

using the standard Kubo formula (cf. Eq. (3.8.8) from Ref. [70]): () = i e2 1 + V mb V
0

eit [jx (t), jx ] dt.

(4.7)

In order to introduce a convenient representation of the impedance function, we give in the next subsection a denition and discuss properties of a scalar product of two operators [cf. [75], Chapter 5]. b. Denition and properties of the scalar product For two operators A and B (i.e., elements of the Hilbert space of operators) the scalar product is dened as

(A, B ) =
0

e L A B .
L

(4.8) acts on the operator

The notation e L A is used in order to indicate that the operator e

A . The time evolution of the operator A is determined by the Liouville operator L: i A 1 = LA [H, A] t (4.9)

with a commutator [H, A] , wherefrom A(t) = eiLt A(0) eiHt/ A(0)eiHt/ . The expectation value in (4.8) is taken over the Gibbs ensemble: A = Tr(0 (H )A) with the equilibtium density matrix when the electric eld is absent 0 (H ) = eH /Tr(eH ). (4.12) (4.11) (4.10)

One can show that (4.8) denes a positive denite scalar product with the following properties (i) (ii) (iii) and [cf. Eq. (5.11) of [75]] (iv) (A, B ) = (B, A) 45 (4.16) (A, B ) = (B , A ), (A, LB ) = (LA, B ), (A, LB ) = 1 A , B , (4.13) (4.14) (4.15)

Demonstration of the property (4.13). Starting from the denition (4.8) and using (4.10), we obtain

(A, B ) =
0

d eH A eH B .

(4.17)

Substituting here (4.11) with (4.12), one nds

(A, B ) =
0

dTr e( )H A eH B /Tr(eH ).

Change of the variable = allows us to represnt this integral as

(A, B ) =
0

d Tr e H A e( )H B /Tr(eH ).

Further, a cyclic permutation of the operators under the trace Tr sign gives

(A, B ) =
0

dTr e( )H BeH A /Tr(eH )

=
0

d eH BeH A =
0

e L B A

=
0

A .

According to the denition (4.8), this nalizes the demonstration of (4.13). Demonstration of the property (4.14). Starting from (4.17) and using (4.9), we obtain

(A, LB ) =
0

d eH A eH LB 1
0

d eH A eH (HB BH ) .

A cyclic permutation of the operators under the average sign gives (A, LB ) = 1
0

d eH A eH HB HeH A eH B .

(4.18)

Using the commutation of H and eH , one nds (A, LB ) = = = =

0

1
0

d eH A HeH B eH HA eH B

(4.19) (4.20) (4.21)

1
0

d eH A H HA eH B

1
0

d eH (HA AH ) eH B
H

d e

[H, A]

eH B

(4.22)

46

With the denition (4.9), this gives

(A, LB ) =
0

d e

(LA) e

B =.
0

(LA) B .

According to the denition (4.8), this nalizes the demonstration of (4.14). Demonstration of the property (4.15). Starting from (4.20) and performing a cyclic permutation of the operators under the average , we nd (A, LB ) = Further we notice that e consequently, (A, LB ) = = = = = 1 1 1
0 H

1
0

d eH A H HA eH B

A H HA e

d eH A eH = , d

d eH A eH B d d eH A eH B d
0

1
0

eH A eH B

A B eH A eH B /Tr(eH ) (4.23)

Tr eH A B eH A eH B

Tr eH A B A eH B /Tr(eH ).

Further, a cyclic permutation of the operators in the second term under the trace Tr sign gives (A, LB ) = = = 1 1 1 Tr eH A B eH BA /Tr(eH ) Tr eH A B BA A B BA = 1 /Tr(eH ) .

A , B

Thus, the property (4.15) has been demonstrated.

47

Demonstration of the property (4.16). We start from the represntation of the scalar product (4.17) and take a complex conjugate: (A, B ) =
0

d eH A eH B

=
0

d B eH AeH .

A cyclic permutation of the operators under the average sign gives then (A, B ) =
0

d eH B eH A =
0

d eH B eH A (4.24)

=
0

e L B A .

According to the denition (4.8), this nalizes the demonstration of (4.16). The above scalar product allows one to represent dierent dynamical quantities in a rather simple way. For example, let us consider a scalar product AB (z ) = =
0

A,

1 B zL e L A d e L A

(4.25) 1 B zL

= i = i = i = i = i = i = i

dtei(z L)t B

(4.26)

0 0 0 0 0 0 0

dteizt
0

d e L A eiLt B d eH A eH eiHt/ BeiHt/

0

dte

izt

dteizt
0

d eiHt/

+H

A eiHt/

B
)/

dteizt
0

d eiH (ti

)/

A eiH (ti

B (4.27) (4.28)

dte

izt 0

d eiL(ti ) A B d A (t i )B (0) .

dteizt
0

c. Representation of the impedance function in terms of the relaxation function The impedance function is related to the relaxation function (z ) x x (z ) = x, 1 x , zL (4.29)

where x is the velocity operator, by the following expression: 1 = ie2 lim ( + i) 0 Z () (z = + i, > 0). 48

(4.30)

Demonstration of the representation (4.29). function entering (4.30):

0

Apply (4.27) to the relaxation

(z ) = i
0

d
0

dteizt

ei(ti

)L

x x

and perform the integration by parts using the formula i e f (t) dt = f (t) eizt z i i = f (0) + z z
izt )L t=0 0

i f (t) izt e dt z 0 t f (t) izt e dt : t

0

eizt

ei(ti

x x dt =

i z i = z

i z 1 e L x x z e
L

x x +

eizt

ei(ti

)L

x x dt

eizt L ei(ti

)L

x x dt.

This allows us to represent the relaxation function in the form

(z ) = i 1 = z =
0

d
0

i z e
L

x x
0

1 z

eizt L ei(ti

)L

x x dt x x dt

i x x + z
0

d
0

eizt L ei(ti x x dt,

)L

1 i + mb z z

d
0

eizt L ei(ti

)L

where the expression (4.55) is inserted in the rst term. Further on, the integral over is taken as follows:

d L ei(ti
0

)L

x x = = = = = = 1 1 1 1

L
0

ei(ti
L

)L

dx x

eiLt e eiLt e e
L

1 x x

x x x

x (t) x (t) x

1 Tr eH eH x (t) eH x (t) x TreH 1 Tr x (t) eH eH x (t) x = TreH 1 = Tr eH (x x (t) x (t)) x TreH 1 1 = x x (t) x (t) x = x (t) , x . 49

eH x (t) eH x (t) x

Hence, (z ) = 1 i mb z z
0

dteizt [x (t) , x ] .

(4.31)

When setting z = + i with +0, we have ( + i) = For = 0, we can set ( + i) = 1 1 i mb

0

1 1 mb + i

i ( + i)

ei(+i)t [x (t) , x ] dt.

(4.32)

ei(+i)t [x (t) , x ] dt.

(4.33)

Multiplying ( + i) by ie2 , we nd that ie2 ( + i) = ie2

1 1 i ei(+i)t [x (t) , x ] dt mb 0 e2 e2 ei(+i)t [x (t) , x ] dt + =i mb 0 e2 1 =i ei(+i)t [jx (t) , jx ] dt, + mb 0

(4.34)

where the electric current density is jx = ex. Substituting further (4.34) in (4.30), we arrive at 1 1 e2 =i + lim 0 Z () mb
0

ei(+i)t [jx (t) , jx ] dt,

what coincides with the expression of the impedance function (4.6) through a frequencydependent conductivity given by the Kubo formula (4.7), q.e.d. d. Application of the projection operator technique Using the Mori-Zwanzig projection operator technique (cf. [75], Chapter 5), the relaxation function (4.29) (z ) = x, 1 x zL

can be represented in a form, which is especially convenient for the application in the theory of the optical absorption of polarons. The projection operator P (Q = 1 P ) is dened as PA = with A an operator and = (x, x ) . 50 (4.36) x (x, A) (4.35)

The projection operator Q = 1 P projects an operator onto the space orthogonal to the space containing x. Here we give some examples of the action of the projection operators: Px = x, Px = Qx = (1 P )x = 0; (4.37) (4.38) (4.39) a k, x = 0, (4.40) (4.41)

x (x, x) x (iLx, x) ix (x, Lx) ix = = = [x, x] = 0, x (x, ak ) x (iLx, ak ) ix (ak , Lx) ix = = =

Qx = (1 P )x = x. P ak =

Qak = (1 P )ak = ak The projection operators P and Q are idempotent: P A=

2 (x,A ) x x, x

x (x, A) = P A;

Q2 = (1 P )2 = 1 2P + P 2 = 1 P = Q. The Liouville operator can be identiaclly represented as L = LP + LQ. Then the operator
1 in z L

the relaxation function (4.29) can be represented as follows: 1 1 = . zL z LQ LP

We use the algebraic operator identity: 1 1 1 1 = y x+y x x x+y with x = z LQ and y = LP : 1 1 1 1 = + LP . zL z LQ z LQ zL Consequently, the relaxation function (4.29) takes the form (z ) = x, 1 1 1 x + x, LP x . z LQ z LQ zL 1 1 1 + 2 LQ + 3 LQLQ + ... x z z z 1 x, x z (4.42)

The rst term in the r.h.s. of (4.42) simplies as follows: x, 1 x z LQ = x, =

because Qx = 0. Using the quantity (4.36) we obtain: x, 1 x z LQ 51 = . z (4.43)

The second term in the r.h.s. of (4.42) contains the operator P 1 x zL 1 x zL x (z ) .

which according to the denition of the projection operator P (4.35) can be transformed as P 1 x x = zL x, = (4.44)

It is remarkable that this term is exactly expressed in terms of the sought relaxation function (4.29). Substituting (4.43) and (4.44) in (4.42), we nd

1 x (z ) + x, L z z LQ z x z (z ) = + x, (z ) L z LQ z LQ + LQ x (z ) L = + x, z LQ x LQ L (z ) = + x, 1+ z LQ (x, Lx ) 1 LQ z (z ) = + x, + Lx (z ) . z LQ (z ) = Introducring the quantity O= (x, Lx )

(4.45)

(4.46)

and the function called the memory function (z ) = 1 x, LQ 1 Lx , z LQ (4.47)

we represent (4.45) in the form of the equation [z O (z )] (z ) = . A solition of this equation gives the relaxation function (z ) represented within the MoriZwanzig projection operator technique: (z ) = . z O (z ) (4.48)

The memory function (4.47) can be still transformed to another useful form.First of all, we apply the property of a scalar product (4.14): 52

(z ) =

1 1 = 1 =

Lx, Q

1 Lx z LQ 1 1 1 P Lx, Q QLx, Q Lx + Lx . z LQ z LQ (P + Q)Lx, Q x (x, B) x (x, A) , B (x, A) (x, x ) (x, B) = (x, B) (x, A) = [(x, B ) (x, B )] = 0,

1 Lx z LQ

(4.49) (4.50) (4.51)

For any two operators A and B (P A, QB ) =

therefore the rst term on the r.h.s. in (4.49) vanishes, and we obtain (z ) = 1 QLx, Q 1 Lx . z LQ (4.52)

In this expression, the operator Q z 1 can be represented in the following form, using the LQ fact that Q is the idempotent operator: Q 1 1 1 1 =Q + 2 LQ + 3 LQLQ + ... z LQ z z z 1 1 1 = Q + 2 QLQ + 3 QLQLQ + ... z z z 1 1 1 = Q + 2 QLQ2 + 3 QLQ2 LQ2 + ... z z z 1 1 1 = + 2 QLQ + 3 QLQQLQ + ... Q z z z 1 1 = Q. Q z LQ z QLQ

(4.53)

A new Liouville operator can be dened, L = QLQ, which describes the time evolution in the Hilbert space of operators, which is orthogonal complement of x. Substituting then (4.53) with the operator L into (4.52), we bring it to the form, which will be used in what follows. 1 1 QLx, QLx . zL For the Hamiltonian (1.16) we obtain the following quantities: (z ) = = (x, x ) = px , iLx mb 53 = i (px , Lx) mb (4.54)

Using (4.15), we nd = and (x, Lx ) 1 = mb [x, x ] = 0. Substituting (4.55) and (4.56) in (4.48), one obtains O= (z ) = The operator Lx =L px 1 = [H, px ] = mb mb 1 mb px ,
k ik r ikx (Vk ak eikr Vk a ) ke ik r kx (Vk ak eikr Vk a ) ke ik r (Vk ak eikr + Vk a ) ke

i mb

[px , x] =

i mb

(i ) =

1 mb

(4.55)

(4.56)

1 1 . mb z (z )

(4.57)

i = mb 1 Lx = mb

(4.58)

does not depend on the velocities. Therefore, multiplying both parts of (4.58) with Q and taking into account (4.39) and (4.41), we obtain QLx = 1 mb
ik r kx (Vk ak eikr Vk a ), ke

what allows us to represent the memory function in the form

(z ) = =

1 1 mb 1 mb

QLx,

In transition to (4.59) we have used the property of the amplitude (1.17): Vk = Vk and taken into account that according to the denition (4.8), the rst operator enters a scalar product in the hermitian conjugate form. Introducing the operators
ik r bk = ak eikr ; b , k = ak e

1 QLx zL ik r ik r kx (Vk ak e Vk ak e ), 1 ik r ik r a e ) e V a k ( V k k k z L x k k ik r ik r (ak e + ak e ), . kx kx Vk Vk ik r 1 i k r ) + ak e (a e k z L k

(4.59)

54

we represent the memory function as (z ) = 1 mb

kx kx Vk Vk (bk + b k ), k k

1 (bk + b k ) . zL

(4.60)

We notice that Qbk = Q(ak eikr ) = ak eikr = bk . It will be represented through the four relaxation functions: (z ) =
+ + (z ) = kk (z ) = kk + (z ) = kk + (z ) = kk

1 mb b k,

+ + kx kx Vk Vk + (z ) + (z ) + + (z ) + (z ) , kk kk kk kk k k

(4.61) (4.62) (4.63) (4.64) (4.65)

1 b zL k 1 bk bk , zL 1 b bk k, zL 1 b bk , zL k

, , , .

There exist relations between the above relaxation functions. For example, the relaxation function (4.63), takes the form
(z ) = kk

bk ,

= i

1 bk zL dteizt eiLt bk (0), bk (0) .

Then the complex conjugate of this relaxation function:

(z ) kk

=i
0

dteiz dteiz

eiLt bk (0), bk (0)

=i
0

bk (0), eiLt bk (0) ,

where the property (4.16) has been used. The property (4.13) gives
(z ) kk

=i
0

dteiz dteiz dteiz dteiz

bk (0), eiHt/ bk (0)eiHt/

iHt/ eiHt/ b , b k (0)e k (0) eiLt b k (0), bk (0) ++ b k (t), bk (0) = kk (z ),

=i
0

=i
0

=i
0

55

wherefrom it follows that

+ (z ) = + (z ) . kk kk

(4.66)

Similarly, the relation

+ (z ) = + (z ) kk kk

(4.67)

is proven. e. Memory function In this subsection we indicate which approximations must be made in the calculation of the relaxation functions in order to obtain the FHIP results for the impedance function. Consider the relaxation function (4.62):
+ (z ) = + kk

b k,

= i

1 b zL k
dteizt eiLt b k (0), bk (0) = i 0

dteizt b k (t), bk (0) ,

iL t where b bk (0), and perform a partial integration: k (t) = e + + (z ) = kk

1 z

d eizt

b k (t), bk (0) 0 0

1 = eizt b k (t)bk (0) z 1 1 bk (0), b k (0) + z z 1 i = bk (0), b k (0) z z = Here we supposed that
t

1 z

dteizt

db k (t) , bk (0) dt (4.68)

dteizt iLb k (t), bk (0) dteizt Lb k (t), bk (0) .

lim eizt b k (t), bk (0) = 0.

In the second term in (4.68),

iL t iQLQt Lb bk , b bk , b k (t), bk (0) = Le k = QLQe k (0) iQHQt/ = QLQeiQHQt/ b , b ke k iQHQt/ = QLeiQHQt/ b , b ke k

56

 2 because Qb k = bk and Q = Q. Further on, we have Lb k (t), bk (0)

=
0

d e L eiQHQt/ bk eiQHQt/ LQb k d e L eiQHQt/ bk eiQHQt/ Lb k

=
0

iQHQt/ = LeiQHQt/ b , b ke k = Lb k (t), bk (0) .

So, we nd from (4.68)

+ + (z ) = kk

1 i bk (0), b k (0) z z i 1 bk (0), b = k (0) z z 1 i = bk (0), b k (0) z z

dteizt Lb k (t), bk (0) dteizt b k (t), Lbk (0)

dteizt

bk (t), b k (0)

(4.69)

The rst term in the r.h.s. of this expression can be represented as follows:
1 1 1 L bk (0), bk (0) = d e bk bk = d e L bk b k z z 0 z 0 1 1 1 e L 1 1 = eH bk eH b = bk b bk b k k z L z L L k 1 1 1 1 Tr eH eH bk eH b bk b = k H z Tre L L k 1 1 1 1 H = Tr b eH bk b k bk e H z Tre L L k 1 1 1 b bk b = k bk z L L k i 1 1 = b bk bk b k z iL iL k i iLt iLt dte bk bk bk dtb = k e z 0 0 i = dtbk (t)b dtb k k bk (t) z 0 0 i = . dt bk (t), b k (0) z 0

Substituting it in the r.h.s. of (4.69), we nd

+ (z ) + kk

i = z i = z

dt

bk (t), b k (0)

i z

dteizt .

bk (t), b k (0) (4.70)

dt 1 eizt

bk (t), b k (0) 57

In a similar way one obtains

+ (z ) = kk

i z

dt 1 eizt

[bk (t), bk (0)] .

(4.71)

Inserting the relaxation functions (4.70), (4.71), (4.66) and (4.67), we nd the memory function (4.61) 1 (z ) = mb =
kx kx Vk Vk k k

i z

dt 1 eizt

bk (t), b k (0) bk (t), b k (0)

+ [bk (t), bk (0)]

1 mb

[bk (t), bk (0)] 2 z

0

kx kx Vk Vk k k

dt 1 eizt ,

Im wherefrom

bk (t), b k (0)
0

+ [bk (t), bk (0)]

(z ) = with F (t) = 2 mb

1 z

dt 1 eizt Im F (t)

(4.72)

kx kx Vk Vk k k

bk (t), b k (0)

+ [bk (t), bk (0)]

(4.73)

f. Derivation of the memory function To calculate the expectation values in Eq. (4.73), we shall make the following approximations (cf. Ref. [76]). The Liouville operator L, which
iL t determines the time evolution of the operator b bk (0), is replaced by Lph + LF , k (t) = e

where Lph is the Liouville operator for free phonons and LF is the Liouville operator for the Feynman polaron model [41]. The Fr ohlich Hamiltonian appearing in the statistical average is imilarly replaced by Hph + HF ,with Hph the Hamiltonian of free phonons and HF the Hamiltonian of the Feynman polaron model. With this approximation, e.g., the average
bk (t)b k (0) = ak (t)ak (0)

eikr(t) eik r = k,k ak (t)a k (0)

eikr(t) eikr .

(4.74)

The time evolution of the free-phonon annihilation operator (4.10),

+ ak (t) = eiHph t/ ak eiHph t/ = exp ik a+ k ak t ak exp ik ak ak t

58

accorting to (4.9) is i dak (t) = k exp ik a+ k ak t dt = k exp ik a+ k ak t

+ a+ k ak , ak exp ik ak ak t + a+ k , ak ak exp ik ak ak t

+ = k exp ik a+ k ak t ak exp ik ak ak t

ak (t) = exp (ik t) ak . Similarly,

a k (t) = exp (ik t) ak .

Hence, we have
ak (t)a k = exp (ik t) ak ak = exp (ik t) 1 + ak ak = exp (ik t) [1 + n(k )] ,

where n(k ) = [exp( k ) 1]1 is the average number of phonons with energy k . The calculation of the Fourier component of the electron density-density correlation function eikr(t) eikr in Eq. (4.74) for an electron described by the Feynman polaron model is given below following the approach of Ref. [76]. We calculate the correlation function eikr(t) eikr( ) = with the Feynman trial Hamiltonian p2 1 p2 f + (r rf )2 . HF = + 2m 2mf 2 Here, r (t) denotes the operator in the Heisenberg representation r (t) = e
it

Tr eHF eikr(t) eikr( ) Tr (eHF )

(4.75)

(4.76)

HF

re

it

HF

(4.77)

We show that the correlation function eikr(t) eikr( ) depends on ( t) rather than on t and independently: eikr(t) eikr( ) = = Tr eHF e
it

HF ikr it HF

HF ikr i HF

Tr (eHF ) Tr eHF eikr e

i( t)

H F ik r

i( t)

HF

Tr (eHF ) (4.78)

= eikr eikr( t) = eikr eikr() , 59

where = t. The normal coordinates are the center-of-mass vector R and the vector of the relative motion : R=
mr+mf rf m+mf

The inverse transformation is:

The same transformation as (4.79) takes place for velocities: r + mf =R m+mf r m f = R

m+mf

r = R + mf m+mf rf = R m
m+mf

= r rf

(4.79)

(4.80)

(4.81)

From (4.79) we derive the transformation for moments mf p p P m = m+mf + m+mf mmf
pf mf

m+mf

P m+mf

m m+mf

p mmf m+mf

The Hamiltonian (4.76) then takes the form HF = with the masses

p = m P + p m+mf pf = mf P p
m+mf

(4.82)

p2 1 2 2 P2 + + 2M 2 2 mmf m + mf

(4.83)

M = m + mf , = and with the frequency

(4.84)

. (4.85) The Cartesian coordinates and moments corresponding to the relative motion can be in the = standard way expressed in terms of the second quantization operators:
1/2

j =

2 2

Cj + Cj , 1/2 . Cj Cj

p,j = i

(4.86)

(j = 1, 2, 3) 60

In these notations, the Hamiltonian (4.83) takes the form P2 HF = + 2M

3

j =1

C Cj + 1 . j 2

(4.87)

Using (4.87), we nd the operators in the Heisenberg representation, (i) for the center-of mass coordinates Xj e HF = ei 2M Pj Xj ei 2M Pj Pj2, Xj = Xj + i 2M = Xj + i Pj2 Xj Pj Xj Pj + Pj Xj Pj Xj Pj2 2M (Pj [Pj , Xj ] + [Pj , Xj ] Pj ) = Xj + i 2M Pj (2i ) = Xj + Pj , = Xj + i 2M M Xj ( ) = Xj + P j , M (ii) for the operators Cj and Cj Xj ( ) = e
HF i Cj ( ) = Cj ei , Cj ( ) = C j e .
i i 2 2

(4.88)

(4.89)

Using the rst formula of (4.80) we nd r ( ) = R ( ) + 2 m2 f 2M 2

1/2 mmf m+mf

mf ( ) m + mf
1/2

r ( ) = R + We denote mf a M =

mf P+ M M
1/2

Cei + C ei .

(4.90)

1/2

2 m2 f

2 (m + mf )2

mf 2mM

1/2

r ( ) = R + Therefore, we obtain

P + a Cei + C ei . M

(4.91)

eikr = exp (ik R) exp iak C + iak C

3

=
j =1

exp (ikj Xj ) exp iakj Cj iakj Cj

61

eikr() = exp ik R i

k P iak Cei + C ei M = exp ik R i k P exp iak Cei iak C ei M 3 i = exp ikj Xj i kj Pj exp iakj Cj ei iakj Cj e . M j =1

(4.92)

The disentangling of the exponents is performed using the formula

1

eA+B = eA T exp
0

deA BeA .

(4.93)

In the case when [A, B ] commutes with both A and B , this formula is reduced to eA+B = eA eB e 2 [A,B] . We perform the necessary commutations:
2 kj 1 2 1 ikj Xj , i kj Pj = kj [ Xj , P j ] = i , 2 M 2 M 2M
1

(4.94)

1 1 1 2 2 iakj Cj , iakj Cj = a2 kj Cj , Cj = a2 kj 2 2 2 1 1 1 i 2 2 e , iakj Cj ei = a2 kj Cj , Cj = a2 kj iakj Cj 2 2 2 eikr = eikR eiakC eiakC e 2 a eikr() = eikR ei M kP e
1 2 2 k

, eiakCe
i

iakC ei

ei 2M 2 a

k2

1 2 2 k

eikr eikr() = ei M kP eiakC eiakC eiakC

i e

eiakCe

ei 2M a

k2

2 k2

It follows from Eq. (4.94) that when [A, B ] commutes with both A and B , eA eB = eB eA e[A,B] . Using (4.95), we nd eiakj Cj eiakj Cj e
i

(4.95)

= eiakj Cj e

i eiakj Cj e[iakj Cj ,iakj Cj e ]

= eiakj Cj e 62

eiakj Cj ea

2 k 2 ei j

Herefrom, we nd
i i a2 k 2 (1ei ) ) ei 2k M eikr eikr() = ei M kP eiakC (1e ) eiakC(1e . 2

(4.96)

The correlation function then is eikr eikr() = ei M kP

i i a2 k 2 (1ei ) ) ei 2k M , eiakC (1e ) eiakC(1e 2

(4.97)

since the variables of the center-of mass motion and of the relative motion are averaged independently.
i i ) = eiQC eiQ C eiakC (1e ) eiakC(1e

Tr e = with Q = a 1 ei k. Let us consider the auxiliary expectation value e

iQC

C C iQC iQ C

Tr e

C C

iQ C

= =

Tr e 1 Tr Tr

C C iQC iQ C

Tr e
CC e

C C n=0 n=0

(iQ)n n!

(iQ )m Tr e m ! m=0
CC

CC

Cm

1
CC e

(1)n |Q|2n Tr e (n!)2

m=0 m=0

Cn .

Tr e

CC

= =

m e e
m

CC

C
n

Cn m

m C

Cn m ,

where |m are the eigenstates of C C . The operators C act on these states as follows: C |m = m |m 1 ,

C | 0 = 0. Therefore, we nd m C m C
n

C n m = m (m 1) . . . (m n + 1) = C n m = 0 for n > m. 63

m! for n m, (m n)!

 T r e and Tr e
C C iQC iQ C CC

Cn =

m=n

m! , (m n)! m! (m n)! m n
k

= = = =

n=0 n=0 n=0 n=0

(1)n |Q|2n (n!)2 (1) |Q| n!

n n 2n

m=n m=n

e e
n

(1) |Q| e n! (1) |Q| e n!

n=0 n 2n

2n

k =0

e 1

k+n n

1 = 1 e 1 = 1 e In particular, for Q = Q = 0, we have T r e

C C

|Q|2 exp e 1

(1)n |Q|2n 1 n! e 1 .

1 e

n+1

1 1 e

(4.98)

As a result, the expectation value eiQC eiQ eiQC eiQ with

C

is (4.99)

= exp n () |Q|2 , 1 e 1

n () Using this result, we obtain the expression eiQC eiQ

C

= exp [n () Q Q ] = exp n () a2 k 2 1 ei = exp 4n () a2 k 2 sin2 1 2

1 ei .

64

The expectation value ei M kP is e

i M

k P

P dP exp 2 iM kP M P dP exp 2 M (iP +k)2 2M

2

dP exp =

P dP exp 2 M k2 2 . = exp 2M

k 2 2 2M

Collecting all factors in Eq. (4.97) together, we nd eikr eikr() = exp i k2 k22 a2 k 2 sin2 4n 2M 2M eikr(t) eikr(+t) = ek with the function D (t) = where v M= w According to (4.78), eikr(t) eikr(+t) = eikr eikr() . Taking = t in (4.100) we nally nd the Fourier component of the electron density2

1 a2 k 2 1 ei 2

. (4.100)

2 D ( )

2M

it +

t2

t) + 4n sin2 + a2 1 exp(i = vLO , a2 = mb ,

t 2

(4.101)

v2 w2 . 2mb LO v 3

(4.102)

density correlation function eikr(t) eikr which enters Eq. (4.74): eikr(t) eikr = exp k 2 D (t) Finally, the correlation functions in (4.73) reduce to

(4.103)

2 bk (t)b k (0) = [1 + n(k )] exp (ik t) exp k D (t) , 2 b k (0)bk (t) = n(k ) exp (ik t) exp k D (t) ,

bk (t)bk (0) = 0, bk (0)bk (t) = 0. Inserting these equations into Eqs. (4.72) and (4.73), one obtains 1 (z ) = z
0

dt 1 eizt Im S (t) 65

(4.104)

with Im S (t) = 2 mb [1 + n( )] exp (i t) exp [k 2 D (t)] k k 2 kx |Vk |2 Im . +n(k ) exp (ik t) exp [k 2 D (t)]

Using the property D (t) =2 D (t) for real vaues of t,one obtains

Im exp (ik t) exp k 2 D (t) } = Im exp (ik t) exp k 2 D (t) } = Im exp (ik t) exp k 2 D (t) } and consequently S (t) = 2 mb
2 kx |Vk |2 exp k 2 D (t) {[1 + n(k )] exp (ik t) + n(k ) exp (ik t)} .

(4.105) Owing to the rotational invariance of |Vk | and k ,we can substitute in (4.105)
2 kx 2

k2 . 3

The resulting expression for S (t) = 2 3mb k 2 |Vk |2 exp k 2 D (t) {[1 + n(k )] exp (ik t) + n(k ) exp (ik t)} (4.106) is identical with Eq. (35) of FHIP [45]. In the case of Fr ohlich polarons, taking into account (1.17), Eq. (4.106) simplies to S (t) = LO mb
3/2

B.

Calculation of the memory function (Devreese et. al. [48])

LO )] exp (iLO t) [D (t)]3/2 . 3 2 +n(LO ) exp (iLO t)

[1 + n(

(4.107)

Upon substituion of (4.30) and (4.57) into Eq. (4.5), we nd the absorption coecient () = 1 e2 ie2 1 1 1 = Re Im n0 c mb () n0 c mb () 2 1 e () = Im n0 c mb [ Re ()]2 + [Im ()]2 1 e2 Im () = n0 c mb [ Re ()]2 + [Im ()]2 66

() =

This general expression was the strating point for a derivation of the theoretical optical absorption spectrum of a single large polaron, at all electron-phonon coupling strengths by Devreese et al. in Ref. [48]. The memory function () as given by Eq. (4.104) with (4.107) contains the dynamics of the polaron and depends on , temperature and .Following the notation, introduced in Ref. [45], () = ()

Im () 1 e2 . n0 c mb [ Re ()]2 + [Im ()]2

(4.108)

(4.109)

we reresent Eq. (4.108) in the form used in Ref. [48]: () = Im () 1 e2 . n0 c mb [2 Re ()]2 + [Im ()]2
0

(4.110)

According to (4.104) and (4.109), Im () = Im

0

dt sin(t)S (t),

Re () = Im

dt [1 cos(t)] S (t).

(4.111)

In the present Notes we limit our attention to the case T = 0 ( ). It was demonstrated in Ref.[48] that the exact zero-temperature limit arises if the limit is taken directly in the expressions (4.111) (see Appendices A and B of Ref.[48]). As follows from (4.107), S (t) = Accorting to (4.101) D (t) = i t t) + a2 1 exp(i 2M ( ). LO mb
3/2

exp (iLO t) [D (t)]3/2 3 2

( ).

(4.112)

Using the Feynman units (where M= and consequently D (t) = with

= 1, LO = 1 and mb = 1), we obtain from (4.102): v w

2

= v, a2 = ,

1 v2 w2 , 2 v3

1 w 1 v2 w2 (1 eivt ) i 3 2 v 2 v

t=

1 w 2 v

R(1 eivt ) it

( ) (4.113)

v2 w2 R= . w2v

67

and according to (4.112) 2 S (t) = 3 v w

3

eit R(1 eivt ) it

3/2

( ).

(4.114)

From (4.111) one obtains immediately 2 Im () = 3 2 Re () = 3 v w v w

3

Im
0

dt dt

sin(t)eit [R(1 eivt ) it]3/2 [1 cos(t)] eit [R(1 eivt ) it]3/2

, .

(4.115) (4.116)

Im
0

In the limit the function Im () was calculated by FHIP [45]. However, to study the optical absorption to the same approximation as FHIPs treatment of the impedance, we have also to calculate Re () and use this result in (4.110). The calculation of Re (), which is a Kramers-Kronig-type transform of Im (), is a key ingredient in Ref. [48]. The details of those calculations are presented in the Appendices A, B and C to Ref. [48]. Developing the denominator of both integrals on the right-hand side of (4.115) and (4.116), the calculations are reduced to the evaluation of a sum of integrals of the type Im
0

dt

sin(t)ei(1+nv)t (R it)3/2+n

Im
0

dt

cos(t)ei(1+nv)t (R it)3/2+n

(4.117)

In Appendix B to Ref. [48] it is shown how such integrals are evaluated using a recurrence formula. For Im () a very convenient result was found in [48]: 2 v Im () = 3 w
3 n=0 n n C 3/2 (1)

R n 2n (2n + 1)...3 1

| 1 nv |n+1/2 e|1nv|R

1 + sgn( 1 nv ) . 2

(4.118)

This expression is a nite sum and not and innite series. FHIP gave the rst two terms of (4.118) explicitly. Using the same recurrence relation it is seen the analytical expression (see Appendix B to Ref. [48]), which was found for Re () is far more complicated. To circumwent the diculty with the numerical treatment of Re (), the corresponding integrals in (4.117) have been

68

transformed in [48] to integrals with rapildy convergent integrands: Im

0

dt

[1 cos(t)] ei(1+nv)t (R it)3/2+n

0

= ln

1 ) (n + 3 2

1 dx (n + )xn1/2 eRx Rxn+1/2 eRx 2

1/2

(1 + nv + x)2 2 (1 + nv + x)2

(4.119)

The integral on the right-hand side of (4.119) is adequate for computer calculations. In Appendix C to Ref. [48] some supplementary details of the computation of (4.119) are given. Another analytical representation for the memory function (4.104) was derived in Ref. [74].

C.

Discussion of optical absorption of polarons at arbitrary coupling

At weak coupling, the optical absorption spectrum (4.108) of the polaron is determined by the absorption of radiation energy, which is reemitted in the form of LO phonons. For 5.9, the polaron can undergo transitions toward a relatively stable RES (see Fig. 7).

The RES peak in the optical absorption spectrum also has a phonon sideband-structure, whose average transition frequency can be related to a FC-type transition. Furthermore, at zero temperature, the optical absorption spectrum of one polaron exhibits also a zerofrequency central peak [ ()]. For non-zero temperature, this central peak smears out and gives rise to an anomalous Drude-type low-frequency component of the optical absorption spectrum. For example, in Fig. 7 from Ref. [48], the main peak of the polaron optical absorption for = 5 at = 3.51LO is interpreted as due to transitions to a RES. A shoulder at the low-frequency side of the main peak is attributed to one-phonon transitions to polaronscattering states. The broad structure centered at about = 6.3LO is interpreted as a FC band. As seen from Fig. 7, when increasing the electron-phonon coupling constant to =6, the RES peak at = 4.3LO stabilizes. It is in Ref. [48] that the all-coupling optical absorption spectrum of a Fr ohlich polaron, together with the role of RES-states, FC-states and scattering states, was rst presented. Recent interesting numerical calculations of the optical conductivity for the Fr ohlich polaron performed within the diagrammatic Quantum Monte Carlo method [77], see Fig. 8, 69

FIG. 7: Optical absorption spectrum of a polaron calculated by Devreese et al. [48] = 5 and 6. The RES peak is very intense compared with the FC peak. The frequency /LO = v is indicated by the dashed lines.)

fully conrm the essential analytical results derived by Devreese et al. in Ref. [48] for

3.

In the intermediate coupling regime 3 < < 6, the low-energy behavior and the position of the RES-peak in the optical conductivity spectrum of Ref. [77] follow closely the prediction of Ref. [48]. There are some minor qualitative dierences between the two approaches in the intermediate coupling regime: in Ref. [77], the dominant (RES) peak is less intense in the Monte-Carlo numerical simulations and the second (FC) peak develops less prominently. There are the following qualitative dierences between the two approaches in the strong coupling regime: in Ref.[77], the dominant peak broadens and the second peak does not develop, giving instead rise to a at shoulder in the optical conductivity spectrum at = 6. This behavior has been tentatively attributed to the optical processes with participation of two [73] or more phonons. The above dierences can arise also due to the fact that, within the Feynman polaron model, one-phonon processes are assigned more oscillator strength and the RES tends to be more stable as compared to the Monte-Carlo result. The nature of the excited states of a polaron needs further study. An independent numerical simulation might be called for.

70

FIG. 8: Left-hand panel : Monte Carlo optical conductivity spectra of one polaron for the weakcoupling regime (open circles) compared to the second-order perturbation theory (dotted lines) for = 0.01 and = 1 and to the analytical DSG calculations [48] (solid lines). Right-hand panel : Monte Carlo optical conductivity spectra for the intermediate coupling regime (open circles) compared to the analytical DSG approach [48] (solid lines). Arrows point to the two- and threephonon thresholds. (From Ref. [77].)

71

In Fig. 9, Monte-Carlo optical conductivity spectrum of one polaron for = 1 compares well with that obtained in Ref. [78] within the canonical-transformation formalism taking into account correlation in processes involving two LO phonons. The dierence between the results of these two approaches becomes less pronounced when decreasing the value of = 1 and might be indicative of a possible precision loss, which requires an independent check.

FIG. 9: One-polaron optical conductivity Re ( ) for = 1 calculated within the Monte Carlo approach [77] (open circles) and derived using the expansion in powers of up to 2 [78] (solid curve).

The coupling constant of the known ionic crystals is too small ( < 5) to allow for the experimental detection of sharp RES peaks, and the resonance condition = Re() cannot be satised for 5.9 as shown in Ref, [48]. Nevertheless, for 3 5.9 the

development of RES is already reected in a broad optical absorption peak. Such a peak, predicted in Ref. [48], was identied, e. g., in the optical absorption of Pr2 NiO4.22 in Ref. [79]. Also, the resonance condition can be fullled if an external magnetic eld is applied; the magnetic eld stabilizes the RES, which then can be detected in a cyclotron resonance peak.

72

1.

Sum rules for the optical conductivity spectra

In this section, we analyze the sum rules for the optical conductivity spectra obtained within the DSG approach [48] with those obtained within the diagrammatic Monte Carlo calculation [77]. The values of the polaron eective mass for the Monte Carlo approach are taken from Ref. [28]. In Tables 3 and 4, we represent the polaron ground-state E0 and the following parameters calculated using the optical conductivity spectra:
max

M0 M1

Re ( ) d,
1 max

(4.120) (4.121)

Re ( ) d,
1

where max is the upper value of the frequency available from Ref. [77], 0 + M 2m
max

Re ( ) d,
1

(4.122)
n0 e 2 , mb LO

where m is the polaron mass, the optical conductivity is calculated in units

m is

measured in units of the band mass mb , and the frequency is measured in units of LO . The values of max are: max = 10 for = 0.01, 1 and 3, max = 12 for = 4.5, 5.25 and 6, max = 18 for = 6.5, 7 and 8.

73

Table 3. Polaron parameters obtained within the diagrammatic Monte Carlo approach M0
(MC)

(MC) M (MC) / E (MC) m(MC) M 0 1 0 0.634 0.010

0.01 0.00249 1.0017 1.5706

1 0.24179 1.1865 1.5657 0.65789 1.013 3 0.67743 1.8467 1.5280 0.73123 3.18 4.5 0.97540 2.8742 1.5219 0.862 4.97 5.25 1.0904 3.8148 1.5022 0.90181 5.68 6 6.5 7 8 1.1994 5.3708 1.4919 0.98248 6.79 1.30 6.4989 1.5417 1.1356 7.44 8.31 9.85 1.3558 9.7158 1.5175 1.2163 1.4195 19.991 1.4981 1.3774

74

Table 4. Polaron parameters obtained within the path-integral approach M0

(DSG) (DSG) (DSG) (Feynman) 0 m(Feynman) M M1 / E0

0.01 0.00248 1 0.24318 3 0.69696 4.5 1.0162 5.25 1.1504 6 1.2608

1.0017 1.1957 1.8912 3.1202 4.3969 6.8367 9.7449 14.395 31.569

1.5706 1.5569 1.5275 1.5196 1.5077 1.4906 1.5269 1.5369 1.5239

0.633 0.65468 0.71572 0.83184 0.88595 0.95384 1.1192 1.2170 1.4340

0.010 1.0130 3.1333 4.8394 5.7482 6.7108 7.3920 8.1127 9.6953

6.5 1.3657 7 8 1.4278 1.4741

The parameters corresponding to the Monte Carlo calculation are obtained using the numerical data kindly provided by A. Mishchenko. The comparison of the zero frequency (MC) and M (DSG) with each other and with the value /2 corresponding to the moments M
0 0

sum rule [80] + 2m

1

Re ( ) d =

2
2 (MC) 0 M , 2

(4.123)
(DSG) 0 M ,

(MC) (DSG) 0 0 is smaller than each of the dierences shows that M M

which appear due to a nite interval of the integration in (4.120), (4.121). We analyze also the fullment of the ground-state theorem [81] E0 () E0 (0) = using the rst-frequency moments M1 3
0

Re (, ) d

(4.124)

(MC)

and M1

(DSG)

. The results of this comparison are

presented in Fig. 10. The dots indicate the polaron ground-state energy calculated using the Feynman variational principle. The solid curve is the value of E0 () calculated numerically using the optical conductivity spectra and the ground-state theorem with the DSG optical conductivity [48] for a polaron, E0
(DSG)

()

Re (DSG) (, ) d.

(4.125)

75

The dashed and the dot-dashed curves are the values obtained using M1
(MC) M1

(DSG)

() and

(), respectively:
0 0

3 (DSG) 0 E ( ) (MC) () 3 E 0

d d

max

Re
0 max

(DSG)

3 (, ) d =

0 0

(DSG) M1 d

( )

, (4.126) (4.127)

Re
0 (DSG)

(MC)

3 (, ) d =

(MC) M1 d

( )

() to a high degree of accuracy coincides with the vari (DSG) () and E (MC) () dier from E (DSG) () ational polaron ground-state energy. Both E 0 0 0 (DSG) (MC) () due to the integration over a nite interval of frequencies. However, E () and E
0 0

As seen from the gure, E0

are very close to each other. Herefrom, a conclusion follows that for integrals over the nite frequency region characteristic for the polaron optical absorption (i. e., except the tails), (MC) () (4.127) reproduces very well the function E (DSG) (). the function E
0 0
10

E0 calculated variationally E0(DSG) E0(DSG) E0(MC)

(E0) (in units hLO)

~ ~

0 0 2 4 6 8

FIG. 10: The ground-state theorem for a polaron using dierent data for the optical conductivity spectra, DSG from Ref. [48] and MC from Ref. [77]. The notations are explained in the text.

76

D. 1.

Scaling relations Derivation of the scaling relations

The form of the Fr ohlich Hamiltonian in n dimensions is the same as in 3D, H= p2 + 2mb k a k ak +
k k ik r Vk ak eikr + Vk ak e ,

(4.128)

except that now all vectors are n-dimensional. In this subsection, dispersionless longitudinal phonons are considered, i.e., k = LO , and units are chosen such that = mb = LO = 1. The electron-phonon interaction is a representation in second quantization of the electron interaction with the lattice polarization, which in 3D is essentially the Coulomb potential 1/r. |Vk |2 is proportional to the Fourier transform of this potential, and as a consequence we have in n dimensions | Vk | 2 = An , Ln k n1 , kn k (4.129)

dimensional vector, can be obtained from |Vk |2 , where k = vector, by summing out one of the dimensions explicitly:
2 |Vk | =

2 where Ln is the volume of the n-dimensional crystal. Note that |Vk | , where k is an (n 1)-

is an n-dimensional

kn

|Vk |2 .

(4.130)

Inserting Eq. (4.129) into Eq. (4.130), we have An1 = n 2 Ln1 k An

kn (n1)/2

(4.131)

Ln

2 + k 2 k n

Replacing the sum in Eq. (4.131) by an integral, i.e., Ln1 Ln we obtain An1 An = n 2 2 k Since

kn

1 2

dkn ,

(4.132)

dkn 2 + k 2 k n
1 (n1)/2

(4.133)

dx 1 = z (z + x ) 77

()

(4.134)

we have

dkn 2 + k2 k n
(n1)/2

1 n 2 k

1 2

n 2 2 n 1 2

n 2 k

n 2 2 n 1 2

(4.135)

where (x) is eh function. Inserting Eq. (4.135) into Eq. (4.133), we obtain 1 2 n 2 An = An1 . 2 n 2 In 3D the interaction coecient is well known, |Vk |2 = 2 2/L3 k 2 , so that A3 = 2 2. Inserting Eq. (4.137) into Eq. (4.136), we immediately obtain 1 A2 = 2 2 2 = 2. 2 (1) Applying Eq. (4.136) n 2 times, we further obtain for n > 3 n 2 n 1 (2 ) An =
j =2 n 2 j =1 j 2

(4.136)

(4.137)

(4.138)

j 2

n 2 (2 ) A2 = 1 2 .

n 1 2

A2 = 2n2 (n3)/2

n1 2

A2

= 2n3/2 (n1)/2

n1 2

(4.139)

So, the interaction coecient in n dimensions becomes [82] |Vk |2 = 2n3/2 (n1)/2 Ln k n1
n 1 2

(4.140)

Following the Feynman approach [41], the upper bound for the polaron ground-state energy can be written down as E = E0 lim 1 S S0
0

(4.141)

where S is the exact action functional of the polaron problem, while S0 is the trial action functional, which corresponds to a model system where an electron is coupled by an elastic force to a ctitious particle (i.e., the model system describes a harmonic oscillator). E0 is the ground-state energy of the above model system, and F
0

F eS0 D r (t) . eS0 D r (t) 78

(4.142)

As indicated above, the Fr ohlich Hamiltonian in n dimensions is the same as in 3D, except that now all vectors are n-dimensional [and the coupling coecient |Vk |2 is modied in accordance with Eq. (4.140)]. Similarly, the only dierence of the model system in n dimensions from the model system in 3D is that now one deals with an n-dimensional harmonic oscillator. So, directly following [41], one can represent lim S S0 0 / as 1 S S0 lim where A=
k 2 0

= A + B,

(4.143)

|Vk |2

0 0

eik[r(t)r(0)]

et dt, ewt dt,

(4.144) (4.145)

B=

w (v w 2 ) 4

[r (t) r (0)]2

w and v are variational parameters, which should be determined by minimizing E of Eq. (4.141). Since the averaging ... in eik[r(t)r(0)]
0

in Eq. (4.144) is performed with the trial action, which cor-

responds to a harmonic oscillator, components of the electron coordinates, rj (j = 1, ..., n),

0

separate [41]:
n

eik[r(t)r(0)]

=
j =1

eikj [rj (t)rj (0)]

(4.146)

For the average eikj [rj (t)rj (0)] 0 , Feynman obtained [41] eikj [rj (t)rj (0)] where D0 (t) =
0

= ekj D0 (t) ,

(4.147)

v2 w2 w2 t + 1 evt . 2v 2 2v 3

(4.148)

Inserting Eq. (4.146) with Eq. (4.147) into Eq. (4.144), we obtain A=
0

et dt
k

|Vk |2 ek

2D

0 (t)

(4.149)

Inserting expression (4.140) for |Vk |2 into Eq. (4.149) and replacing the sum over k by an integral [see (4.132)], we have A=2
n3/2 (n1)/2

n1 2 n1 2

e dt et dt 79

ek D0 (t) dk k n1 (2 )n dn
0

= 2n3/2 (n1)/2

ek D0 (t) k n1 dk , k n1 (2 )n

(4.150)

where dn is the elemental solid angle in n dimensions. Since the integrand in Eq. (4.150) depends only on the modulus k of k, one have simply n = So, we obtain for A the result 21/2 1/2 A= n 2 = 23/2
n 1 2 n 2 n 1 2

dn = n with (4.151)

2 n/2 . n 2 2

e dt
0

k 2 D0 (t)

21/2 1/2 dk = n 2

n 1 2

et D0 (t)

dt

et dt. D0 (t)

(4.152)

Like in Ref. [41], B can be easily calculated by noticing that

n n

[r (t) r (0)]

2 0

=
j =1 n

[rj (t) rj (0)]

2 0

=
j =1

2 eik[r(t)r(0)] 2 kj

0 k=0

=
j =1

2D0 (t) = 2nD0 (t) ,

(4.153)

so that B= = = = nw (v 2 w 2) D0 (t) ewt dt 2 0 nw (v 2 w 2) w 2 wt v 2 w 2 wt te + e e(v+w)t 2 3 2 2 v 2 v 0 v2 w2 1 1 nw (v 2 w 2) w 2 1 + 2 2 3 2 2v v 2v w v+w 2 2 n (v w ) . 4v

dt

(4.154)

Inserting Eq. (4.140) with A and B , given by Eqs. (4.152) and (4.154), together with the ground-state energy of the model system [41] (an isotropic n-dimensional harmonic oscillator), E0 = into Eq. (4.141), we obtain n (v w ) n (v 2 w 2 ) 23/2 E= 2 4v = n (v w ) 4v 2 2
2 n 1 2 n 1 2 n 2

n (v w ) , 2

(4.155)

et dt D0 (t) (4.156)

n 2 0

D0 (t)

dt.

80

In order to make easier a comparison of E for n dimensions with the Feynman result [41] for 3D, 3 (v w )2 1 E3D () = 4v 2
0

et D0 (t)
0

dt,

(4.157)

it is convenient to rewrite Eq. (4.156) in the form 1 n 3 (v w )2 1 3 n 2 EnD () = 3 4v 2 2n n 2

et D0 (t)

dt .

(4.158)

It is worth recalling that the parameters w and v must be determined by minimizing E . Thus, in the case of Eq. (4.158) one has to minimize the expression in the square brackets. The only dierence of this expression from the r.h.s. of Eq. (4.157) is that is multiplied by the factor 3 an = 2n
n 1 2 n 2

(4.159)

This means that the minimizing parameters w and v in nD at a given will be exactly the same as those calculated in 3D for the Fr ohlich constant as large as an : vnD () = v3D (an ) , wnD () = w3D (an ) . (4.160)

Therefore, comparing Eq. (4.158) to Eq. (4.157), we obtain the scaling relation [8284] EnD () = n E3D (an ) , 3 (4.161)

where an is given by Eq. (4.159). As discussed in Ref. [82], the above scaling relation is not an exact relation. It is valid for the Feynman polaron energy and also for the groundstate energy to order . The next-order term (i.e., 2 ) no longer satises Eq. (4.161). The reason is that in the exact calculation (to order 2 ) the electron motion in the dierent space directions is coupled by the electron-phonon interaction. No such a coupling appears in the Feynman polaron model [see, e.g., Eq. (4.146)]; and this is the underlying reason for the validity of the scaling relation for the Feynman approximation. In Refs. [82, 84], scaling relations are obtained also for the impedance function, the eective mass and the mobility of a polaron. The inverse of the impedance function Z ( ) is given by i 1 = , Z ( ) ( ) where the memory function ( ) can be expressed as [85] 1 (z ) = z
0

(4.162)

dt 1 eizt ImS (t) , 81

(4.163)

with z = + i0+ and S (t) =

k 2 2k1 |Vk |2 ek
2 D (t)

T (t) ,

(4.164)

T (t) = [1 + n (1)] et + n (1) eit , (4.165) 2 2 2 2 w v w t vt ivt 2 D (t) = 2 it + + . (4.166) 1 e + 4 n ( v ) sin 2v 2v 3 2 Here, is the inverse temperature and n ( ) is the occupation number of phonons with frequency (recall that in our units LO = 1). As implied from Eqs. (4.162) and (4.163), scaling of ( ) and Z ( ) is determined by scaling of S (t). For an isotropic crystal, since |Vk |2 , D (t) and T (t) do not depend on the direction of k, one can write
k 2 kn |Vk |2 e k 2 D (t) k 2 k1 |Vk |2 ek
2 D (t)

T (t) =

2 k2 |Vk |2 ek

2 D (t)

T (t) = ... =

T (t) , so that S (t) = 2 n k 2 |Vk |2 ek

2 D (t)

T (t) .

(4.167)

Inserting expression (4.140) for |Vk |2 and replacing the sum over k by an integral, we have S (t) = 2 2n3/2 (n1)/2 n (2 )n 2 2n3/2 (n1)/2 = n (2 )n =
1 2 n 2 n n 2 0 n 1 2 n 1 2

dn

0 0

k 2 ek

2 D (t)

T (t) dk

2 n/2 n 2
2 D (t)

k 2 ek

2 D (t)

T (t) dk (4.168)

k 2 ek

T (t) dk.

In particular, for 3D one has from Eq. (4.168) 2 2 2 S3D (; t) = k 2 ek D(t) T (t) dk. 3 0 For nD, Eq. (4.168) can be rewritten is the form 1 2 2 3 n 2 2 SnD (; t) = k 2 ek D(t) T (t) dk n 3 2n 2 0 2 2 2 an = k 2 ek D(t) T (t) dk. 3 0

(4.169)

(4.170)

So, the only dierence of the expression for S3D (t) from S3D (t) is that is multiplied by an . Since for the minimizing parameters w and v , which enter D (t), scaling is determined by the same product with an [see Eq. (4.160)], we can write SnD (; t) = S3D (an ; t) , 82 (4.171)

so that [84] nD (; ) = 3D (an ; ) , and ZnD (; ) = Z3D (an ; ) . (4.173) (4.172)

The polaron mass at zero temperature can be obtained from the impedance function as [85, 86] m Re ( ) = 1 lim , 0 mb

(4.174)

so that from the scaling relation (4.172) for the memory function we also have a scaling relation for the polaron mass [84]: m m (an ) n D ( ) = 3D . (mb )nD (mb )3D Since the mobility can be obtained from the memory function as [87] mb Im ( ) 1 = lim , e 0 (4.176) (4.175)

fullment of the scaling relation (4.172) implies also a scaling relation for the mobility [84]: nD () = 3D (an ) . (4.177)

In the important particular case of 2D, the above scaling relations take the form [8284]: 2 E2D () = E3D 3 Z2D (; ) = Z3D 3 , 4 3 ; , 4 (4.178) (4.179) (4.180) (4.181)

m 3 m 2D () = 3D 4 , (mb )nD (mb )3D 2D () = 3D 3 . 4

83

2. energy

Check of the scaling relation for the path integral Monte Carlo result for the polaron free

The fullment of the PD-scaling relation [84] is checked for the path integral Monte Carlo results [43] for the polaron free energy. The path integral Monte Carlo results of Ref.[43] for the polaron free energy in 3D and in 2D are given for a few values of temperature and for some selected values of . For a check of the scaling relation, the values of the polaron free energy at = 10 are taken from Ref. [43] in 3D (Table I, for 4 values of ) and in 2D (Table II, for 2 values of ) and plotted in Fig. 11, upper panel, with squares and open circles, correspondingly. The PD-scaling relation for the polaron ground-state energy as derived in Ref. [84] reads: 2 E2D () E3D 3 3 4 . (4.182)

In Fig. 11, lower panel, the available data for the free energy from Ref [43] are plotted in the following form inspired by the l.h.s. and the r.h.s parts of Eq. (1) : F2D () (squares)
2 and 3 F3D 3 4

(open triangles). As follows from the gure, the path integral Monte Carlo

results for the polaron free energy in 2D and 3D very closely follow the PD-scaling relation of the form given by Eq. (4.182): 2 F2D () F3D 3 3 4 . (4.183)

84

FIG. 11: Upper panel: The values of the polaron free energy in 3D (squares) and 2D (open circles) obtained by Ciuchi2001 [43] for = 10. The data for F3D () are interpolated using a polynomial t to the available four points (dotted line). Lower panel: Demonstration of the PDscaling cf. PD1987: the values of the polaron free energy in 2D obtained by Ciuchi2001 [43] for = 10 (squares) are very close to the PD-scaled according to PD1987 [84] values of the polaron free energy in 3D from Ciuchi2001 for = 10 (open triangles). The data for interpolated using a polynomial t to the available four points (solid line).
2 3 F3D 3 4

are

Appendix 1. Weak coupling: LLP approach

Inspired by the work of Tomonaga on quantum electrodynamics (Q. E. D.), Lee, Low and Pines (L.L.P.) [29] derived (1.22) and m = mb (1 + /6) from a canonical transformation formulation, which establishes (1.22) as a variational upper bound for the ground-state energy. 85

The wave equation corresponding to the Fr ohlich Hamiltonian (1.16) is

H = E . We shall take advantage of the fact that the total momentum of the system ka k ak + p
k

(4.184)

Pop =

(4.185)

(where p = i is the momentum of the electron) is a constant of motion because it commutes with the Hamiltonian (1.16) Indeed, [p, H ] = [p,
k ik r (Vk ak eikr + Vk a )] = ke k ik r k(Vk ak eikr Vk a ); ke ik r (Vk ak p, eikr + Vk a ) k p, e

=
k

ka k ak , H =
k k

ka k ak ,
k

ik r ) = (Vk ak eik r + Vk a k e

=
k

ik r ik r k a + Vk a ) = k ak , (Vk ak e ke ik r ik r = k Vk a + Vk a k ak , ak e k ak , ak e k ik r k Vk ak eikr Vk a ; ke

= =

[Pop , H ] =
k

ka k ak + p, H = 0.

(4.186)

Because of the commutation (4.186), the operators H and Pop have a common set of basis functions: H = E and Pop = P. It is possible to transform to a representation in which Pop becomes a c number P, and in which the Hamiltonian no longer contains the electron coordinates. The unitary (canonical) transformation required is = S1 , where i (P 86 ka k ak )r .

S1 = exp

(4.187)

Derivation of the transformations of the operators.

1 p S1 pS1 =

i = exp (P i = exp (P i = exp (P =P

ka k ak )r p exp

(P i

ka k ak )r ka k ak )r
i

ka k ak )r (i ) exp ka k ak )r (P
k

(P

ka k ak ) exp
i

+ exp

(P

ka k ak )r

(P

ka k ak )r

(i )

(4.188)

ka k ak + p,
k

1 Pop S1 Pop S1 =

i = exp (P = exp =
k

ka k ak )r

ka k ak + p exp
k

(P

ka k ak )r

i
k

ka k ak r +P

ka k ak exp

i
k

1 ka k ak r + S1 pS1

ka k ak

ka k ak
k

+ p = P + p,

(4.189)

87

1 ak S1 ak S1 =

i = exp (P = exp = exp i

k

ka k ak )r ak exp i
k

(P

ka k ak )r

ka k ak r ak exp ak exp

ka k ak r

ika k ak r

= exp ika k ak r ak = exp ika k ak r = exp ik ra k ak = exp ik ra k ak

n 1 ika ak r k n=0 n! n 1 ak ika ak r k n=0 n! n see() 1 (ik r)n ak a ak = k n=0 n! 1 n [ik r a k ak + 1 ] ak n=0 n!

ika k ak r

= exp ik ra k ak exp ik r(ak ak + 1) ak

= ak exp (ik r) . Here the property was used: ak a k ak

n

(4.190)

= a k ak + 1

ak .

(*)

It is evident for n = 0.For n = 1 it is demonstrated as follows:

ak a ak a k = ak ak + 1 = k ak = (ak ak + 1)ak ;

then for n 2 the validity of (*) is straightforwardly demonstrated by induction. Finally,

1 1 a k S1 ak S1 = [S1 ak S1 ] = ak exp (ik r) .

(4.191)

Using (4.188), (4.189), (4.190) and (4.191), one arrives at

1 H H = S1 HS1 =

ka k ak

2mb

+
k

LO a k ak +
k

(Vk ak + Vk a k ), (4.192)

88

where p is set 0.

The wave equation (4.184) takes the form HS1 = ES1 = H = E.

(4.193)

Our aim is to calculate for a given momentum P the lowest eigenvalue E (P ) of the Hamiltonian (4.192). For the low-lying energy levels of the electron E (P ) may be well represented by the rst two terms of a power series expansion in P 2 : E (P ) = E0 + P 2 /2mp + O (P 4),where mp is the eective mass of the polaron. The canonical transformation (4.187) formally eliminates the electron operators from the Hamiltonian. LLP use further a variational method of calculation. The trial wave function is chosen as

= S2 0

(4.194)

where 0 is the eigenstate of the unperturbed Hamiltonian with no phonons present (vacuum state). Specically, 0 is dened by ak 0 = 0, and the second canonical transformation: S2 = exp
k (a k fk ak fk ) ,

(0 , 0 ) = 1

(4.195)

(4.196)

where fk are treated as variational functions and will be chosen to minimize the energy. The physical signicance of Eq. (4.196) is that it dresses the electron with the virtual phonon eld, which describes the polarization. Viewed as a unitary transformation, S2 is a displacement operator on ak and a k :
2 1 Transformation of the equation Pop = P leads to S1 Pop S1 = P.At the same time, applying Eq. 1 (4.189), we obtain S1 Pop S1 = P + p.Setting the gauge p = 0 eliminates the operator p from the problem.

89

1 ak S2 ak S2 =

= exp = exp

(a k fk ak fk ) ak exp

(a k fk ak fk ) see()

(a k fk

ak fk )

ak exp 1 2

(a k fk

ak fk )

= ak + ak , (a k fk ak fk ) +

ak , (a k fk ak fk ) , (ak fk ak fk ) + ...

= ak + fk ,

(4.197)

1 a k S2 ak S2 = ak + fk .

(4.198)

Here the relation was used exp [V ] a exp [V ] = a + [a, V ] + 1 1 [[a, V ] , V ] + [[[a, V ] , V ] , V ] + ... 2 3! (**)

Further we seek to minimize the expression for the energy,

1 E = (, H ) = 0 , S2 HS2 0 .

(4.199)

In virtue of (4.197) and (4.198), we obtain:

1 S2 H S2 =

k a k + fk (ak + fk )

2mb LO
k

a k
k

+ fk (ak + fk ) + k

Vk (ak + fk ) + Vk a k + fk k a k fk + ak fk 2

= +

(P LO
k

ka k ak ) a k ak
2

k |fk |2 2mb
+ ak fk

+ |fk | +

a k fk

+
k

Vk (ak + fk ) + Vk a k + fk

= H0 + H1 ,

90

where H0 = +
k

(P |fk |2 ak
k

ka k ak )

k |fk |2 +

2mb
2 2 kP k LO + mb 2mb

+
k 2

[Vk fk + Vk fk ]

mb

ka k ak

k |fk |2

+ +
k

Vk +

fk

2 2 2 kP k k LO + + mb 2mb mb

k |fk |2 k |fk |2 (4.200)

a k

Vk + fk

LO

kP k k + + mb 2mb mb

2 2

+
k

LO a k ak ;

H1 =
k, k

k k fk + 2a ak ak fk k ak fk fk + ak ak fk fk + 2mb

+
k, k

k k ak ak ak fk . + ak ak ak fk 2mb

Using (4.195), we obtain from (4.199) that E = H0 = +

k

P2 +

2mb LO

k |fk |2

+
k 2 2

[Vk fk + Vk fk ]

|fk |2

k kP + mb 2mb

(4.201)

We minimize (4.201) by imposing E = 0, fk This results in

Vk + fk

E
fk

= 0.

LO

2 2 2 k kP + + mb 2mb mb

k |fk |2 k

=0

(4.202)

and the appropriate complex conjugate equation for fk . Upon comparing (4.202) and (4.200), we see that the linear terms in a k and ak are identically zero if (4.202) is satised, and hence that H0 is diagonal in a representation in which a k ak is diagonal. So, the variational
calculation by LLP is equivalent to the use of (4.200) as the total Hamiltonian provided fk

91

satises (4.202). An estimate of the accuracy of the LLP variational procedure was obtained by an estimate of the eect of H1 using a perturbation theory. Now we evaluate the energy of the ground state of the system, which is given by Eq.
(4.201) with fk satisfying Eq. (4.202). The only preferred direction in this problem is P.

Therefore one may introduce the parameter dened as P =

k

k |fk |2 .

(4.203)

Then Eq. (4.202) leads to

fk = Vk

LO

2 2 kP k (1 ) + , mb 2mb

(4.204)

and we obtain the following implicit equation for : P=

k

k |Vk |

2 2 k kP (1 ) + LO mb 2mb

V = (2 )3

dk k

LO k

4 V

1 2

2mb LO

2 2 kP k LO (1 ) + mb 2mb

Let us introduce spherical coordinates with a polar axis along P and denote x = cos(k P):
2 3LO P = 2 2
1 2

2mb LO
1 2

2
1 1

dxx
0 0

dkk

2 2 kP x k LO (1 ) + mb 2mb 2 2 kP x k 12 (1 ) + 2mb LO 2mb LO

= 2 2

2mb LO

2
1

dxx

dkk

Further, we introduce the parameter q= and a new variable = k (2mb LO )1/2 . P (2mb LO )1/2 (1 ) (4.205)

92

This gives = =
1 2

2mb LO
1/2

2mb LO 2P
1

dxx
1 0

1 2qx + 2
2

(2mb LO ) P

dxx
1 1 0

( qx)2 + (1 q 2 x2 ) 2 + (1 q 2 x2 )

(2mb LO ) = P = (2mb LO ) P (2mb LO ) P

1/2

dxx
1 1

qx

d ( + qx)

1/2

dxx
1 1

1/2

dxx
1

+qx

2(1q 2 x2 )[2 +(1q 2 x2 )]

2[2 +(11q2 x2 )] +
2 2

1 2(1q 2 x2 )3/2

arctan

[1q 2 x2 ]1/2

qx

1 qx q x qx + + + arcsin (qx) . 2 2 3 / 2 2 2 2 4(1 q x ) 2(1 q x ) 2(1 q 2 x2 )3/2

(2mb LO )1/2 q 1 x2 dx = P 4 1 (1 q 2 x2 )3/2 1 x2 = (1 ) dx 2 2 3/2 4 1 (1 q x ) = = 2 =

(2mb

q 1 q 2 arcsin (q ) (1 ) 2 q3 1 q2 (1 )
P (1 LO )1/2

1 q2 q

arcsin (q ) arcsin (q )

2 (1 )2
2

2mb LO P2

3/2

1 q2
3/2

So, we have arrived at the equation (1 ) = 2 2mb LO P2 q 1 q2 q 1 q2 arcsin (q ) , (4.206)

or equivalently, using the denition (4.205), = 3 1 2q P 2 + ( P) 2 2 2mb arcsin (q ) . (4.207)

Using Eqs. (4.203) and (4.204), we can simplify the energy (4.201) as follows: E= +
k

|Vk |2 LO
k P (1 mb

| Vk | 2 k P LO m (1 ) + b
2 k2 2mb

2 k2 2mb

) + 93

LO

2 2 k kP + mb 2mb

= +

(1 + 2 ) P 2 2 2mb

| Vk | 2
k

|Vk |2 k P LO m (1 ) + b
2 k2 2mb

2 k2 2mb

LO

k P (1 mb

) +

LO

2 2 k kP (1 + ) + mb 2mb

= +

(1 + 2 ) P 2 2 2mb

| Vk | 2
k

|Vk |2 k P LO m (1 ) + b
2 k2

2 k2 2mb

LO LO

k P (1 mb

) +
2

LO kP mb

2 2 k kP (1 ) + mb 2mb

2mb

|Vk |
k k P (1 mb

) +

2 k2 2mb

(1 + 2 ) P 2 2 = 2mb +
k

LO
k

k |Vk |2 k P (1 mb

| Vk | 2 k P LO m (1 ) + b ) +
2 k2 2mb

2 k2 2mb

P mb

k |Vk |2
k P (1 mb

LO

) +
2 k2

2 k2

2mb

(1 + 2 ) P 2 2mb =

|Vk |2 k P (1 ) + LO m b P P mb
k k

2mb

P mb |Vk | k P (1 mb
2

k |fk |2
2 k2 2mb

(1 + 2 ) P 2 2mb

LO

) +

P2 1 2 E= 2mb

|Vk |2 k P (1 ) + LO m b

2 k2 2mb

(4.208)

94

The sum over k in Eq. (4.208) is calculated as follows: |Vk |2 k P (1 ) + LO m b V (2 )3

2 LO k 4 V
2 k2 2mb 1 2 1 4

2mb LO

dk
2 LO

LO 4 V

k P (1 mb

) +
1 2

2 k2 2mb

V = (2 )3

2mb LO
1 2

dk k2 k2 2
k P

1 LO
k P (1 mb

) + .

2 k2 2mb

2 mb LO = 2

2mb LO

dk

k2

1 (1 ) +

2mb LO

For the calculation of this integral, we can use the auxiliary identity 1 = ab Setting a = k2 2 b = k2 , we nd |Vk |2 k P LO m (1 ) + b
1 2 2 k2 2mb

1 0

1 . [ax + b (1 x)]2 2mb LO

(4.209)

kP

(1 ) +

2 mb LO = 2 2 mb LO = 2 2 mb LO = 2 2 mb LO = 2

2mb LO
1 2

dx
0 1

dk dk dk

1 x k2 2
k P

(1 ) + 1

2mb LO

+ (1 x) k 2
2

2mb LO
1 2

dx
0 1

k 2 2 kP (1 )x + k P (1 )x
2

2mb LO

2mb LO
1 2

dx
0 1

1 +
2mb LO

x
2

P2
2

(1 )2 x2

2mb LO

dx
0

dk

1 k2 +
2mb LO

P2
2

(1 )2 x2

As long as P 2 / (2mb ) is suciently small so that no spontaneous emission can occur (roughly P 2 / (2mb ) LO ), the quantity A 2mb LO x 95 P2
2

(1 )2 x2

is supposed to be positive for 0 < x < 1. Therefore, we can use the integral 1 2 , d k = (k 2 + A)2 A what gives |Vk |2 k P LO m (1 ) + b
1 2 2 k2 2mb

2 mb LO = 2

2mb LO
1

dx
0

2
2mb LO

P2
2

(1 )2 x2

1 = LO 2 1 = LO 2

dx
0 1

1 x
(1)2 P 2 2 x 2mb LO

dx
0

1 x q 2 x2
1

We change the variable x = t2 , what gives

1 0

1 x q 2 x2

dx = 2
0

1 1 q 2 t2 =

dt =

2 arcsin q, q

and hence
k

| Vk | 2 k P LO m (1 ) + b

2 k2

2mb

LO arcsin q. q

(4.210)

As a result, the energy (4.208) is expressed in a closed form E= P2 LO 1 2 arcsin q. 2mb q (4.211)

Further, we expand the r.h.s. of Eq. (4.207) to the second order in powers of P (or, what is the same, in powers of q ) using the relation q 1 what results in 1 3 = 3 q + O q5 1 2q 3 /6 +O 1 + /6 = + O q2 . 6 q2 1 arcsin (q ) = q 3 + O q 5 3 (4.212)

Solving this equation for , we nd =

P2 2mb LO

(4.213)

96

We also apply the expansion in powers of q up to q 2 to the energy (4.211): E= P2 1 LO q + q3 + O q5 1 2 2mb q 6 2 1 P 1 2 LO LO q 2 + LO O q 4 = 2mb 6 2 P 2 (1 )2 P 2 1 + LO O q 4 = LO + 2mb 12mb 2 2 P P (1 )2 2 = LO + 1 + LO O q 4 2mb 12mb P2 = LO + (1 ) ((6 + ) + 6) + LO O q 4 12mb /6 /6 P2 1 (6 + ) + 6 + LO O q 4 = LO + 12mb 1 + /6 1 + /6 2 P = LO + + LO O q 4 . 2mb (1 + /6)

Finally, we have arrived at the LLP polaron energy E = LO + P2 + LO O 2mb (1 + /6) P2 2mb LO
2

(4.214)

97

Appendix 2. Expansion in Stieltjes continuous fractions [52]

In this derivation it is shown that the approximation used in the evaluation of the function, which determines the polaron mass [see Eqs. (40) and (B1) from Ref. [52]]
0

g (k, z ) = with

d eiz exp k 2 C (0) exp k 2 C ( )

(4.215)

C ( ) =

1 3

ik k 2 2 e | | f k 2 2 m k

(4.216)

is equivalent to an expansion in a continued fraction limited to the rst step. Moreover, it is proved that the choice of the coecients of the continued fraction can be justied by a variational principle, at least when z is real and positive. Expanding the last exponential of Eq. (4.215) in a power series leads to
0

g (k, z ) = exp k 2 C (0)

d eiz

1 n! n=0

1 3m2

k1 ,...,kn

2 2 2 k1 k2 ...kn |fk1 |2 |fk2 |2 ... |fkn |2 2 2 2 k ...k n 1 k2

2 ) exp i(k1 + k2 + ...k n

= i exp k C (0)

(3m2 )n n! n=0

(4.217)

k1 ,...,kn

2 2 2 k1 k2 ...kn |fk1 |2 |fk2 |2 ... |fkn |2 1 . 2 2 2 k1 k2 ...kn k1 + k2 + ... + kn + z

The multiple sum over the k s is in fact an integral with 3n variables. It is possible to change the variables in that one of the new variables is xn = k1 + k2 + ... + kn . (4.218)

Then the multiple sum which appears in the last term of Eq. (4.217) is of the following type:

J (z ) =
n

L(xn ) dxn , xn + z

(4.219)

98

where

L(xn ) = 0 is the result of the integration over the n 1 other variables. An expansion of integrals of the type (4.219) into Stieltjes continued fractions is known to give good results when z is real and not located on the cut of J (z ), i.e., when z > n. The rst nontrivial step in the continued fraction expansion is J (z ) = with

(4.220)

a0 a1 + z

(4.221)

a0 =
n

L(xn )dxn ,

(4.222)

xn L(xn )dxn a1 = n L(xn )dxn

n

(4.223)

A variational principle can be established, which gives a rather strong argument in favour of the approximation (4.221). Let us introduce a variational parameter x writing

J (z ) =
n

L(xn ) dxn . (xn x ) + (z + x )

(4.224)

Performing two steps of the division, this relation becomes

1 J (z ) = z+x

L(xn )dxn

1 (z + x )2

(xn x )L(xn )dxn + K (z, x )

(4.225)

99

with 1 K (z, x ) = (z + x )2

(xn x )2 L(xn ) dxn . xn + z

(4.226)

The approximation consists of neglecting the term K (z, x ) in Eq. (4.225). As this term is positive [cf. (4.220)], the best approximation is obtained when it is minimum. Therefore let us use the freedom in the choice of x to minimize the expression (4.226),

K (z, x ) K (z, x ) = 2 x z+x 1 2 (z + x )2 which gives 1 2 (z + x )2 or 1 2 (z + x )3

(xn x )2 L(xn ) dxn = 0, xn + z

(4.227)

xn x +1 z+x

xn x L(xn )dxn = 0 xn + z

(4.228)

(xn x )L(xn )dxn = 0.

(4.229)

This provides us with the best value of the variational parameter

xn L(xn )dxn x = n L(xn )dxn

n

(4.230)

which is a1 [cf. Eq. (4.223)]. With this value of x and neglecting K (z, x ), the expression (4.225) of the calculated quantity J (z ) becomes 1 J (z ) = z+x

L(xn )dxn = J (z ) =

a0 , a1 + z

(4.231)

which is the rst step (4.221) of a Stieltjes continued fraction. 100

To prove that this value of x gives a minimum of K (z, x ), let us calculate the second derivative

2 K (z, x ) 6 = x 2 (z + x )3 + 2 (z + x )3

(xn x )L(xn )dxn (4.232)

L(xn )dxn .

Now the parameter x is replaced by its expression (4.230). The relation (4.232) becomes 2 2 K (z, x ) = x 2 (z + a1 )3 which is positive of z

L(xn )dxn ,

(4.233)

n , since it follows from relation (4.230) that a1 > n.

Our approximation is related to that used by Feynman which is based on the following inequality: esx es x , (4.234)

where the brackets denote the expectation value of the random variable x. For instance,

L(x)esx dx

sx

, L(x)ds

(4.235)

where L(x) is the non-normalized probability density of x. The Laplace transform of Eq. (4.234) gives
0

esz esx ds
0

esz es x ds,

(4.236)

which after integration becomes

a

L(x) dx x+z

L(x)ds =

x +z

a0 . a1 + z

(4.237)

The last inequality shows the relation with our procedure.

101

Part II

Many polarons
V. OPTICAL CONDUCTIVITY OF AN INTERACTING MANY-POLARON GAS A. Kubo formula for the optical conductivity of the many-polaron gas

The derivations in the present section are based on Ref. [51]. The Hamiltonian of a system of N interacting continuum polarons is given by:

H0 =
j =1

p2 j + 2mb
N

LO b+ q bq
q

+
q j =1

eiqrj bq Vq + eiqrj b+ q Vq +

e2 2

j =1 (=j )=1

1 , | ri rj |

(5.1)

where rj , pj represent the position and momentum of the N constituent electrons (or holes) with band mass mb ; b+ q , bq denote the creation and annihilation operators for longitudinal optical (LO) phonons with wave vector q and frequency LO ; Vq describes the amplitude of the interaction between the electrons and the phonons; and e is the elementary electron charge. This Hamiltonian can be subdivided into the following parts: H = He + Hee + Hph + Heph where He =
j =1 N

(5.2)

p2 j 2mb

(5.3)

is the kinetic energy of electrons, H e e e2 = 2

N N

j =1 (=j )=1

1 | ri rj |

(5.4)

is the potential energy of the Coulomb electron-electron interaction, Hph =

q

LO b+ q bq

(5.5)

102

is the Hamlitonian of phonons, and

N

Heph =

eiqrj bq Vq + eiqrj b+ q Vq q j =1

(5.6)

is the Hamiltonian of the electron-phonon interaction. Further on, we use the second quantization formalism for electrons, in which the terms He , Hee and Heph are He =
k,

k + a ak, , 2mb k, vq
+ a+ k+q, ak q, ak , ak, =

2 2

(5.7) 1 2 vq : q q :, (5.8) (5.9)

H e e = Heph =

1 2

q=0

k, k ,

q=0

+ Vq bq q + Vq bq q ,

where : ... : is the symbol of the normal product of operators, vq = 4e2 q 2 V (5.10)

is the Fourier component of the Coulomb potential, and

N

q =
j =1

eiqrj =
k,

a+ k+q, ak,

(5.11)

is the Fourier component of the electron density. The ground state energy of the many-polaron Hamiltonian (5.1) has been studied by L. Lemmens, J. T. Devreese and F. Brosens (LDB) [89], for weak and intermediate strength of the electron-phonon coupling. They introduce a variational wave function: |LDB = U | el where el
(0) (0)

(5.12)

represents the ground-state many-body wave function for the electron (or hole)

system and | is the phonon vacuum, U is a many-body unitary operator which determines a canonical transformation for a fermion gas interacting with a boson eld:
N

U = exp
j =1 q

+ iq r j fq aq eiqrj fq aq e

(5.13)

In Ref. [89], this canonical transformation was used to establish a many-fermion theory. The fq were determined variationally [89] resulting in fq = Vq q LO + 2mb S (q) 103
2 2

(5.14)

for a system with total momentum P =

pj = 0. In this expression, S (q) represents the

static structure factor of the constituent interacting many electron or hole system :
N N

NS (q) =
j =1 j =1

eiq(rj rj )

(5.15)

The angular brackets ... represent the expectation value with respect to the ground state. The many-polaron optical conductivity is the response of the current-density, in the system described by the Hamiltonian (5.1), to an applied electric eld (along the x-axis) with frequency . This applied electric eld introduces a perturbation term in the Hamiltonian (5.1), which couples the vector potential of the incident electromagnetic eld to the currentdensity. Within linear response theory, the optical conductivity can be expressed through the Kubo formula as a current-current correlation function: Ne2 1 ( ) = i + V mb V
0

eit [Jx (t), Jx (0)] dt.

(5.16)

In this expression, V is the volume of the system, and Jx is the x-component of the current operator J, which is related to the momentum operators of the charge carriers: J= q mb
N

pj =
j =1

q P, mb

(5.17)

with q the charge of the charge carriers (+e for holes, e for electrons) and P the total momentum operator of the charge carriers. The real part of the optical conductivity at temperature zero, which is proportional to the optical absorption coecient, can be written as a function of the total momentum operator of the charge carriers as follows : Re ( ) = e2 Re V m2 b 1
0

eit [Px (t), Px (0)] dt .

(5.18)

104

B.

Force-force correlation function

Let us integrate over time in (5.18) twice by parts as follows:

0 1 d [Px (t) , Px ] eitt t=0 dt Px (t) , Px i dt 0 it it 1 d = e H Px e H , Px eitt dt i 0 dt i 1 it it = dt e H [H, Px ] e H , Px eitt i 0 it it i 1 dt [H, Px ] , e H Px e H = eitt i 0 it it 1 dt Fx (0) , e H Px e H eitt = i 0 2 it it 1 Fx (0) , e H Px e H eitt = i t=0 it d it dt Fx (0) , e H Px e H eitt dt 0

dt [Px (t) , Px ] eitt eitt

= = =

1 i 1 i

[Fx , Px ] +
2

0 0

dt dt

Fx (0) , e Fx (0) , e

it

[H, Px ] e
it

it

eitt

[Fx , Px ] +
2

it

Fx (0) e

eitt eitt

it it 1 [Fx , Px ] + dt e H Fx (0) e H , Fx (0) i 0 1 [Fx , Px ] + dt [Fx (t) , Fx (0)] eitt , = ( + i )2 0

where F i [H, P] is the operator of the force applied to the center of mass of the electrons. Since both Fx and Px are hermitian operators, the average [Fx , Px ] is imaginary. Hence, for = 0, this term does not give a contribution into Re ( ) . As a result, integrating by parts twice, the real part of the optical conductivity of the many-polaron system is written with a force-force correlation function: Re ( ) = 1 e2 Re V 3 m2 b i
0

eit [Fx (t), Fx (0)] dt .

(5.19)

The force operator is determined as Fx = i [H, Px ] = [He + Hee + Hph + Heph, Px ] . 105

Further, we use the commutators: a+ k+q, ak, , Px =

k + kx a+ k+q, ak, , ak , ak ,

=
k

= =

kx k

+ + + a+ k+q, ak, ak , ak , + ak+q, ak , ak, ak , + a+ k , ak+q, ak , ak,

+ a+ k , ak , ak+q, ak,

+ kx kk a+ k+q, ak , k ,k+q ak , ak, + kx kk a+ k+q, ak, k ,k+q ak+q, ak,

= a+ k+q, ak,
k

kx (kk k ,k+q ) = qx a+ k+q, ak, ,

[q , Px ] = qx q . Hence, [He , Px ] = 0, [Hee , Px ] = 0, [Heph , Px ] =

+ Vq bq [q , Px ] + Vq bq [q , Px ] + qx Vq bq q Vq bq q ,

So, the commutator of the Hamiltonian (5.1) with the total momentum operator of the charge carriers leads to the expression for the force F = i
+ q Vq bq q Vq bq q .

(5.20)

This result for the force operator claries the signicance of using the force-force correlation function rather than the momentum-momentum correlation function. The operator product Fx (t)Fx (0) is proportional to |Vk |2 , the charge carrier - phonon interaction strength. This will be a distinct advantage for any expansion of the nal result in the charge carrier - phonon interaction strength, since one power of |Vk |2 is factored out beforehand. Substituting (5.20) into (5.19), the real part of the optical conductivity then takes the form: 1 e2 Re Re ( ) = V 3 m2 b
0

dteitt
q,q

qx qx

+ + Vq bq (t) + V q bq (t) q (t) , Vq bq + Vq bq q

(5.21)

Up to this point, no approximations other than linear response theory have been made. 106

C.

Canonical transformation

The expectation value appearing in the right hand side of expression (5.21) for the real part of the optical conductivity is calculated now with respect to the LDB many-polaron wave function (5.12). The unitary operator (5.13) can be written as U = exp
q

Aq q ,

+ Aq = fq bq f q bq ,

(5.22)

The transformed Hamiltonian (5.2) is denoted as = U 1 HU. H (5.23)

The momentum operator of an electron pj , the operator of the total momentum of electrons P and the phonon creation and annihilation operators are transformed by the unitary transformation (5.22) as follows: U 1 pj U = pj +
q

qAq eiqrj , qAq q ,

q + U 1 b+ q U = bq fq q .

(5.24) (5.25) (5.26)

U 1 P U = P +

U 1 bq U = bq fq q ,

As a result, after the transformation (5.22), the Hamiltonian takes the form (see Ref. [89]): = He + H ee + Hph + H eph + HN + Hppe, H where the terms are e e = 1 H 2 v q : q q :, v q = vq + 2
LO |fq |2 Vq fq Vq fq ,

(5.27)

(5.28)

q=0

eph = H +

q 2

(Vq LO fq ) bq q + Vq LO fq b+ q q

2mb

Aq
q k,

q2 + 2k q a+ k+q, ak, , N a+ k, ak,

k,

(5.29)

HN = N
q

LO |fq |2 Vq fq Vq fq ,

(5.30)

107

Hppe =

2mb

qq

q q Aq Aq q+q .

(5.31)

The exact expression for the real part of the conductivity (5.21) after the replacement of |0 by |LDB = U | el Re ( ) = 1 e2 Re V 3 m2 b el
(0) 0 (0)

is transformed to the approximate one

dteitt
q, q

qx qx
it

U 1
0

it

+ Vq bq + V q bq q e

Vq bq + dteitt
q,q

+ Vq bq

el

(0)

1 e2 Re V 3 m2 b el
(0)

qx qx it H

it

Vq bq +

+ V q bq

Uq e

Vq bq +
qx qx q, q

+ Vq bq

Uq

el

(0)

1 e2 Re V 3 m2 b
(0)

dteitt

0 H

el

it

+ Vq bq fq q + V q bq fq q

q e .

it

+ Vq bq f q q + Vq bq fq q

el

(0)

So, we have arrived at the expression Re ( ) = el

(0)

1 e2 Re V 3 m2 b

dteitt
q,q

qx qx

it

+ Vq bq fq q + V q bq fq q

q e
(0)

it

+ Vq bq f q q + Vq bq fq q

el

= Vq f Since q q = q q , and Vq fq q , the terms proportional to q q vanish after the

summation over q:
qx Vq fq q q = qq qx Vq fq q q = 0.

(5.32)

108

Hence we obtain the real part of the optical conductivity in the form Re ( ) = 1 e2 Re V 3 m2 b
(0) 0
it

dteitt
q,q

qx qx
it

el

+ Vq bq + V q bq q e

, (5.33)

+ Vq bq + Vq bq q

el

(0)

Introducing the factor J (q, q ) = el

(0)

it

+ Vq bq + V q bq q e

it

, (5.34)

+ Vq bq + Vq bq q

el

(0)

the optical conductivity can be written as Re ( ) = 1 e2 Re V 3 m2 b

0

dteitt
q,q

qx qx J (q, q ).

(5.35)
it

in Eq. (5.33) by the Hamiltonian interaction (5.28) and (5.31)]. Namely, we replace H 0 = He + H ee + Hph + HN . H In this case, we nd J (q, q ) = el
(0)

In the case of a weak electron-phonon coupling, we can neglect in the exponent e H of eph and Hppe [i. e., the renormalized Hamiltonian of the electron-phonon (5.33) the terms H

(5.36)

it

0 H

+ Vq bq + V q bq q e

it

0 H

+ Vq bq + Vq bq q

el

(0)

= |Vq |2 qq el

(0)

eiH0 t/ q bq eiH0 t/ q b+ q

iH0 t/ q bq eiH0 t/ q b+ qe

el

(0) (0)

= 2i|Vq |2 qq Im

el

(0)

eiH0 t/ q bq eiH0 t/ q b+ q

el

The time-dependent phonon operators are eiH0 t/ bq eiH0 t/ = bq eiLO t ,

109

so that we have J (q, q ) = 2i|Vq |2 qq Im eiLO t el = 2i|Vq |2 qq Im e = He + H e e + H N . where H Taking the expectation value with respect to the phonon vacuum, we nd J (q, q ) = 2i|Vq |2 qq Im eiLO t el eiHe t/ q eiHe t/ q el The optical conductivity (5.33) then takes the form: Re ( ) = 2e2 Im 3 V m2 b
0 (0) (0) (0) (0)

eiH0 t/ q eiH0 t/ q bq b+ q
(0)

el ,

(0)

el eiHe t/ q eiHe t/ q el

bq b+ q

(5.37)

dteitt
q e t/ iH

2 qx | Vq | 2 (0)

Im eiLO t

(0) el

q eiHe t/ q el
1 3

(5.38)
2 2 2 qx + qy + qz q2, =1 3

2 For an isotropic electron-phonon system, qx in 3D can be replaced by

what gives us the result Re 3D ( ) = e2 3V 3 m2 b 2 q 2 |Vq |2 Im

0

dteitt Im eiLO t F (q, t) ,

(5.39)

where the two-point correlation function is F (q, t) = el e

(0)
it

e H

q e

it

e H

q el

(0)

(5.40)

The same derivation for the 2D case, provides the expression Re 2D ( ) = 1 e2 A 3 m2 b q 2 |Vq |2 Im
0

dteitt Im eiLO t F (q, t) ,

(5.41)

where A is the surface of the 2D system.

D.

Dynamic structure factor

To nd the formula for the real part of the optical conductivity in its nal form, we introduce the standard expression for the dynamic structure factor of the system of charge carriers interacting through a Coulomb potential, 1 S (q, ) = 2N

el

(0) j,

eiq.(rj (t)r (0)) el

(0)

eit dt.

(5.42)

110

The dynamic structure factor is expressed in terms of the two-point correlation function as follows: 1 S (q, ) = 2N = 1 2N

el e

(0)

it

e H

q e

it

e H

q el

(0)

eit dt

F (q, t) eit dt =

1 F (q, ) 2N

1 S (q, ) = F (q, ) , 2N where F (q, ) is the Fourier image of F (q, t):

(5.43)

F (q, ) =

F (q, t) eit dt.

(5.44)

The function F (q, t) obeys the following property: F (q, t) = el + q e = el q e = el q e

(0) (0)
it it

(0)

it

e + it H e H q e

el el e

(0)

e H e H

q e

it

e H

(0)
it

q el

(0)

0 E

0 is the eigenvalue of the Hamiltonian H e: where E 0 (0) . e (0) = E H el el Herefrom, we nd that F (q, t) = e
it

0 E

el q e
it

(0)

it

e H

q el

(0)

= el e From (5.45), for the function

(0)

e H

q e

it

e H

q el

(0)

= F (q, t) .

(5.45)

B (q, t) Im eiLO t F (q, t) the following equality is derived: B (q, t) = Im eiLO t F (q, t) = Im eiLO t F (q, t) = Im eiLO t F (q, t) = B (q, t) , 111

(5.46)

B (q, t) = B (q, t) . The integral in Eq. (5.39) Im

0 0

(5.47)

dteitt Im eiLO t F (q, t) = Im

dteitt B (q, t)

is then transformed as follows: Im

0

dteitt B (q, t) =

1 2i 1 = 2i 1 = 2i 1 = 2i 1 = 2i 1 = 4
(0)

0 0 0

dteitt B (q, t) dteitt B (q, t) dteitt B (q, t) +

0 0 0

dteitt B (q, t) dteit+t B (q, t) dteit+t B (q, t)

dteit|t| B (q, t) dteit|t| 1 iLO t e F (q, t) eiLO t F (q, t) 2i

dteit|t| eiLO t F (q, t) eiLO t F (q, t) .

We can show that, as far as el

is the ground state, the integral

( n)

dteit|t| F (q, t)

is the total basis set of the eigenfunctions of for positive is equal to zero. Let el e . Using these functions we expand F (q, t): the Hamiltonian H F (q, t) =
n

el e
(n)

(0)

it

e H

q e
(0)

it

e H

el

( n)

el |q | el

(n)

(0)

=
n it|t|

el |q | el
(n) el n (n)

it e (E 0 E n ) ,

dte

F (q, t) = =

(0) |q | el (0)

i dteit+ (En E0 )t|t|

el |q | el

2 +

n E 0 E

= 0,

because for > 0, + gas results in:

n E 0 E

is never equal to zero.

So, rewriting expression (5.39) with the dynamic structure factor of the electron (or hole) e2 Re 3D ( ) = 6V 3 m2 b 1 Re 2D ( ) = 1 e2 4A 3 m2 b

q |Vq |

dtei(LO )t|t| F (q, t) , dtei(LO )t|t| F (q, t) ,

q 2 | Vq | 2 112

and we obtain Re 3D ( ) = Re 2D ( ) = where n0 e2 3 3 m2 b n0 e 2 3 m2 b

2

q 2 |Vq |2 S (q, LO ), q 2 |Vq |2 S (q, LO ),

(5.48a) (5.48b)

is the density of charge carriers.

N/V in 3D, n0 = N/A in 2D

For an isotropic medium, the dynamic structure factor does not depend on the direction of q, so that S (q, ) = S (q, ), where q = |q|. Let us simplify the expression (5.48a) using explicitly the amplitudes of the Fr ohlich electron-phonon interaction. The modulus squared of the Fr ohlich electron-phonon interaction amplitude is given by 1/2 ( LO )2 4 in 3D q2 V 2mb LO 2 | Vq | = 1/2 ( LO )2 2 in 2D, q A 2mb LO

(5.49)

where is the (dimensionless) Fr ohlich coupling constant determining the coupling strength between the charge carriers and the longitudinal optical phonons [83, 84]. In 3D and 2D, respectively, the sums over q is transformed to the integrals as follows: 3D:
q

... = ... =
q

V (2 )3 A (2 )2

dq . . . dq . . .

2D:

n0 e2 V Re 3D ( ) = 3 3 3 m2 b (2 )

dqq

LO iq

4 V

1/2

1/4 2

2mb LO
0

S (q, LO )

2 n0 e2 2 LO Re 3D ( ) = 3 m2 2mb LO b 3 In the same way, we transform Re 2D ( ):

1/2

q 2 S (q, LO )dq.

(5.51)

Re 2D ( ) =

2 n0 e2 LO 3 m2 b 2

1/2

2mb LO 113

q 2 S (q, LO )dq.

Using the Feynman units ( = 1, mb = 1, LO = 1), Re ( ) is 2 2 1 Re 3D ( ) = n0 e q 2 S (q, 1)dq, 3 3 0 1 2 q 2 S (q, 1)dq. Re 2D ( ) = n0 e 3 2 2 0 From these expressions, it is clear that the scaling relation Re 2D (, ) = Re 3D (, 3 ) 4

(5.52) (5.53)

(5.54)

which holds for the one-polaron case introduced in ref. [83, 84], is also valid for the manypolaron case if the corresponding 2D or 3D dynamic structure factor is used.

1.

Calculation of the dynamic structure factor using the retarded Greens functions

The dynamic structure factor S (q, ) is expressed through the two-point correlation function by Eq. (5.43). The correlation function F (q, ) can be found using the retarded Greens function of the density operators GR (q, t) = i (t) el
(0)

it

0 H

q e

it

0 H

, q

el

(0)

(5.55)

where (t) is the step function. Let us consider the more general case of a nite temperature, GR (q, t) = i (t) where the average is ... T r e H0 . . . Tr
0 e H

it

0 H

q e

it

0 H

, q

(5.56)

, =

1 . kB T

(5.57)

The Fourier image GR (q, ) of the retarded Greens function GR (q, t) is GR (q, ) =

GR (q, t) eit dt
0

= i = i

it

0 H

q e

it

0 H

, q

eit dt
it

it

0 H

q e

it

0 H

q q e

0 H

q e

it

0 H

eit dt

114

The imaginary part of GR (q, ) then is Im GR (q, ) = Re

0

it

0 H

q e

it

0 H

q q e

it

0 H

q e

it

0 H

eit dt

it it it it 1 = e H0 q e H0 q q e H0 q e H0 eit dt 2 0 it it it it 1 q e H0 q e H0 e H0 q e H0 q eit dt 2 0 it it 1 it H 1 0 0 it H it e q e H0 q e H0 eit dt. = q e q e dt + 2 2

In the second integral here, we replace t by (t + i ):

T r e H0 q e
i

it

0 H

q e
it

it

0 H

0 T r e H

eit dt
0 H 0 H 0 e H

T r e H0 q e
i

q e
it

it

0 + H 0 H

Tr Tr e
it

ei(t +i q

dt

= e

0 H 0 H

q e

0 H

Tr

0 e H
it

eit dt . q tends to zero at |t| ),

As far as the integral over t converges (i. e., e

0 H

q e

it

0 H

we can shift the integration contour to the real axis, what gives us the result

T r e H0 q e Tr

it

0 H

q e

it

0 H

0 e H
it

eit dt
it

=e

Tr e

0 H 0 H

q e

0 H

Tr

0 e H

eit dt,

q e

it

0 H

q e e
it

it

0 H

eit dt
it

= e Herefrom, we nd that

0 H

q e

0 H

q eit dt.

Im GR (q, ) =

1 1 e 2

it

0 H

q e

it

0 H

q eit dt

Im GR (q, ) =

1 1 e 2

F (q, ) .

115

So, the equation follows from the analytical properties of the Greens functions: F (q, ) =
1

2 Im GR (q, ) . 1 e

(5.58)

The formula (5.58) is related to arbitrary temperatures. In the zero-temperature limit ( ), the factor 1 e leads to the formula F (q, )|T =0 = 2 ( ) Im GR (q, ) 1 ( ) Im GR (q, ) N
T =0

(5.58) turns into the Heavicide step function ( ), what

(5.59)

S (q, )|T =0 = following equation:

T =0

(5.60)

The retarded Green function is related to the dielectric function of the electron gas by the GR (q, ) = 1 1 1 . vq (q, )

(5.61)

Within the random phase approximation (RPA), following [70], the expression for GR (q, ) is GR (q, ) = [1 vq P (q, )]1 P (q, ) , where the polarization function P (q, ) is (see, e. g., p. 434 of [70]) P (q, ) = 1
k,

(5.62)

k2 2mb

fk+q, fk, +
(k+q)2 2mb

+ i

, +0

(5.63)

with the Fermi distribution function of electrons fk, . For a nite temperature, the explicit analytic expression for the imaginary part of the structure factor P3D (q, ) is obtained (see [70]), 1 + exp E (+) (q, ) V m2 b ln , Im P3D (q, ) = 2 4 q 1 + exp { [ E () (q, )]} E () (q, )
2 q2 2mb

q 4 2m b

2 2

(5.64)

with the chemical potential . The real part of the structure factor is obtained using the Kramers-Kronig dispersion relation: Re P (q, ) = 1

Im P (q, ) d .

(5.65)

116

Analytical expressions for both real and imaginary parts of P (q, ) can be written down for the zero temperature (see [70]), kF q q2 2 2 m 2mb mb q2 2 ln kF 2 qb2 2m 2 kF q q b 2m + m b b V mb 2 k q q F 2 2 + + 2 Re P3D (q, ) = 2 2 , m 2mb mb q 2 ln + kF 2 qb2 + 2m 4 q kF q q2 b + 2m m b b +2kF q 2 2 m2 m2 q2 q2 2 2 b k kF 2 qb2 2m 2 2 F q 2mb b V mb Im P3D (q, ) = 2 2 m2 m2 4 2 q q2 q2 2 2 b 2 qb2 + 2m + k kF 2 q2 F 2mb b where kF = (3 2 N/V )
1/3

(5.66)

is the Fermi wave number.

After substituting into Eq. (5.58) the retarded Greens function (5.62) in terms of the polarization function we arrive at the formula F (q, ) = 2 1 e S (q, ) = S (q, )|T =0 Im P (q, ) , N (1 e ) [1 vq Re P (q, )]2 + [vq Im P (q, )]2 Im P (q, ) . = ( ) N [1 vq Re P (q, )]2 + [vq Im P (q, )]2 (5.67) (5.68)

Im

P (q, ) 1 vq P (q, )

With this dynamic structure factor, the optical conductivity (5.52) (in the Feynman units) takes the form Re 3D ( ) = e 2 1 ( 1) 3V 3 Im P3D (q, 1) 2 2 2 q dq. [1 vq Re P3D (q, 1)] + [vq Im P3D (q, 1)] 0
2

(5.69)

Correspondingly, in the 2D case we obtain the expression 1 ( 1) Re 2D ( ) = e2 2 2V 3 Im P2D (q, 1) 2 2 2 q dq. [1 vq Re P2D (q, 1)] + [vq Im P2D (q, 1)] 0 117

(5.70)

2.

Plasmon-phonon contribution

The RPA dynamic structure factor for the electron (or hole) system can be separated in two parts, one related to continuum excitations of the electrons (or holes) Scont , and one related to the undamped plasmon branch: SRPA (q, ) = Apl (q ) ( pl (q )) + Scont (q, ), (5.71)

where pl (q ) is the wave number dependent plasmon frequency and Apl is the strength of the undamped plasmon branch. In Eqs. (5.69), (5.70), the contribution of the continuum excitations corresponds to the region (q, ) where Im P (q, ) = 0. The contribution related to the undamped plasmons is provided by a region of (q, ) , where the equations Im P (q, ) = 0 . 1 vq Re P (q, ) = 0 Im

(5.72)

are fullled simultaneously. Using (5.61), we nd that Eqs. (5.73) are equivalent to 1 1 = 0, Re = 0. (q, ) (q, )
Im P (q, ) [1vq Re P (q, )]2 +[vq Im P (q, )]2

(5.73) is propor-

In the region where Im P (q, ) = 0, the expression

tional to the delta function, which gives a nite contribution to the memory function after the integration over q : Im P (q, ) [1 vq Re P (q, )]2 + [vq Im P (q, )]2

Im P (q, )=0

1 (1 vq Re P (q, )) . = vq

(5.74)

Using Eq. (5.74), the coecients Apl (q ) in Eq. (5.71) can be expressed in terms of the polarization function P (q, ) as follows: Im P (q, ) [1 vq Re P (q, )]2 + [vq Im P (q, )]2 =
2 vq

Im P (q, )=0

1 Re P (q, )

=pl (q )

( pl (q ))

118

Apl (q ) =

2 vq

1 Re P (q, )

.
=pl (q )

(5.75)

The contribution derived from the undamped plasmon branch Apl (q ) ( pl (q )) is denoted in Ref. [51] as the plasmon-phonon contribution. The physical process related to this contribution is the emission of both a phonon and a plasmon in the scattering process.

E.

Comparison to the infrared spectrum of Nd2x Cex CuO2y

Calvani and collaborators have performed doping-dependent measurements of the infrared absorption spectra of the high-Tc material Nd2x Cex CuO2y (NCCO). The region of the spectrum examined by these authors (50-10000 cm1 ) is very rich in absorption features: they observe is a Drude-like component at the lowest frequencies, and a set of sharp absorption peaks related to phonons and infrared active modes (up to about 1000 cm1 ) possibly associated to small (Holstein) polarons. Three distinct absorption bands can be distinguished: the d -band (around 1000 cm1 ), the Mid-Infrared band (MIR, around 5000 cm1 ) and the Charge-Transfer band (around 104 cm1 ). Of all these features, the d band and, at a higher temperatures, the Drude-like component have (hypothetically) been associated with large polaron optical absorption [90]. For the lowest levels of Ce doping, the d -band can be most clearly distinguished from the other features. The experimental optical absorption spectrum (up to 3000 cm1 ) of Nd2 CuO2 ( < 0.004), obtained by Calvani and co-workers [90], is shown in Fig. 12 (shaded area) together with the theoretical curve obtained by the present method (full, bold curve) and, for reference, the one-polaron optical absorption result (dotted curve). At lower frequencies (600-1000 cm1 ) a marked dierence between the single polaron optical absorption and the many-polaron result is manifest. The experimental d -band can be clearly identied, rising in intensity at about 600 cm1 , peaking around 1000 cm1 , and then decreasing in intensity above that frequency. At a density of n0 = 1.5 1017 cm3 , we found a remarkable agreement between our theoretical predictions and the experimental curve.

119

FIG. 12: The infrared absorption of Nd2 CuO2 ( < 0.004) is shown as a function of frequency, up to 3000 cm1 . The experimental results of Calvani and co-workers [90] is represented by the thin black curve and by the shaded area. The so-called d-band rises in intensity around 600 cm1 and increases in intensity up to a maximum around 1000 cm1 . The dotted curve shows the single polaron result. The full black curve represents the theoretical results obtained in the present work for the interacting many-polaron gas with n0 = 1.5 1017 cm3 , = 2.1 and mb = 0.5 me . (From Ref. [51].)

F.

Experimental data on the optical absorption in manganites: interpretation in

terms of a many-polaron response

In Refs. [91, 92], the experimental results on the optical spectroscopy of La2/3 Sr1/3 MnO3 (LSMO) and La2/3 Ca1/3 MnO3 (LCMO) thin lms in the mid-infrared frequency region are presented. The optical conductivity spectra of LCMO lms are interpreted in [91, 92] in terms of the optical response of small polarons, while the optical conductivity spectra of LSMO lms are explained using the large-polaron picture (see Fig. 13). The real part of the optical conductivity Re ( ) is expressed in Ref. [92] by the formula 2 e2 ( 0 )2 Re ( ) = np 3 m2 3

2m0

q 2 S (H) (q, 0 )dq,

(5.76a)

where is the electron-phonon coupling constant, np is the polaron density, m is the electron band mass, 0 is the LO-phonon frequency, and S (H) (q, ) is the dynamic structure factor,

120

FIG. 13: Comparison of the low-temperature MIR optical conductivity to ( ) from various model calculations: the solid line refers to the weak-coupling approach of Tempere and Devreese [51] modied for an on-site Hubbard interaction, the dashed line is the result of the phenomenological approach for self-trapped large polarons by Emin [93], the dotted curve is the weak-coupling singlepolaron result [69]. (From Ref. [91].)

determined in Ref. [92] through the dielectric function of an electron gas (q, ): S (H) (q, ) = 1 q2 . Im 2 np 4e (q, ) (5.77)

In Ref. [51], the other denition for the dynamic structure factor is used, which is equivalent to that given by (5.77) (see Ref. [70]) with the factor N (the number of electrons) S
(TD)

1 (q, ) = 2

el |q (t) q (0)| el eit dt = NS (H) (q, ).

(5.78)

Here, el denotes the ground state of the electron subsystem (without the electron-phonon interaction), q =
N iqrj j =1 e

is the Fourier component of the electron density.

In Ref. [51], the dynamic structure factor is calculated within the random-phase approximation (RPA) taking into account the Coulomb interaction between electrons with the Fourier component of the Coulomb potential vq = 4e2 , q 2 (5.79)

where is the high-frequency dielectric constant of the crystal. In Refs. [91, 92], the dynamic structure factor is calculated taking into account the local Hubbard electron-electron 121

interaction instead of the Coulomb interaction. The local Hubbard interaction is used in the small-polaron formalism and describes the potential energy of two electrons on one and the same site (see, e. g., Refs. [94, 95]). In its simplest form the Hubbard interaction is (see Eq. (1) of Ref. [94]) VH = U
i

ni ni ,

(5.80)

where U is the coupling constant of the Hubbard interaction, ni is the electron occupation number for the i-th site. In Refs. [91, 92], there are no details of the calculation using the interaction term (5.80). The following procedure can be supposed. The transition from the summation over the lattice sites to the integral over the crystal volume V is performed taking into account the normalization condition N() =
i

ni() =

n () (r) dr,

where the density n () (r) is to be determined through ni() . As far as the lattice cell volume 0 V , the integral
V

n() (r) dr can be written as the sum over the lattice sites:
V

n() (r) dr = 0

n () (ri ) ,

where {ri } are the vectors of the lattice. Therefore, from the equality
i

ni() = 0

n () (ri )

we nd that ni() = n () (ri ) . The potential (5.80) is then transformed from the sum over sites to the integral: VH = U 0 ni ni 0 = U 0 (r r ) n (r) n (r ) d r d r . (5.81)

Consequently, in the continuum approach the Hubbard model is described by the -like interparticle potential U 0 (r r ). This development of the approach [51] performed in Refs. [91, 92] seems to be contradictory by the following reason. For a many-polaron system, both the electron-phonon and electron-electron interactions are provided by the electrostatic potentials. Therefore, it would be consistent to consider them both within one and the same approach. Namely, 122

the Coulomb electron-electron interaction with the potential (5.79) is relevant for large and small polarons with the Fr ohlich electron-phonon interaction, while the Hubbard electronelectron interaction is relevant for small Holstein polarons. Nevertheless, as recognized in Ref. [91], this model reproduces the observed shape of the polaron peak quite convincingly and provides a better agreement with the experiment [91, 92] than the phenomenological approach [93] and the one-polaron theory [69].

123

VI. A.

INTERACTING POLARONS IN A QUANTUM DOT The partition function and the free energy of a many-polaron system

We consider a system of N electrons with mutual Coulomb repulsion and interacting with the lattice vibrations following Ref. [96]. The system is assumed to be conned by a parabolic potential characterized by the frequency parameter 0 . The total number of electrons is represented as N =

N , where N is the number of electrons with the

spin projection = 1/2. The electron 3D (2D) coordinates are denoted by xj, with j = 1, , N . The bulk phonons (characterized by 3D wave vectors k and frequencies k ) are described by the complex coordinates Qk , which possess the property [41] Q k = Qk . (6.1)

{Qk } . The full set of the electron and phonon coordinates are denoted by x {xj, } and Q Throughout the present treatment, the Euclidean time variable = it is used, where t is the real time variable. In this representation the Lagrangian of the system is ;x Q = Le x + Leph x , ,Q ,x L x , Q V C (x ) Ub (x ) + Lph Q, , Q (6.2)

,x where Le x is the Lagrangian of an electron with band mass mb in a quantum dot:

N

,x Le x =

=1/2 j =1

mb 2 2 x j, + 2 0 xj, , 2

dx , d

(6.3)

0 is the connement frequency, Vb (x ) is the potential of a background charge (supposed to be static and uniformly distributed with the charge density enb in a sphere of a radius R),
N

Ub (x ) =
j =1 2

Vb (rj, ) , r |x| ,

(6.4) (6.5)

V b (r ) =

3R 2 r 2 R3 4e nb (r < R ) , + (R r ) 3 0 2 r

where 0 is the static dielectric constant of a crystal, VC (x ) is the potential energy of the electron-electron Coulomb repulsion in the medium with the high-frequency dielectric constant : V C (x ) =
, =1/2 j =1 l=1 (j,)=(l, ) N N

1 e2 , 2 |xj, xl, |

(6.6)

124

 Q ; Q, Q is the Lagrangian of free phonons: Lph Q, Q = 1 Lph Q, 2

2 (Q k Qk + k Qk Qk ), k

dQ . Q d

(6.7)

Q is the Lagrangian of the electron-phonon interaction: Further, Leph x , Q, Leph = x , Q 2k

k 1/2

Vk Qk k ,

(6.8)

where k is the Fourier transform of the electron density operator:

N

k =
=1/2 j =1

eikxj, .

(6.9)

Vk is the amplitude of the electron-phonon interaction. Here, we consider electrons interacting with the long-wavelength longitudinal optical (LO) phonons with a dispersionless frequency k = LO , for which the amplitude Vk is [24] LO Vk = q 2 2 V
1/2 1/4

mb LO

(6.10)

where is the electron-phonon coupling constant and V is the volume of the crystal. We consider a canonical ensemble, where the numbers N are xed. The partition function Z ({N } , ) of the system can be expressed as a path integral over the electron and phonon coordinates: Z ({N } , ) = (1)P N1/2 !N1/2 !
Px Q Q

dx
x

Dx ( )

dQ

( ),Q( )] ( ) eS [x , DQ

(6.11)

( ) is the action functional: where S x ( ) , Q ( ) = 1 S x ( ) , Q

0

;x d. ,Q L x , Q

(6.12)

The parameter 1/ (kB T ) is inversely proportional to the temperature T . In order to take the Fermi-Dirac statistics into account, the integral over the electron paths {x ( ) } in Eq. (6.11) contains a sum over all permutations P of the electrons with the same spin projection, and P denotes the parity of a permutation P . Therefore, the path The action functional (6.12) is quadratic in the phonon coordinates Q. integral over the phonon variables in Z ({N } , ) can be calculated analytically following 125

Ref. [41]. Let us describe this path integration in detail. First, we introduce the real phonon coordinates through the real and imaginary parts of the complex phonon coordinates Qk Re Qk , Q k Im Qk . According to the symmetry property (6.1), they obey the equalities
Qk = Qk , Q k = Qk . 2Q , k 0, x k qk 2Q , kx < 0. k

(6.13) (6.14) |Qk |2 is transformed in the

In this representation, the sum over phonon coordinates following way using the symmetry property (6.13): |Qk |2 = =2
k (kx 0)

(Qk ) + (Q k)
k

(Qk ) + 2
k (kx <0) 2 qk = k (kx <0) 2 qk +

(Q k)

=
k (kx 0)

2 qk . k

Therefore, the phonon Lagrangian (6.7) with the real phonon coordinates is Lph = 1 2
2 2 2 (q k + k qk ). k

(6.15)

The Lagrangian of the electron-phonon interaction (6.8) with the real phonon coordinates is transformed in the following way using (6.13): Leph = = +i
k (kx <0)

2k
k

1/2

Vk k (Qk iQ k)
1/2

2k
k (kx 0)

(Vk k + Vk k ) Qk
1/2

2k

(Vk k Vk k ) Q k.

Let us introduce the real forces: 1 2 k 1

2k 1/2 2k 1/2

(Vk k + Vk k ) , kx 0, (Vk k Vk k ) , kx > 0. 126

(6.16)

i 2

This gives us the Lagrangian of the electron-phonon interaction in terms of the real forces and real phonon coordinates: Leph = k qk .
k

(6.17)

So, the sum of the Lagrangians of phonons and of the electron-phonon interaction is expressed through ordinary real oscillator variables: Lph + Leph = 1 2
2 2 2 (q k + k qk + k qk ). k

(6.18)

The path integration for each phonon mode with the coordinate qk is performed independently as described in Sec. 2 of Ref. [41] and gives the result
qk

dqk
qk

Dqk ( ) exp

1
0

1 2 2 2 d (q k + k qk + k qk ) 2

1
k 2

2 sinh 1 4

exp

d
0 0

cosh [k (| | /2)] k ( ) k ( ) , k sinh ( k /2)

where the exponential is the inuence functional of a driven oscillator { [41], Eq. (3.43)}. Therefore, the path integral over all phonon modes is d {qk } =
k {qk } { qk }

D {qk ( )} exp
k 2

1
0

d
k

1 2 2 2 (q + k qk + k qk ) 2 k

1 2 sinh 1 4
k

exp Here, the product

d
0 0

d
k

cosh [k (| | /2)] k ( ) k ( ) . k sinh ( k /2)

. . . is the partition function of free phonons, and the exponential is

the inuence functional of the phonon subsystem on the electron subsystem. This inuence functional results from the above described elimination of the phonon coordinates and is usually written down as e , where is = 1 4

d
0 0

d
k

cosh [k (| | /2)] k ( ) k ( ) . k sinh ( k /2)

(6.19)

127

The sum over the phonon wave vectors k can be simplied as follows: cosh [k (| | /2)] k ( ) k ( ) k sinh ( k /2) 1
k (kx 0)

cosh [k (| | /2)] sinh ( k /2)

[Vk k ( ) + Vk k ( )] [Vk k ( ) + Vk k ( )] 1
k (kx <0)

cosh [k (| | /2)] sinh ( k /2)

[Vk k ( ) Vk k ( )] [Vk k ( ) Vk k ( )] = 2
k (kx 0)

cosh [k (| | /2)] V k V k sinh ( k /2)

[k ( ) k ( ) + k ( ) k ( )] = 2
k

cosh [k (| | /2)] |Vk |2 k ( ) k ( ) . sinh ( k /2)

Herefrom, we nd that = | Vk | 2 2 2

d
0 0

cosh k | | sinh
k 2

k ( ) k ( ) .

(6.20)

As a result, the partition function of the electron-phonon system (6.11) factorizes into a product Z ({N } , ) = Zp ({N } , ) 1 2 sinh ( k /2) (6.21)

of the partition function of free phonons with a partition function Zp ({N } , ) of interacting polarons, which is a path integral over the electron coordinates only: Zp ({N } , ) = The functional S p [x ( )] = 1
0

(1)P N1/2 !N1/2 !

Px

dx
x

( )] Dx ( ) eSp [x .

(6.22)

( ) , x Le x ( ) V C (x ( )) d + [x ( )]

(6.23)

describes the phonon-induced retarded interaction between the electrons, including the retarded self-interaction of each electron. 128

Using (6.3) and (6.6) we write down Sp [x ( )] explicitly: Sp [ x ( )] = 1

N 0

|Vq |2 2 2

j =1

mb 2 2 x j, + 2 0 xj, + 2 ,

N N

j =1 l=1 (j,)=(l, )

d
0 0

cosh [LO (| | /2)] q ( ) q ( ) . sinh ( LO /2)

d 2 |xj, xl, | (6.24)

e2

The free energy of a system of interacting polarons Fp ({N } , ) is related to their partition function (6.22) by the equation: 1 Fp ({N } , ) = ln Zp ({N } , ) . (6.25)

At present no method is known to calculate the non-gaussian path integral (6.22) analytically. For distinguishable particles, the Jensen-Feynman variational principle [41] provides a convenient approximation technique. It yields a lower bound to the partition function, and hence an upper bound to the free energy. It can be shown [97] that the path-integral approach to the many-body problem for a xed number of identical particles can be formulated as a Feynman-Kac functional on a state space for N indistinguishable particles, by imposing an ordering on the conguration space and by the introduction of a set of boundary conditions at the boundaries of this state space. The resulting variational inequality for identical particles takes the same form as the Jensen-Feynman variational principle: Fp Fvar , 1 Sp S0
S0

(6.26) , (6.27)

Fvar = F0 +

where S0 is a model action with the corresponding free energy F0 . The angular brackets mean a weighted average over the paths ( ) =
(1)P P N1/2 !N1/2 ! (1)P P N1/2 !N1/2 !

dx

Px x

S0

( )] Dx ( ) () eS0 [x ( )] Dx ( ) eS0 [x

dx

Px x

(6.28)

In the zero-temperature limit, the polaron ground-state energy

0 Ep = lim Fp

(6.29)

129

obeys the inequality following from (6.26) with (6.27):

0 Ep

Evar

with
0 Evar = E0 + lim 0 E0 = lim F0 .

1 Sp S0

S0

(6.30) (6.31)

B.

Model system

We consider a model system consisting of N electrons with coordinates x {xj, } and Nf ctitious particles with coordinates y {yj } in a harmonic connement potential with elastic interparticle interactions as studied in Refs. [96, 98]. The Lagrangian of this model system takes the form LM mb ,y ;x x , y = 2 mf 2
Nf 2 y j j =1 N N N , j =1 l=1

x 2 j,
j =1

2 x2 j, k 2

mb 2 + 4
N Nf

(xj, xl, )2 (6.32)

2 2 f yj

j =1 l=1

(xj, yl )2 .

The frequencies , , f , the mass of a ctitious particle mf , and the force constant k are variational parameters. Clearly, this Lagrangian is symmetric with respect to electron permutations. Performing the path integral over the coordinates of the ctitious particles in the same way as described above for phonons, the partition function Z0 ({N } , ) of the model system of interacting polarons becomes a path integral over the electron coordinates: Z0 ({N } , ) = (1)P N1/2 !N1/2 !
Px ( )] Dx ( ) eS0 [x , x

dx

(6.33)

with the action functional S0 [x ( )] given by S 0 [x ( )] = 1

0 0 N

j =1

mb 2 x j, ( ) + 2 x2 j, ( ) d 2 mb 2 [xj, ( ) xl, ( )]2 d 4 d

0 cosh [f

N N , j =1 l=1

k 2 N 2 Nf 4mf f

d
0

(| | /2)] X ( ) X ( ) , sinh ( f /2)

(6.34)

130

where X is the center-of-mass coordinate of the electrons, 1 X= N

1.
N

xj, .
j =1

(6.35)

Analytical calculation of the model partition function

The partition function Z0 ({N } , ) [Eq. (6.33)] for the model system of interacting polarons can be expressed in terms of the partition function ZM ({N } , Nf , ) of the model system of interacting electrons and ctitious particles with the Lagrangian LM [Eq. (6.32)] as follows: Z0 ({N } , ) = ZM ({N } , Nf , ) , Zf (Nf , wf , ) (6.36)

where Zf (Nf , wf , ) is the partition function of ctitious particles, Zf (Nf , ) = with the frequency wf = 2 f + kN/mf (6.38) 1 2 sinh 1 wf 2
DNf

(6.37)

and D=3(2) for 3D(2D) systems. The partition function ZM ({N } , Nf , ) is the path integral for both the electrons and the ctitious particles: ZM ({N } , Nf , ) = (1)P N1/2 !N1/2 !
Px y

dx
x

Dx ( )

dy
y

( ),y ( )] Dy ( ) eSM [x

(6.39)

with the action functional S M [x ( ) , y ( )] = 1

0

,y ;x LM x , y d,

(6.40)

where the Lagrangian is given by Eq. (6.32). Let us consider an auxiliary ghost subsystem with the Lagrangian mf Nf 2 2 2 2 2 g, Y g , Xg , Yg = mb N X 2 Lg X Yg + wf Yg g + w Xg 2 2 with two frequencies w and wf , where w is given by w= 2 N 2 + kNf /mb . 131 (6.42) (6.41)

The partition function Zg of this subsystem

Xg Yg

Zg =

dXg

dYg
Xg

D X g ( )
Yg

D Yg ( ) exp {Sg [Xg ( ) , Yg ( )]} ,

(6.43)

with the action functional

Sg [Xg ( ) , Yg ( )] =

1
0

g , Xg , Y g , Yg d Lg X

(6.44)

is calculated in the standard way, because its Lagrangian (6.41) has a simple oscillator form. Consequently, the partition function Zg is Zg = 1 2 sinh
w 2 D

1 2 sinh
wf 2 D

(6.45)

The product Zg ZM of the two partition functions Zg and ZM ({N } , Nf , ) is a path integral in the state space of N electrons, Nf ctitious particles and two ghost particles M of this system is a sum of LM with the coordinate vectors Xg and Yg . The Lagrangian L and Lg , M x g, Y g , Xg , Yg . g, Y g; x , ;x ,y ,X L , y , Xg , Yg LM x y , y + Lg X (6.46)

M can be The ghost subsystem is introduced because the center-of-mass coordinates in L explicitly separated much more transparently than in LM . This separation is realized by the linear transformation of coordinates, x = x + X X , j, g j, yj = y + Y Yg ,
j

(6.47)

where X and Y are the center-of-mass coordinate vectors of the electrons and of the ctitious particles, correspondingly: 1 X= N
N

xj, ,
j =1

1 Y= Nf

Nf

yj .
j =1

(6.48)

Before the transformation (6.47), the independent variables are x , y , Xg , Yg , with the center-of-mass coordinates X and Y determined by Eq. (6.48). When applying the trans-

132

formation (6.47) to the centers of mass (6.48), we nd that 1 X= N Y= 1 Nf

N

xj,
Nf j =1

+ X Xg

1 = N
Nf

j =1

xj, + X Xg ,

(6.49)

j =1

yj + Y Yg =

1 Nf

j =1

yj + Y Yg .

(6.50)

As seen from Eqs. (6.49), (6.50), after the transformation (6.47) the independent variables are (x , y , X, Y) , while the coordinates (Xg , Yg ) obey the equations 1 Xg = N
N

xj, ,
j =1

1 Yg = Nf

Nf yj . j =1

(6.51)

In order to nd the explicit form of the Lagrangian (6.46) after the transformation (6.47), we use the following relations for the quadratic sums of coordinates:
N N Nf 2 xj, Nf 2 yj j =1 N

x2 j,
j =1

=
j =1

+N X
2

X2 g

=
j =1 2

yj

2 + Nf Y 2 Yg ,

N N , j =1 l=1 Nf

(xj, xl, ) = 2N (yj yl ) = 2Nf

2 2

xj,
Nf yj j =1 N 2 j =1

2 2 2Nf Xg ,

Nf

j =1 l=1 N Nf

2 2 2Nf Yg , Nf 2

j =1 l=1

(xj, yl ) = Nf

2 xj,

+N
j =1

yj

+ (6.52)

j =1

2 NNf X2 + Y 2 2X Y X2 g Yg .

The substitution of Eq. (6.48) into Eq. (6.46) then results in the following 3 terms:
M x ,Y ;x , X; Y ,Y , , , X , x , y L y ,y , X, Y = Lw x + Lwf y + LC X

(6.53)

, x where Lw x

, y and Lwf y

are Lagrangians of non-interacting identical oscillators

with the frequencies w and wf , respectively, Lw , x x =

mb 2

x j,
=1/2 j =1 Nf y j, j =1 2

+ w 2 xj,
2

(6.54)

,y Lwf y

mf = 2

2 + wf yj,

(6.55)

133

 , X; Y , Y describes the combined motion of the centers-of-mass of The Lagrangian LC X the electrons and of the ctitious particles,
2 2 2 X2 mf Nf Y , X; Y , Y = mb N X 2+ 2 + wf Y + kNNf X Y , (6.56) LC X 2 2

with = 2 + kNf /mb . (6.57)

The Lagrangian (6.56) is reduced to a diagonal quadratic form in the coordinates and the velocities by a unitary transformation for two interacting oscillators using the following replacement of variables: X= Y= with the coecients a1 = 1+ 2
1/2

1 (a1 r + a2 R) , mb N 1 (a2 r + a1 R) mf Nf
1/2

(6.58)

a2 =

1 2 , 1/2

, k NNf . mb mf

(6.59) (6.60)

2 2 f 2 2 f
2

+ 4 2

As a result, four independent frequencies 1 , 2 , w and wf appear in the problem. Three of them (1 , 2 , w ) are the eigenfrequencies of the model system. 1 is the frequency of the relative motion of the center of mass of the electrons with respect to the center of mass of the ctitious particles; 2 is the frequency related to the center of mass of the model system as a whole; w is the frequency of the relative motion of the electrons with respect to their center of mass. The parameter wf is an analog of the second variational parameter w of the one-polaron Feynman model. Further, the Lagrangian (6.56) takes the form LC = 1 2 1 2 2 2 , R + 2 r + 2 2R 1r 2 2 134

The eigenfrequencies of the center-of-mass subsystem are then given by the expression 2 1 2 2 + 2 2 + 4 2 , = + 1 f f 2 (6.61) 2 2 = 1 2 + 2 2 2 + 4 2 . f f 2

(6.62)

leading to the partition function corresponding to the combined motion of the centers-ofmass of the electrons and of the ctitious particles ZC = 1 2 sinh
1 2 D

1 2 sinh
2 2 D

(6.63)

Taking into account Eqs. (6.45) and (6.63), we obtain nally the partition function of the model system for interacting polarons sinh sinh
w 2 1 2

sinh sinh

Here

Z0 ({N } , ) =

wf 2 2 2

F ({N } , w, ) . Z

(6.64)

F ({N } , w, ) = ZF N1/2 , w, ZF N1/2 , w, Z

(6.65)

is the partition function of N = N1/2 + N1/2 non-interacting fermions in a parabolic connement potential with the frequency w. The analytical expressions for the partition function of N spin-polarized fermions ZF (N , w, ) were derived in Ref. [99].

C.

Variational functional

In order to obtain an upper bound to the free energy Evar , we substitute the model action functional (6.34) into the right-hand side of the variational inequality (6.26) and consider the limit : Evar ({N }) mb 2 0 2 + N 2 = EF ({N } , w ) + 2 mb N 2
2 2

x2 j (0)
j =1 2 S0

X2 (0)

S0

+ Ub (x )

S0

+
q=0

2e [G (q, 0| {N } , ) N ] V q 2
S0

k 2 N 2 Nf + lim 4mf f lim |Vq |2 2 2

0

d
0 0

cosh [f (| | /2)] X ( ) X ( ) sinh ( f /2)

d
0

cosh [LO (| | /2)] G (q, | {N } , ) . sinh ( LO /2)

(6.66)

135

Here, EF (N, w ) is the energy of N non-interacting fermions in a parabolic connement potential with the connement frequency w , EF ({N } , w ) = w
L 1 =1/2 n=0

n+ 3 2

3 2 ,

g (n) (6.67)

+ (N NL ) L +

where is the spin of an electron, L is the lower partly lled or empty level for N electrons with the spin projection . The rst term in the curly brackets of Eq. (6.94) (the upper line) is the number of electrons at fully lled energy levels, while the second term (square brackets) is the number of electrons at the next upper level (which can be empty or lled partially). The energy levels of a 3D oscillator are degenerate, so that g (n) = 1 (n + 1) (n + 2) 2 (6.68)

is the degeneracy of the n-th energy level. The parameter 1 NL = L (L + 1) (L + 2) 6 (6.69)

is the number of electrons at all fully lled levels. The summation in Eq. (6.67) is performed explicitly, what gives us the result EF ({N } , w ) = w 3 1 L (L + 1)2 (L + 2) + (N NL ) L + 8 2 . (6.70)

In Eq. (6.66), G (q, | {N } , ) is the two-point correlation function for the electron density operators: G (q, | {N } , ) = q ( ) q (0) The averages X ( ) X ( ) Xk ( ) Xk ( )
S0 S0 S0

(6.71)

are calculated using the generating function method: 2 exp [i ( X ( ) + X ( ))] k k 3 = 2mN
2 S0

=0, =0

(6.72)

= X ( ) X ( )

S0

i=1

a2 i cosh [i (| | /2)] . i sinh ( i /2)

(6.73)

136

Substituting this expression into Eq. (6.66) and performing integrations over and analytically, we obtain the result k 2 N 2 Nf 4mf f = 3 4
2

d
0 0

cosh [f (| | /2)] X ( ) X ( ) sinh ( f /2) coth ( i /2) coth ( f /2) , i f

S0

a2 i

2 f i=1

2 i

(6.74)

and in the zero-temperature limit we have k 2 N 2 Nf lim 4mf f 3 = 4 The average

j =1 2

d
0 0

cosh [f (| | /2)] X ( ) X ( ) sinh ( f /2) .

S0

i=1 N

a2 i 2 2 i f x2 j
S0

1 1 i f

(6.75)

is transformed, using the described above operations with the

ghost subsystem,

xj = xj + X Xg , and taking into account the rst of equations (6.52)

N N

(6.76)

x2 j
j =1

=
j =1

xj

+ N X2 X2 g .

(6.77)

Consequently, averaging the left-hand side of Eq. (6.77) on the model action functional S0 , one obtains
N N N

x2 j
j =1 S0

=
j =1 N

x2 j
SM

=
j =1

x2 j
SM + Sg

=
j =1 N

x2 j
Sw

+N

X2

SC

X2 g

Sg

(6.78)

The term
j =1

x2 j
Sw

is expressed using the virial theorem through the ground-state energy

EF (N, w ) of N independent 3D fermion oscillators with the frequency w and with the mass mb ,
N

x2 j
j =1 Sw

1 EF (N, w ) = ln ZI (N ) , 2 mb w mb w 2 137

(6.79)

Two other terms in Eq. (6.78) are [cf. Eq. (6.73)]: X

2 SC

3 = 2mb N =

i=1

a2 i coth ( i /2) , i (6.80)

X2 g So, we obtain
N

Sg

3 coth (w/2) . 2mb N w

2

x2 j
j =1 S0

3 EF ({N } , w ) + = 2 mb w 2mb

i=1

a2 1 i i w

(6.81)

The averaging of the background-charge potential gives us the result n (1)k n + 2 3 2 f1 (n, |, N )| Ub (x ) S0 = nk (1 ) n=0 k! k =0 k1 2 Ak1/2
1 F1

1 2w

1 1 R2 k ; ; 2 2 4A
2

1 3 R2 1 F1 k ; ; 2 2 4A ,

(6.82)

/0 , A

4mb N

i=1

a2 N 1 i + i w

where f1 (n, |, N ) is the one-particle distribution function of fermions (the distribution functions are considered in more details in the next subsection). Collecting all terms together, we arrive at the variational functional Evar (1 , 2 , w, f ) =
2 (w, N ) 3 E 3 2 0 +w w + (1 + 2 f ) 2 2w 2 2 2

3 2 2 2 + 2 1 2 + f 4 0 + Ub (x )

i=1 S0

3 2 a2 i + i 4f

i=1

a2 i i (i + f ) (6.83)

+ EC + Eeph ,

where EC and Eeph are the Coulomb and polaron contributions, respectively: EC = Eeph e2 4 2 dq 1 G (q, 0| {N } , )| N , q2 d exp (LO ) G (q, | {N } , )| .
0

(6.84) (6.85)

2 = 2 4

1 dq 2 q

The correlation function (6.71) is calculated analytically in the next subsection. With this correlation function, the variational ground-state energy is calculated and minimized numerically. 138

D.

Two-point correlation functions

The two-point correlation function (6.71) is represented as the following path integral: G (q, | {N } , ) = 1 Z0 ({N } , )
Px

(1)P N1/2 !N1/2 ! (6.86)

dx
x

( )] Dx ( ) eS0 [x q ( ) q (0) .

We observe that G (q, | {N } , ) can be rewritten as an average within the model action S M [x ( ) , y ( )] of interacting electrons and ctitious particles: 1 G (q, | {N } , ) = ZM ({N } , Nf , )
Px

(1)P N1/2 !N1/2 !

y ( ),y ( )] Dy ( ) eSM [x y

dx
x

Dx ( )

dy

q ( ) q (0) . to (6.86).

(6.87)

Indeed, one readily derives that the elimination of the ctitious particles in (6.87) leads The representation (6.87) allows one to calculate the correlation function G (q, | {N } , ) in a much simpler way than through Eq. (6.86), using the separation of the coordinates of the centers of mass of the electrons and of the ctitious particles. This separation is performed for the two-point correlation function (6.87) by the same method as it has been done for the partition function (6.39). As a result, one obtains G (q, | {N } , ) = g (q, | {N } , ) exp [iq (X ( ) X ( ))] SC , exp [iq (Xg ( ) Xg ( ))] Sg (6.88)

where g (q, | {N } , ) is the time-dependent correlation function of N non-interacting electrons in a parabolic connement potential with the frequency w , g (q, | {N } , ) = q ( ) q (0)
Sw

(6.89)

,x The action functional Sw [x ] is related to the Lagrangian Lw x [Eq. (6.54)] S w [x ] = 1

0

,x Lw x d.

(6.90)

The averages in (6.88) are calculated using Feynmans method of generating functions [41]. Namely, according to [41], the average i G [f ( )] exp

f ( ) x d
0

,
S

(6.91)

139

where S is the action functional of a one-dimensional harmonic oscillator with the frequency and with the mass m, results in 1 d G [f ( )] = exp 4m
0

d
0

The diagonalization procedure for the Lagrangian LC (6.56) allows us to represent that

cosh [ (| | /2)] f ( ) f ( ) . sinh (/2)

(6.92)

Lagrangian as a sum of Lagrangians of independent harmonic oscillators, what gives the following explicit expressions for averages in Eq. (6.88): exp [iq (X ( ) X ( ))] SC 2 2 sinh q = exp a2 i Nmb i=1

i | | 2

sinh

i ( | |) 2

i sinh

i 2

exp [iq (Xg ( ) Xg ( ))] Sg w | | 2 sinh sinh w( 2 q = exp Nmb w sinh 2w

| |) 2

1.

The correlation function g (q, | {N } , )

As seen from the formula (6.89), g (q, | {N } , ) is the time-dependent correlation function of N non-interacting fermions in a parabolic connement potential with the frequency w . Let us consider rst of all a system of N identical spin-polarized oscillators with the Lagrangian m L= 2 The corresponding Hamiltonian is
N N 2 2 x 2 j xj .

(6.93)

j =1

= H
j =1 N

p 2 m 2 x 2 j j + 2m 2

(6.94)

= H
j =1

j , h

m 2 x 2 2 p + . h 2m 2

(6.95)

is determined as follows: A set of eigenfunctions of the one-particle Hamiltonian h n (x) = n n (x) , h 140 (6.96)

where n (n1 , n2 , n3 ) , n = n = n + n n1 + n2 + n3 , 3 , 2 n (x) = n1 (x1 ) n2 (x2 ) n3 (x3 ) , (6.97)

n (x) is the n-th eigenfunction of a one-dimensional oscillator with the frequency . The Hamiltonian (6.94) can be written down in terms of the annihilation ( an ) and creation ( a+ n ) operators: = H
n

n a + n = na
n

n , n N

n a N + n . na

(6.98)

The many-particle quantum states in the representation of occupation numbers are written down as |. . . Nn . . . , where Nn is the number of particles in the n-th one-particle quantum state. The states |. . . Nn . . . are dened as the eigenstates of the operator of the n : number of particles in the n-th state N n |. . . Nn . . . = Nn |. . . Nn . . . . N Let us determine a set of quantum states with a nite total number of particles Nn = N
n

(6.99)

(6.100)

as follows: |. . . Nn . . . |
n

Nn = N

N,{Nn } .

(6.101)

Further on, we use the basis set of quantum states (6.101) for the derivation of the partition function, of the density function and of the two-point correlation function. Partition function The density matrix of the canonical Hibbs ensemble is , = exp H quantum states (6.101): ZI ( |N ) = N,{Nn } exp H exp n Nn
n

1 . kB T

The partition function of this ensemble is the trace of the density matrix on the set of

{Nn }

N,{Nn } .
Nn = N

(6.102)

{Nn }

141

This expression can be written down also in the form ZI ( |N ) = where j,k = is the delta symbol. exp 1, n Nn
n

N,

Nn ,

(6.103)

{ Nn }

j=k j=k

0,

Let us introduce the generating function for the partition function in the same way as in Ref. [99]: (, u) =
N =0

uN ZI ( |N ) = exp exp

N =0

uN
{Nn }

exp
n

n Nn
n

N,

Nn

=
{ Nn }

n Nn
n

u
Nn

Nn

N,

Nn

N =0

=
{ Nn }

n Nn
n

(, u) =
n Nn

[u exp (n )]Nn

(6.104)

Fermions For fermions, the number Nn can take only values Nn = 0 and Nn = 1. Hence, for fermions (denoted by the index F ), we obtain: F (, u) =
n

[1 + u exp (n )] .

Since the n-th level of a 3D oscillator is degenerate with the degeneracy g (n) = (n + 1) (n + 2) , 2

we nd that the generating function F (, u) is given by F (, u) = Bosons 142

n=0

[1 + u exp (n )]g(n) .

(6.105)

For bosons (denoted by the index B ), Nn = 0, 1, . . . , . The summations over {Nn } in Eq. (6.104) gives: B (, u) =
n=0

1 1 u exp (n )

g ( n)

(6.106)

The results (6.104) and (6.106) prove (for the partition function) the equivalence of the path-integral approach for identical particles [99] and of the second-quantization method. Integral representation Let us use the Fourier representation for the delta symbol: N, 1 = 2
2

Nn

exp i
0 n

Nn N

( i ) d,

(6.107)

where is an arbitrary constant. Substituting Eq. (6.107) into Eq. (6.103) we obtain ZI ( |N ) = exp
2

n Nn
n

{Nn }

1 2

exp i
0 n

Nn N

( i ) d Nn ( i )

1 = 2 = 1 2

d exp [iN ( i )]
0 2

{Nn }

exp

n Nn + i
n n

d exp (iN N ) , ei+

0 2

1 ZI ( |N ) = 2

d exp ln , ei+ N iN .
0

(6.108)

The partition function for a nite number of particles can be obtained from the generation function also by the inversion formula [100] ZI ( |N ) = 1 2i 1 = 2 (, z ) dz z N +1
2 0
i e[ln (,ue )N ln u] eiN d.

(6.109) (6.110)

Let us denote in Eq. (6.108): ln u. (6.111)

In these notations, Eqs. (6.108) and (6.110) are identical. For the numerical calculation, it is more convenient to choose in Eq. (6.107) the interval of the integration over as [, ] 143

instead of [0, 2 ] , what gives: 1 ZI ( |N ) = 2 with the function N () = exp ln , uei N ln u iN . (6.113)

N () d,

(6.112)

The aforesaid method of derivation of the partition function [Eqs. (6.107) to (6.108)] is heuristically useful, because it allows a simple generalization to spin-mixed systems with various polarization distributions. The two-point density-density correlation function in the operator formalism is g (q, | {N } , ) = q ( ) q (0) , where q (t) is the density operator in the Heisenberg representation: q ( ) = exp H q exp H . (6.115) (6.114)

In the second-quantization representation, q (t) is q ( ) =

n,n eiqx nn

a + n ( ) n ( ) a a + n exp na (n n ) . (6.116)

=
n,n

eiqx

nn

After substituting Eq. (6.116) into (6.114), we nd that g (q, | {N } , ) =

eiqx n,n m,m nn eiqx mm

exp

(n n )

+ m . a + n a ma na (6.117)

+ m has non-zero diagonal matrix elements in the basis of quantum n a The operator a + ma na states N,{Nn } only in the cases n = n Hence, the average + m = a + n a ma na

m = m

or

n = m

m = n

(6.118)

1 ZI ( |N )

{ Nn }

a + m N,{Nn } N,{Nn } exp H + n a ma na 144

(6.119)

is not equal to zero only when the condition (6.118) is fullled. This allows us to write down the average (6.119) as + m = n n m m (1 mn ) a + n a + m + m n n m (1 mn ) a a + n a + m a + n ma na ma na na ma + n n m m mn a + n a + n na na n N m + m n n m (1 mn ) = n n m m (1 mn ) N n + n n m m mn N n N m + m n n m N n 1 N m = n n m m N n = N 1 ZI ( |N ) . (6.120) n N n N m N

n : Here, the notation is used for the average occupation number N N n N,{Nn } . N,{Nn } exp H N,{Nn } exp H exp n Nn
n

(6.121)

{ Nn }

ZI ( |N ) =

In the same way as Eq. (6.102), the average (6.121) can be written down in the form 1 n = n Nn . Nn exp N ZI ( |N ) n
{Nn }
n

{Nn }

N,{Nn } .
Nn = N

(6.122)

{Nn }

Nn =N

n = N n N

1 ZI ( |N ) , ZI ( |N ) n

1 = 2 ZI ( |N )

N () d. exp (n i) + 1

(6.123)

n depends only on n. Since n = n , N n as Using Eq. (6.111), we can write N n N 1 = 2 ZI ( |N ) 1 = 2 ZI ( |N ) = 1 2 ZI ( |N )

1 u

N () d exp (n i) + 1

exp ln , uei N ln u iN d 1 exp (n i) + 1 u , uei exp [i (N 1) n ] d. u N 1 1 + u exp (i n ) 145 (6.124)

n N m The averages N sentation: n N m N =

for m = n can be also expressed in terms of the integral repre-

m =n

1 ZI ( |N )

{Nn }

N n N m N,{Nn } N,{Nn } exp H

m =n

n N m for We obtain the integral representation for the average of the product of operators N m = n: n N m N 1 = 2 ZI (N )

1 ZI ( |N ) ZI ( |N ) 2 m n 2 1 1 = ZI ( |N ) 2 m n 2 =

1 ZI ( |N )

{Nn } 2

Nn Nm exp

n Nn

Nn =N,

m =n

d exp ln , ei+ N iN

m =n

N () d. (6.125) [exp (n i) + 1] [exp (m i) + 1]

Let us introduce the notation f (, ) 1 , exp ( i) + 1 (6.126)

which formally coincides with the Fermi distribution function of the energy with the chemical potential ( + i) /. Using this notation, the averages (6.123) and (6.125) can be written down in the form n N 1 = 2 ZI ( |N ) = 1 2 ZI (N )

f (n , ) N () d,

(6.127)

n N m N

m =n

f (n , ) f (m , ) N () d.

(6.128)

We can develop the aforesaid procedure for the average of a product of any number of n1 N n2 . . . N n , where all quantum numbers n1 , n2 , . . . , nK are dierent. The operators N
K

result is: n1 N n2 . . . N n N K 1 = 2 ZI (N )

nj =nl

f (n1 , ) f (n2 , ) . . . f (nK , ) N () d. (6.129)

146

It should be emphasized, that all expressions above [including Eq. (6.129)] are derived for a canonical Hibbs ensemble (i. e., for a xed number of particles) and for both closed-shell and open-shell systems. Let us substitute the average (6.120) into the correlation function g (q, | {N } , ): g (q, | {N } , ) =
eiqx n,n m,m nn eiqx mm

exp

(n n )

n N m + m n n m N n 1 N m n n m m N = g (q, | {N } , ) = +
n,m The matrix elements eiqx eiqx n,m eiqx 2 nm nn eiqx mm

n N m N n 1 N m N . (6.130)

exp

( n m )

nm

has the following form

2 m2 eiq2 x n2 3 m3 eiq3 x n3 ,

eiqx

nm

1 = m1 eiq1 x n1

where m eiqx n is the matrix element of a one-dimensional oscillator with the frequency

w:

m e

iq x

q w,

2mw

() Ln

, |n are the quantum states of the one-dimensional oscillator with the frequency

2 (i )n> n< n = exp 2 n max (n, m) ; > n< min (n, m) ,

n< ! (n> n< ) 2 , L n> ! n< (6.131)

(z ) is the generalized Laguerre polynomial.

System with mixed spins The correlation functions for a system with mixed spins can be explicitly derived by the generalization of Eqs. (6.112) and (6.129) to the case of the particles with the non-zero spin. We use the fact, that the derivation of Eqs. (6.112) and (6.129), performed in this section,

147

does not depend on the concrete form of the energy spectrum n . Hence, in the formulae, derived above, the replacement should be made: |n |n, , n N n, = a+ an, , N n, (6.132)
mn

where is the electron spin projection. Consequently, the matrix elements eiqx

are

replaced by
eiqx mn m, eiqx n, = eiqx mn

(6.133)

Taking into account Eqs. (6.132) and (6.133), the two-point correlation function (6.130) becomes g (q, | {N } , ) = +
n,m eiqx n,m 1 ,2 eiqx 2 nm nn eiqx mm

n,1 N m,2 N n, 1 N m, N . (6.134)

exp

( n m )

n, The averages N distribution functions,

n, N m, and N 1 2

are, respectively, one-particle and two-particle

n, f1 (n, |N , ) , N n, N n , f2 (n, ; n , | {N } , ) . N

(6.135) (6.136)

The one-electron distribution function f1 (n, |N , ) is the average number of electrons with the spin projection at the n-th energy level, while the two-electron distribution function f2 (n, ; n , | {N } , ) is the average product of the numbers of electrons with the spin following integrals [see (6.127), (6.128)]: 1 f (n , ) (, , N ) d, (6.137) f1 (n, |N , ) = 2 ZF (N , w, ) 1 f (n , ) f (n , ) (, , N ) d, if = ; 2 ZF (N ,w, ) f2 (n, ; n , | {N } , ) = f1 (n, |N , ) f1 (n, |N , ) , if = (6.138) with the notations (, , N ) = exp
n=0

projections and at the levels n and n . These functions are expressed through the

ln 1 + ei+n N ( + i) , 148

(6.139)

1 . (6.140) exp ( i) + 1 The function f (, ) formally coincides with the Fermi-Dirac distribution function of the f (, ) energy with the chemical potential ( + i) /. Here we consider the zero-temperature limit, for which the integrals (6.137) and (6.138) can be calculated analytically. The result for the one-electron distribution function is 1, n < L ; (6.141) f1 (n, |, N )| = 0, n > L ; N NL , n = L . gL

According to (6.141), L is the number of the lowest open shell, and 1 (n + 1) (n + 2) (3D ) , 2 gn = n+1 (2D ) . with the spin projection , NL
L 1 n=0

is the degeneracy of the n-th shell. NL is the number of electrons in all the closed shells 1 L (L + 1) (L + 2) 6 gn = 1 L (L + 1)
2

(3D ) , (2D ) .

(6.142)

The two-electron distribution function f2 (n, ; n , | {N } , ) at T = 0 takes the form f2 (n, ; n , |, {N })| n = n or = f1 (n, |, N )| f1 (n , |, N )| , 1, = and n = n < L ; = 0, = and n = n > L ; N NL N NL 1 , = and n = n = L . gL g L 1

(6.143)

In summary, we have obtained the following expression for g (q, | {N } , ): g (q, | {N } , ) = +

n,n ,

eiqx
n,,n ,

nn 2

eiqx exp

n n

f2 (n, ; n , | {N } , )

eiqx

nn

( n n ) (6.144)

[f1 (n, | {N } , ) f2 (n, ; n , | {N } , )] .

This formula is valid for both closed and open shells. The correlation functions derived in this subsection are used both for the calculation of the ground-state energy and, a shown 149

below, for the calculation of the optical conductivity of an N -polaron system in a quantum dot.

E.

Many-polaron ground state in a quantum dot: extrapolation to the homoge-

neous limit and comparison to the results for a polaron gas in bulk [88]

The correlation function given by Eq. (6.144) can be subdivided as g (q, | {N } , ) = g 1 (q, | {N } , ) + g 2 (q, | {N } , ) , with g 1 (q, | {N } , ) eiqx
n,n , 2 nn

(6.145)

exp

( n n ) (6.146) (6.147)

[f1 (n, | {N } , ) f2 (n, ; n , | {N } , )] , g 2 (q, | {N } , ) eiqx

n,,n , nn

eiqx

n n

f2 (n, ; n , | {N } , ) .

In accordance with the subdivision (6.145) of the correlation function, we subdivide the Coulomb and polaron contributions: EC = EC + EC , Eeph = Eeph + Eeph .
(1) (2) (1) (2)

(6.148) (6.149)
(1)

We have numerically checked whether the polaron contribution per particle Eeph /N tends to a nite value at N . In Figs. 14 and 15, we have plotted the polaron contributions Eeph /N as a function of N for a quantum dot in ZnO and in a polar medium with = 5, = 0.3, respectively.
3 (1)

As seen from Fig. 14, the polaron contribution Eeph /N

(1)

(1)

in ZnO as a function of N oscillates taking expressed maxima for N corresponding to the closed shells N = 2, 8, 20, 40, . . .. There exist kinks of Eeph /N at N corresponding to the half-lled shells, but these kinks are extremely small. In the case of the medium with = 5,
= 20 (what corresponds to the density n0 1.14 1018 cm3 ), the polaron = 0.3, for rs
3

Since, as discussed above, for a single polaron only the whole polaron contribution Eeph = Eeph + Eeph has a physical meaning, the plots for Eeph in Figs. 3 and 4 start from N = 2. The total polaron contribution Eeph for N = 1 is plotted below, in Fig. 9.
(1)

(1)

(2)

150

contribution Eeph /N oscillates taking maximal values at the numbers of fermions, which correspond to the closed shells for a spin-polarized system with parallel spins. In Figs. 14 and 15, the dashed curves are the envelopes for local maxima (closed shells) and local minima of Eeph /N.We see that when these envelopes are extrapolated to larger number of fermions, the distance between the envelopes decreases. Therefore, the magnitude of the variations of Eeph /N related to the shell lling diminishes with increasing N , and it is safe to suppose that in the limit of large N, the envelopes tend to one and the same value. That value corresponds to the homogeneous (bulk) limit limN Eeph /N .
(1) (1) (1)

(1)

151

-0.4

Polaron contribution per particle (in units

LO

-0.5

Spin

-0.6

0 0 8 16 24 32

-0.7

Number of fermions

ZnO: = 0.849 = 0.4908

-0.8 0 4 8 12 16

Closed shells

Half-filled shells

20

24

28

32

36

40

44

48

Number of fermions

FIG. 14: Polaron contribution Eeph /N as a function of N for a ZnO quantum dot.The material
= 2 corresponds to n = 4.34 1019 parameters for ZnO are taken from Ref. [89]. The value rs 0

(1)

cm2 . Inset: the total spin as a function of N .

Polaron contribution per particle (in units

LO

-3.5

20

15

-4.0

Spin

10

-4.5

16

24

32

40

= 5 = 0.3

Number of fermions

ap = 3 nm
-5.0 0 4 8 12 16

Closed shells

20

24

28

32

36

40

44

48

Number of fermions

FIG. 15: Polaron contribution Eeph /N as a function of N for a quantum dot of a polar medium
= 20 corresponds to n = 1.14 1018 cm3 . Inset: the total with = 5, = 0.3. The value rs 0

(1)

spin as a function of N .

In Fig. 16, we compare the polaron contribution Eeph /N calculated within our variational path-integral method for dierent numbers of fermions with the polaron contribution to the ground-state energy per particle for a polaron gas in bulk, calculated (i) in Ref. [101] within an intermediate-coupling approach (the thin solid curve), (ii) in Ref. [102], using a 152

(1)

variational approach developed rst in Ref. [89]. As seen from this gure, our all-coupling variational method provides lower values for the polaron contribution than those obtained in Refs. [101, 102]. The dierence between the polaron contribution calculated within our method and that of Ref. [101] is smaller at low densities and increases in magnitude with increasing density. The dierence between the polaron contribution calculated within our method and that of Ref. [102] very slightly depends on the density. The result of Ref. [102] becomes closer to our result only at high densities.

FIG. 16: Polaron contribution to the polaron ground-state energy per particle Eeph in an N polaron quantum dot as a function of the eective density. The parameters are taken for ZnO (see Ref. [89]): = 0.849, 0 = 8.15, = 4.0, LO = 73.27 meV, mb = 0.24me , where me is the electron mass in the vacuum. This polaron contribution is compared with the polaron contribution to the ground-state energy of a polaron gas in bulk calculated in Refs. [101, 102].

(1)

153

F.

Optical conductivity

In Ref. [96] we have extended the memory-function approach to a system of arbitrarycoupling interacting polarons conned to a parabolic connement potential. The optical conductivity relates the current J (t) per electron to a time-dependent uniform electric eld E (t) in the framework of linear response theory. Further on, the Fourier components of the electric eld are denoted by E ( ) : E (t) = 1 2

E ( ) eit d,

(6.150)

and the similar denotations are used for other time-dependent quantities. The electric current per electron J (t) is related to the mean electron coordinate response R (t) by J (t) = e and hence J ( ) = ie R ( ) . (6.152) dR (t) , dt (6.151)

Within the linear-response theory, both the electric current and the coordinate response are proportional to E ( ): J ( ) = ( ) E ( ) , R ( ) = ( ) E ( ) , ie (6.153)

where ( ) is the conductivity per electron. Because we treat an isotropic electron-phonon system, ( ) is a scalar function. It is determined from the time evolution of the center-ofmass coordinate: 1 R (t) N The symbol ( )
S N

xj (t)
j =1 S

(6.154)

denotes an average in the real-time representation for a system with

x x

action functional S : ( )
S

dx

dx 0

dx 0
x 0

Dx (t)
x 0

Dx (t) e

S [x (t),x (t)]

( ) x 0 | (t0 )| x 0

t0

(6.155) where x 0 | (t0 )| x 0 is the density matrix before the onset of the electric eld in the innite past (t0 ). The corresponding action functional is
t

S [x (t) , x (t)] =

(t) , x (t) , x Le x (t) , t Le x (t) , t

dt i [x (t) , x (t)] , (6.156)

154

,x where Le x , t is the Lagrangian of N interacting electrons in a time-dependent uniform electric eld E (t)
N

,x Le x , t =
j =1

2 mb 2 mb x 2 0 xj, j, exj, E (t) 2 2

N N ,

e2 2 |xj, xl, |

j =1 l=1 (j,)=(l, )

(6.157) The inuence phase of the phonons [x ( s) , x (s)] =

|Vq |2
2 q

ds

ds q (s) q (s) (6.158)

T (s s ) q (s ) Tq (s s ) q (s ) q

describes both a retarded interaction between dierent electrons and a retarded selfinteraction of each electron due to the elimination of the phonon coordinates. This functional contains the free-phonon Greens function: T (t) = The equation of motion for R (t) is d2 R (t) mb + mb 2 0 R (t) + eE (t) = Fph (t) , 2 dt where Fph (t) is the average force due to the electron-phonon interaction, Fph (t) = Re 2 |Vq |2 q N
t ds T (t s) q (t) q (s) LO S

eit 1 e

eit . e 1

(6.159)

(6.160)

(6.161)

The two-point correlation function

q (t) q (s)

should be calculated from Eq. (6.155)

using the exact action (6.156), but like for the free energy above, this path integral cannot be calculated analytically. Instead, we perform an approximate calculation, replacing the two-point correlation function in Eq. (6.161) by q (t) q (s)
S0 ,

where S0 [x (t) , x (t)]

is the action functional with the optimal values of the variational parameters for the model system considered in the previous section in the presence of the electric eld E (t). The functional S0 [x (t) , x (t)] is quadratic and describes a system of coupled harmonic oscillators in the uniform electric eld E (t). This eld enters the term eE (t)
N j =1 xj,

in the

155

Lagrangian, which only aects the center-of-mass coordinate. Hence, a shift of variables to the frame of reference with the origin at the center of mass x (t) = x n (t) + R (t) , n x (t) = x (t) + R (t) ,
n n

(6.162)

results in

q (t) q (s)

S0

q (t) q (s)

iq[R(t)R(s)] . S0 E =0 e

(6.163)

This result (6.163) is valid for any quadratic model action S0 . The applicability of the parabolic approximation is conrmed by the fact that a selfinduced polaronic potential, created by the polarization cloud around an electron, is rather well described by a parabolic potential whose parameters are determined by a variational method. For weak coupling, our variational method is at least of the same accuracy as the perturbation theory, which results from our approach at a special choice of the variational parameters. For strong coupling, an interplay of the electron-phonon interaction and the Coulomb correlations within a connement potential can lead to the assemblage of polarons in multi-polaron systems. Our choice of the model variational system is reasonable because of this trend, apparently occurring in a many-polaron system with arbitrary N for a nite connement strength. The correlation function q (t) q (s)
S0 E=0

corresponds to the model system in the ab-

sence of an electric eld. For t > s, this function is related to the imaginary-time correlation function G (q, | {N } , ) , described in the previous section: q (t) q (s)
S0 E=0,t>s

= G (q, i (t s) | {N } , ) .

(6.164)

Using the transformation (6.162) and the relation (6.164), we obtain from Eq. (6.161) Fph (t) = Re 2 |Vq |2 q N
t T (t s) eiq[R(t)R(s)] G (q, i (t s) | {N } , ) ds. (6.165) LO

Within the linear-response theory, we expand the function eiq[R(t)R(s)] in Eq. (6.165) no contribution into Fph (t) due to the symmetry of |Vq |2 and of fq (t s) with respect to the inversion q q. In this approach, the Cartesian coordinates of the force (j = 1, 2, 3) 156 as a Taylor series in [R (t) R (s)] up to the rst-order term. The zeroth-order term gives

become
3

(Fph (t))j =
k =1 q

2 |Vq |2 qj qk N

[Rk (t) Rk (s)] (6.166)

Im T (t s) G (q, i (t s) | {N } , ) ds. LO

Further on, we perform the Fourier expansion: R (t) = 1 2

R ( ) eit d.

(6.167)

In Eq. (6.166), we make the replacement t s, what gives

3

s = t ,

(Fph (t))j =
k =1 3 q

2 |Vq |2 qj qk N
2

d [Rk (t) Rk (t )] Im T ( ) G (q, it| {N } , ) LO

=
k =1 q

2 |Vq | qj qk 1 N 2 dFj ( ) eit ,

dRk ( ) eit
0

1 ei Im T ( ) fq ( ) LO

1 2

where the Fourier component of the force is

3

(Fph ( ))j =
k =1 q

2 |Vq |2 qj qk N

dt 1 eit Im T ( ) G (q, it| {N } , ) Rk ( ) . LO 0

(6.168) The expression (6.168) can be written down as

3

(Fph ( ))j = mb where jk ( ) are components of the tensor jk ( ) =

q

jk ( ) Rk ( ) ,
k =1

(6.169)

2 |Vq |2 qj qk N mb

dt eit 1 Im T (t) G (q, it| {N } , ) . LO 0

(6.170)

In the abstract tensor form, Eq. (6.169) is F ( ) = ( ) R ( ) . 157 (6.171)

In particular, for the isotropic electron-phonon interaction and in the absence of the magnetic eld, the tensor ( ) is proportional to the unity tensor I, ( ) = ( ) I , where ( ) is the scalar memory function: ( ) =
q

(6.172)

2 |Vq |2 q 2 3N mb

dt eit 1 Im T (t) G (q, it| {N } , ) . LO

(6.173)

Let us perform the Fourier transformation of the equation of motion (6.160):

2 mb 2 R ( ) + eE ( ) = Fph ( ) . 0

(6.174)

With Eq. (6.171), this equation takes the form 2 mb 2 R ( ) + eE ( ) = mb ( ) R ( ) 0 mb 2 2 0 ( ) R ( ) = eE ( ) . Comparing Eqs. (6.153) and (6.175) between each other, we nd that mb 2 2 0 ( ) so that ( ) E ( ) = eE ( ) , ie (6.175)

ie2 2 1 2 . 0 ( ) mb In the case when Eq. (6.172) is valid, we obtain the conductivity in the scalar form ( ) = ( ) = The real part of the conductivity is Re ( ) = Re
2 mb [( 2 2 0 ) Re ( )] + [Im ( )]

ie2 . mb [ 2 2 0 ( )]

ie2 [( 2 2 0 ) Re ( ) + i Im ( )]
2

e2 Im ( ) = . 2 2 mb [( 0 ) Re ( )]2 + [Im ( )]2 ( ) (cf. Ref. [48]), Re ( ) =

In summary, the optical conductivity can be expressed in terms of the memory function Im ( ) e2 , 2 mb [ 2 0 Re ( )]2 + [Im ( )]2 158

(6.176)

where ( ) is given by ( ) =
q

2 |Vq |2 q 2 3N mb

dt eit 1 Im T (t) G (q, it| {N } , ) . LO 0

(6.177)

It is worth noting that the optical conductivity (6.176) diers from that for a translationally invariant polaron system both by the explicit form of ( ) and by the presence of the term 2 0 in the denominator. For 0, the optical conductivity tends to a -like peak at = 0 ,
0

lim Re ( ) =

e2 ( 0 ) . 2mb

(6.178)

For a translationally invariant system 0 0, and this weak-coupling expression (6.178) reproduces the central peak of the polaron optical conductivity [80]. The further simplication of the memory function (6.177) is performed in the following way. With the Fr ohlich amplitudes of the electron-phonon interaction, we transform the summation over q to the integral and use the Feynman units ( = 1, LO = 1, mb = 1), in which |Vq |2 =
2 2 . q2 V

We also use the fact that in an isotropic crystal, G (q, it| {N } , ) =

0

G (q, it| {N } , ). As a result, we nd V ( ) = (2 )3

2 2 2 2 q 4q 2 dq 2 3N q V

2 2 = 3N

dt eit 1 Im T (t) G (q, it| {N } , ) LO 0

q dq
0

dt eit 1 Im T (t) G (q, it| {N } , ) . LO

In the zero-temperature case, T (t) eit , and we arrive at the expression LO

2 2 ( ) = 3N

q 2 dq
0

dt eit 1 Im eit G (q, it| {N } , ) .

(6.179)

Substituting the two-point correlation function G (q, it| {N } , ) with the one-electron (6.141) and the two-electron (6.143) distribution functions into the memory function (6.179) and expanding G (q, it| {N } , ) in powers of eiwt , ei1 t and ei2 t , the integrations over q and t in Eq. (6.179) are performed analytically. The similar transformations are performed also in the 2D case. As a result, the memory function (6.177) is represented in the unied

159

form for 3D and 2D interacting polarons: 2 ( ) = lim +0 3N

m

3 4

3D

LO A

3/2

p1 =0 p2 =0 p3

(1)p3 p !p !p ! =0 1 2 3

a2 1 N 1 A

p1

a2 2 N 2 A

p2

1 NwA

p3

m=0 n=0

[f1 (n, | {N } , ) f2 (n, ; m, | {N } , )]|

1 LO [p1 1 +p2 2 +(p3 m+n)w ]+i

l=0

n+D1 nk

(1)nm+l+k p1 + p2 + p3 + k + l + k !l! k =nm+l 2k kln+m

+P

2 LO +p1 1 +p2 2 +(p3 m+n)w

1 +LO +p1 1 +p2 2 +(p3 m+n)w +i

3 2

1 wA

l +k

+
n

1 LO (p1 1 +p2 2 +p3 w )+i

+P

2 LO +p1 1 +p2 2 +p3 w

1 +LO +p1 1 +p2 2 +p3 w +i

m=0 n=0 , m

f2 (n, ; m, | {N } , )|
3 2

k =0 l=0

(1)k+l p1 + p2 + p3 + k + l + k !l! m+D1 ml ,

1 wA

k +l

n+D1 nk

(6.180)

where D = 2, 3 is the dimensionality of the space, P denotes the principal value, A is dened as A
2 i=1

a2 i /i + (N 1) /w /N , 1 , 2 , and w are the eigenfrequencies of the model

system, a1 and a2 are the coecients of the canonical transformation which diagonalizes the model Lagrangian (6.32).

1.

Selected results: the manifestations of the shell lling in optical conductivity

The shell lling schemes for an N -polaron system in a quantum dot can manifest themselves in the spectra of the optical conductivity. In Fig. 17, optical conductivity spectra for N = 20 polarons are presented for a quantum dot with the parameters of CdSe: = 0.46, 160

 = 0.656 [5] and with dierent values of the connement energy 0 . increasing 0 in the interval 0.0421H < 0 < 0.0422H .

In this case,

the spin-polarized ground state changes to the ground state satisfying Hunds rule with

a b
1 CdSe

LO

N
0.01

)]

= 0.46, = 20 = 0.03

= 0.656

e mb

[in units

/(

= 0.0421

= 10

= 10

LO

1
0

c d
= 0.0422

0.18

S
Re
0.01

*
0.17

= 0

0.040
0

(in

0.045

= 0.06

= 0

(in units

LO

FIG. 17: Optical conductivity spectra of N = 20 interacting polarons in CdSe quantum dots with = 0.46, = 0.656 for dierent connement energies close to the transition from a spin-polarized ground state to a ground state obeying Hunds rule. Inset : the rst frequency moment of the optical conductivity as a function of the connement energy. (From Ref. [96].)

In the inset to Fig. 17, the rst frequency moment of the optical conductivity
Re ( ) d 0 , Re ( ) d 0

(6.181)

as a function of 0 shows a discontinuity, at the value of the connement energy corresponding to the change of the shell lling schemes from the spin-polarized ground state to the ground state obeying Hunds rule. This discontinuity might be observable in optical measurements.
4

For the numerical calculations, we use eective atomic units, where , the electron band mass mb and e/ have the numerical value of 1. This means that the unit of length is the eective Bohr radius 2 a / mb e2 , while the unit of energy is the eective Hartree H = mb e4 / 2 2 . B =

161

The shell structure for a system of interacting polarons in a quantum dot is clearly revealed when analyzing the addition energy and the rst frequency moment of the optical conductivity in parallel. In Fig 18, we show both the function (N ) |N +1 2 |N + |N 1 , and the addition energy (N ) = E 0 (N + 1) 2E 0 (N ) + E 0 (N 1) . for interacting polarons in a 3D CdSe quantum dot.
LO

(6.182)

(6.183)

CdSe 0.01 = 0.46 = 0.656 = 0.1

) (in units

a
H
*

0.00

) (in units

LO

0.6

0.4

N
0.2 0

12

16

20

24

Number of electrons

FIG. 18: The function (N ) and the addition energy (N ) for systems of interacting polarons in CdSe quantum dots with = 0.46, = 0.656 for 0 = 0.1H . (From Ref. [96].)

As seen from Fig 18, distinct peaks appear in (N ) and (N ) at the magic numbers corresponding to closed-shell congurations at N = 8, 20 and to half-lled-shell congurations at N = 5, 14. We see that each of the peaks of (N ) corresponds to a peak of the addition energy. The lling patterns for a many-polaron system in a quantum dot can be therefore determined from the analysis of the rst moment of the optical absorption for dierent numbers of polarons.

162

VII.

VARIATIONAL PATH-INTEGRAL TREATMENT OF A TRANSLATION

INVARIANT N -POLARON SYSTEM A. The many-polaron system

In the present section, the ground-state properties of a translation invariant N -polaron system are theoretically studied in the framework of the variational path-integral method for identical particles, using a further development [103] of the model introduced in Refs. [96, 98, 104]. In order to describe a many-polaron system, we start from the translation invariant N polaron Hamiltonian
N

H=
j =1

p2 1 j + 2m 2

e2 + |r rl | j =1 l=1,=j j

LO a k ak
k

+
j =1 k

Vk ak eikrj + H.c. , (7.1)

where m is the band mass, e is the electron charge, LO is the longitudinal optical (LO) phonon frequency, and Vk are the amplitudes of the Fr ohlich electron-LO-phonon interaction Vk = i LO k 4 V
1/2 1/4

2mLO

e2 2 LO

2mLO

1/2

1 1 0

(7.2)

with the electron-phonon coupling constant > 0, the high-frequency dielectric constant > 0 and the static dielectric constant 0 > 0, and consequently e2 > 2 LO m
1/2

2 <

H LO

1/2

U.

(7.3)

In the expression (7.3), H is the eective Hartree H = e2 , a B a B =

2

me2 /

(7.4)

where a B is the eective Bohr radius. The partition function of the system can be expressed as a path integral over all electron and phonon coordinates. The path integral over the phonon variables can be calculated analytically [41]. Feynmans phonon elimination technique for this system is well known and leads to the partition function, which is a path integral over the electron coordinates only: Z=
k

e 2 LO 2 sinh 1 LO 2 163

eS D r

(7.5)

where r = {r1 , , rN } denotes the set of electron coordinates, and

r( )= r

D r denotes the path

integral over all the electron coordinates, integrated over equal initial and nal points, i.e. eS D r d r
r(0)= r

eS D r ( ) .

Throughout this paper, imaginary time variables are used. The eective action for the N -polaron system is retarded and given by

S= + 1 2

0 0 0

m 2

j =1 N

d r j ( ) d

1 + 2

j =1 l=1,=j

e2 |rj ( ) rl ( )|

j,l=1 k

|Vk |2 eik(rj ( )rl ())

1 cosh LO 2 | | dd. 1 sinh 2 LO

(7.6)

Note that the electrons are fermions. Therefore the path integral for the electrons with parallel spin has to be interpreted as the required antisymmetric projection of the propagators for distinguishable particles. We below use units in which = 1, m = 1, and LO = 1. The units of distance and energy are thus the eective polaron radius [ / (mLO )]1/2 and the LO-phonon energy LO .

B.

Variational principle

For distinguishable particles, it is well known that the Jensen-Feynman inequality [41] provides a lower bound on the partition function Z (and consequently an upper bound on the free energy F ) Z= e D r=
S

e D r

S0

S S0

e D r e

S0

S S0

with A
0

A ( r) eS0 D r , S 0 e D r (7.7) (7.8)

eF eF0 e S S0 0 = F F0

S S0

for a system with real action S and a real trial action S0 .The many-body extension (Ref. [97, 105]) of the Jensen-Feynman inequality, requires that the potentials are symmetric with respect to all particle permutations, and that the exact propagator as well as the model propagator are dened on the same state space. Within this interpretation we consider the

164

following generalization of Feynmans trial action

S0 =

1 2

j =1 N

d r j ( ) d
0 0

2 + w2 v2 + 4N

j,l=1

(rj ( ) rl ( ))2 d (7.9)

w v2 w2 8 N

j,l=1

(rj ( ) rl ( ))2

1 | | cosh w 2 dd w sinh 1 2

with the variational frequency parameters v, w, . Using the explicit forms of the exact (7.6) and the trial (7.9) actions, the variational inequality (7.8) takes the form F ( |N , N ) F0 ( |N , N ) + +w v 4N
2 2 2 0 0 0

U 2

0 N

1 | r j ( ) r l ( )| j,l=1,=j (rj ( ) rl ( ))2 (rj ( ) rl ( )) eik(rj ( )rl ())

2 0

d
0

d
0

j,l=1 N

w v2 w2 8 N 1 2
0 0

j,l=1 N

| | cosh w 1 2 dd w sinh 1 2

|Vk |2

j,l=1

1 cosh LO 2 | | dd. 1 sinh 2 LO

(7.10) In the zero-temperature limit ( ), we arrive at the following upper bound for the E 0 (N , N ) Evar (N , N |v, w, ) , with Evar (N , N |v, w, ) = 1 1 3 (v w )2 3 + EF (N ) + EF (N ) 4 v 4 2 2

ground-state energy E 0 (N , N ) of a translation invariant N -polaron system

+ EC (N ) + EC (N ) + EC (N , N ) + E (N ) + E (N ) + E (N , N ) , (7.11)

where EF (N ) is the energy of N spin-polarized fermions conned to a parabolic potential with the connement frequency , EC spins, E N() is the Coulomb energy of the electrons with parallel spins, EC (N , N ) is the Coulomb energy of the electrons with opposite N() is the electron-phonon energy of the electrons with parallel spins, and E (N , N ) is the electron-phonon energy of the electrons with opposite spins. 165

C.

Results

Here, we discuss some results of the numerical minimization of Evar (N , N |v, w, ) with respect to the three variational parameters v , w , and . The Fr ohlich constant and the Coulomb parameter U 1 0 0 with = (7.12) 1 2 characterize the strength of the electron-phonon and of the Coulomb interaction, obeying the physical condition 0 [see (7.3)]. The optimal values of the variational parameters v,w,and are denoted vop ,wop ,and op , respectively. The optimal value of the total spin was always determined by choosing the combination (N , N ) for xed N = N + N which corresponds to the lowest value E 0 (N ) of the variational functional E 0 (N ) min Evar (N , N N |vop , wop , op ) .
N

(7.13)

FIG. 19: The phase diagrams of a translation invariant N -polaron system. The grey area is the non-physical region, for which > 0 . The stability region for each number of electrons is determined by the equation c < < 0 . (From Ref. [103].)

In Fig. 19, the phase diagrams analogous to the bipolaron phase diagram of Ref. [106] are plotted for an N -polaron system in bulk with N = 2, 3, 5, and 10. The area where 166

0 is the non-physical region. For > 0 , each sector between a curve corresponding to a well dened N and the line indicating 0 = shows the stability region where op = 0, while the white area corresponds to the regime with op = 0. When comparing the stability region for N = 2 from Fig. 19 with the bipolaron phase diagram of Ref. [106], the stability region in the present work starts from the value c 4.1 (instead of c 6.9 in Ref. [106]). The width of the stability region within the present model is also larger than the width of the stability region within the model of Ref. [106]. Also, the absolute values of the ground-state energy of a two-polaron system given by the present model are smaller than those given by the approach of Ref. [106]. The dierence between the numerical results of the present work and of Ref. [106] is due to the following distinction between the used model systems. The model system of Ref. [106] consists of two electrons interacting with two ctitious particles and with each other through quadratic interactions. But the trial Hamiltonian given by Eq. (6) of Ref. [106] is not symmetric with respect to the permutation of the electrons. It is only symmetric under the permutation of the pairs electron + ctitious particle. As a consequence, this trial system is only applicable if the electrons are distinguishable, i.e. have opposite spin. In contrast to the model of Ref. [106], the model used in the present paper is described by the trial action (9), which is fully symmetric with respect to the permutations of the electrons, as is required to describe identical particles. The phase diagrams for N > 2 demonstrate the existence of stable multipolaron states (see Ref. [107]). As distinct from Ref. [107], here the ground state of an N -polaron system is investigated supposing that the electrons are fermions. As seen from Fig. 20, for N > 2, the stability region for a multipolaron state is narrower than the stability region for N = 2, and its width decreases with increasing N . A consequence of the Fermi statistics is the dependence of the polaron characteristics and of the total spin of an N -polaron system on the parameters (, 0 ,N ). In Fig. 20, we present the ground-state energy per particle, the connement frequency op and the total spin S as a function of the coupling constant for 0 / = 1.05 and for a dierent numbers of polarons. The ground-state energy turns out to be a continuous function of , while op and S reveal jumps. For N = 2 (the case of a bipolaron), we see from Fig. 20 that the ground state has a total spin S = 0 for all values of , i. e., the ground state of a bipolaron is a singlet. This result is in agreement with earlier investigations on the large-bipolaron 167

FIG. 20: The ground-state energy per particle (a ), the optimal value op of the connement frequency (b ), and the total spin (c ) of a translation invariant N -polaron system as a function of the coupling strength for 0 / = 0.5. The vertical dashed lines in the panel c indicate the critical values c separating the regimes of > c , where the multipolaron ground state with op = 0 exists, and < c , where op = 0. (From Ref. [103].)

problem (see, e. g., [108]). In summary, using the extension of the Jensen-Feynman variational principle to the systems of identical particles, we have derived a rigorous upper bound for the free energy of a translation invariant system of N interacting polarons. The developed approach is valid for an arbitrary coupling strength. The resulting ground-state energy is obtained taking into account the Fermi statistics of electrons.

168

VIII. A.

RIPPLONIC POLARONS IN MULTIELECTRON BUBBLES Ripplon-phonon modes of a MEB

Spherical shells of charged particles appear in a variety of physical systems, such as fullerenes, metallic nanoshells, charged droplets and neutron stars. A particularly interesting physical realization of the spherical electron gas is found in multielectron bubbles (MEBs) in liquid helium-4. These MEBs are 0.1 m 100 m sized cavities inside liquid helium, that contain helium vapor at vapor pressure and a nanometer-thick electron layer anchored to the surface of the bubble [110]. They exist as a result of equilibrium between the surface tension of liquid helium and the Coulomb repulsion of the electrons [111]. Recently proposed experimental schemes to stabilize MEBs [112] have stimulated theoretical investigation of their properties. We describe the dynamical modes of an MEB by considering the motion of the helium surface (ripplons) and the vibrational modes of the electrons together. In particular, we analyze the case when the ripplopolarons form a Wigner lattice [109]. First, we derive the Lagrangian of interacting ripplons and phonons within a continuum approach. The shape of the surface of a bubble is described by the function R (, ) = Rb + u (, ) , where u (, ) is the deformation of the surface from a sphere with radius Rb . The deformation can be expanded in a series of spherical harmonics Ylm (, ) with amplitudes Qlm , u (, ) =
l

Qlm Ylm (, ) . l (l + 1) |Qlm | Rb .

(8.1)

l=1 m=l

We suppose that the amplitudes are small in such a way that

The ripplon contribution (Tr ) to the kinetic energy of an MEB, and the contributions to the potential energy due to the surface tension (U ) and due to the pressure (UV ) were described in Ref. [113]:
2 3 1 lm , Tr = Rb Q 2 l=1 m=l l + 1 l

U =

2 4Rb

4 3 UV = pRb + pRb |Qlm |2 . 3 l=1 m=l 169

(l2 + l + 2) |Qlm |2 , + 2 l=1 m=l

l

(8.2)

Here 145 kg/m3 is the density of liquid helium, 3.6 104 J/m2 is its surface tension, and p is the dierence of pressures outside and inside the bubble. Expanding the surface electron density n (, ) in a series of spherical harmonics with amplitudes nlm , n (, ) =
l

nlm Ylm (, ) ,

(8.3)

l=0 m=l

the kinetic energy of the motion of electrons can be written as 1 Tp = 2

l=1 6 4me Rb |n lm |2 , l(l + 1)N m=l l

(8.4)

where me is the bare electron mass and N is the number of electrons. Finally, the electrostatic energy (UC ) of the deformed MEB with a non-uniform surface electron density (8.3) is calculated using the Maxwell equations and the electrostatic boundary conditions at the surface. The result is: UC = e2 N 2 3 + 2e2 Rb 2Rb e2 N 2 3 8Rb
l=1 l l=1

|nlm |2 l + (l + 1) m=l

l=1 m=l l

l2 (l + 1) |Qlm |2 l + (l + 1) (8.5)

e2 N

l+1 nlm Q lm , l + ( l + 1) m=l

with the dielectric constant of liquid helium 1.0572. The last term in Eq. (8.5) describes the ripplon-phonon mixing. Only ripplon and phonon modes which have the same angular momentum couple to each other. After the diagonalization of the Lagrangian of this ripplonphonon system, we arrive at the eigenfrequencies: 1,2 (l) = 1 2 2 (l ) + r (l ) 2 p
2 (l ) 2 (l ) p r 2 1/2

+ 4 2 (l )

(8.6)

where r (l) is the bare ripplon frequency, r (l ) = l+1 l2 + l + 2 3 Rb e2 N 2 l2 (l + 1) + 2pRb 3 4Rb l + (l + 1) 170
1/2

(8.7)

while p (l) is the bare phonon frequency, p (l ) = e2 N l (l + 1) 3 me Rb l + (l + 1)

1/2

(8.8)

and (l) describes the ripplon-phonon coupling: e2 N (l ) = 3 Rb

B.

Nl 3 4me Rb

1/2

(l + 1)2 . l + (l + 1)

(8.9)

Electron-ripplon interaction in the MEB

The interaction energy between the ripplons and the electrons in the multielectron bubble can be derived from the following considerations: (i) the distance between the layer electrons and the helium surface is xed (the electrons nd themselves conned to an eectively 2D surface anchored to the helium surface) and (ii) the electrons are subjected to a force eld, arising from the electric eld of the other electrons. For a spherical bubble, this electric eld lies along the radial direction and equals E= Ne e. 2 r 2Rb (8.10)

A bubble shape oscillation will displace the layer of electrons anchored to the surface. The interaction energy which arises from this, equals the displacement of the electrons times the force eE acting on them. Thus, we get for the interaction Hamiltonian int = H
j

j ). e|E| u(

(8.11)

Here u() is the radial displacement of the surface in the direction given by the spherical j is the (angular) position operator for electron j . The displacement can be angle ; and rewritten using (8.1) and we nd int = H
j

e|E|

m Ym ( j ). Q
,m

(8.12)

Using the relation ,m = (1)(m|m|)/2 Q

(+1) ( a,m 2R3 b

+a + ,m ),

(8.13)

the interaction Hamiltonian can be written in the suggestive form int = H

,m j

j )( M,m Y,m( a,m + a + ,m ), 171

(8.14)

with the electron-ripplon coupling amplitude for a MEB given by M,m = (1)(m|m|)/2
C. Locally at approximation

Ne2 2 2Rb

( + 1) 3 2Rb

(8.15)

Substituting M,m into (8.14), we get int = H

,m j

Ne2 2 2Rb

( + 1) 3 2Rb

(8.16)

(1)(m|m|)/2

j) Y,m ( ( a,m + a + ,m ). Rb

In this expression, we consider the limit of a bubble so large that the surface becomes at on all length scales of interest. Hence we let Rb but keep /Rb = q a constant. This means we have to let as well. In this limit, i lim Y,0 () = sin[( + 1/2) + /4], sin (8.17)

and Y,0 () varies locally as a plane wave with wave vector q = /Rb . The wave function j )/Rb is furthermore normalized with respect to integration over the surface (with Y,m (
2 total area 4Rb ). Thus, we get in the locally at approximation

int = H
q j

Ne2 2 2Rb

q rj eiq. ( aq + a + q ), 2 (q )

(8.18)

or int = H
q j rj Mq eiq. ( aq + a + q ),

Mq = e|E|

q . 2 (q )

(8.19)

This corresponds in the limit of large bubbles to the interaction Hamiltonian expected for a at surface.

D.

Ripplopolaron in a Wigner lattice: the mean-eld approach

In their treatment of the electron Wigner lattice embedded in a polarizable medium such as a semiconductors or an ionic solid, Fratini and Qu emerais [114] described the eect of 172

the electrons on a particular electron through a mean-eld lattice potential. The (classical) lattice potential Vlat is obtained by approximating all the electrons acting on one particular electron by a homogenous charge density in which a hole is punched out; this hole is centered in the lattice point of the particular electron under investigation and has a radius given by the lattice distance d. Within this particular mean-eld approximation, the lattice potential can be calculated from classical electrostatics and we nd that for a 2D electron gas it can be expressed in terms of the elliptic functions of rst and second kind, E (x) and K (x), Vlat (r) = 4rd (d r )2 4rd + (d + r ) sgn (d r ) K ( d r )2 |d r | E 2e2 d2

(8.20)

Here, r is the position vector measured from the lattice position. We can expand this potential around the origin to nd the small-amplitude oscillation frequency of the electron lattice:
r d

lim Vlat (r) =

2e2 1 2 2 + me lat r + O r4 , d 2 e2 . me d3

(8.21)

with the connement frequency lat = (8.22)

In the mean-eld approximation, the Hamiltonian for a ripplopolaron in a lattice on a locally at helium surface is given by 2 = p H + Vlat ( r) + 2me +
q

(q ) a+ q qa
q

Mq eiq.r a q + a + q ,

(8.23)

where r is the electron position operator. Now that the lattice potential has been introduced, we can move on and include eects of the bubble geometry. If we restrict our treatment to the case of large bubbles (with N > 105 electrons), then both the ripplopolaron radius and the inter-electron distance d are much smaller than the radius of the bubble Rb . This gives us ground to use the locally at approximation using the auxiliary model of a ripplonic polaron in a planar system described by (8.23), but with a modied ripplon dispersion relation and an modied pressing eld. 173

We nd for the modied ripplon dispersion relation in the MEB: (q ) = 3 p q + q, Rb (8.24)

where Rb is the equilibrium bubble radius which depends on the pressure and the number of electrons. The bubble radius is found by balancing the surface tension and the pressure with the Coulomb repulsion. The modied electron-ripplon interaction amplitude in an MEB is given by Mq = e|E| q . 2 (q ) (8.25)

The eective electric pressing eld pushing the electrons against the helium surface and determining the strength of the electron-ripplon interaction is E=
E.

Ne e. 2 r 2Rb

(8.26)

Ripplopolaron Wigner lattice at nite temperature

To study the ripplopolaron Wigner lattice at nite temperature and for any value of the electron-ripplon coupling, we use the variational path-integral approach [41]. This variational principle distinguishes itself from Rayleigh-Ritz variation in that it uses a trial action functional Strial instead of a trial wave function. The action functional of the system described by Hamiltonian (8.23), becomes, after elimination of the ripplon degrees of freedom,

S=
0

1
0

me 2 r ( ) + Vlat [r ( )] + 2

|Mq |2 (8.27)

d
0

dG(q) ( )eiq[r( )r()] ,

with G ( ) = cosh[ (| | /2)] . sinh( /2) (8.28)

In preparation of its customary use in the Jensen-Feynman inequality, the action functional (8.27) is written in imaginary time t = i with = 1/(kB T ) where T is the temperature.

174

We introduce a quadratic trial action of the form

Strial =

1
0

me 2 me 2 2 r ( ) + r ( ) 2 2

Mw 2 4
0

d
0

dGw ( )r( ) r( ).

(8.29)

where M, w, and are the variationally adjustable parameters. This trial action corresponds to the Lagrangian L0 = me 2 M 2 2 K r + R r (r R )2 , 2 2 2 2 (8.30)

from which the degrees of freedom associated with R have been integrated out. This Lagrangian can be interpreted as describing an electron with mass me at position r, coupled through a spring with spring constant to its lattice site, and to which a ctitious mass M at position R has been attached with another spring, with spring constant K . The relation between the spring constants in (8.30) and the variational parameters w, is given by w= = K/me , ( + K )/me . (8.31) (8.32)

Based on the trial action Strial , Feynmans variational method allows one to obtain an upper bound for the free energy F of the system (at temperature T ) described by the action functional S by minimizing the following function: F = F0 1 S Strial ,

(8.33)

with respect to the variational parameters of the trial action. In this expression, F0 is the free energy of the trial system characterized by the Lagrangian L0 , = 1/(kbT ) is the inverse temperature, and the expectation value S Strial is to be taken with respect to the ground state of this trial system. The evaluation of expression (8.33) is straightforward

175

though lengthy. We nd F = 2 ln 2 sinh 1 2 + 2 ln 2 sinh

2

2 2 i 2 (8.34)

w 2 a2 i coth ln 2 sinh 2 2 i=1 i 2 d2 d2 e d2 /(2D) + I1 I0 e D 2D 2D 1 2

1/Rb
2

/2

dqq |Mq |2
2

d
0

cosh[ (q )( /2)] sinh[ (q )/2] .

q exp 2m e

a2 j
j =1

cosh( j /2)cosh[ j ( /2)] j sinh( j /2)

In this expression, I0 and I1 are Bessel functions of imaginary argument, and

2

D=

me

j =1

a2 j coth ( j /2) , j w 2 2 2 . 2 2 1 2

(8.35)

a1 =

Finally, 1 and 2 are the eigenfrequencies of the trial system, given by 2 1,2 = 1 2 + w2 2 (2 w 2 )2 + 4K/(Mme ) . (8.37)

2 2 1 w ; a2 = 2 2 1 2

(8.36)

Optimal values of the variational parameters are determined by the numerical minimization of the variational functional F as given by expression (8.34).

F.

Melting of the ripplopolaron Wigner lattice

The Lindemann melting criterion [115] states in general that a crystal lattice of objects (be it atoms, molecules, electrons, or ripplopolarons) will melt when the average motion of the objects around their lattice site is larger than a critical fraction 0 of the lattice parameter d. It would be a strenuous task to calculate from rst principles the exact value of the critical fraction 0 , but for the particular case of electrons on a helium surface, we can make use of an experimental determination. Grimes and Adams [116] found that the Wigner lattice melts when = 137 15, where is the ratio of potential energy to the kinetic energy per electron. At temperature T the average kinetic energy in a lattice potential Vlat is Ekin = lat coth 2 176 lat 2kB T , (8.38)

and the average distance that an electron moves out of the lattice site is determined by r2 = me lat coth lat 2kB T = 2Ekin . 2 me lat (8.39)

From this we nd that for the melting transition in Grimes and Adams experiment [116], the critical fraction equals 0 0.13. This estimate is in agreement with previous (empirical) estimates yielding 0 0.1 [117], and we shall use it in the rest of this section. Within the approach of Fratini and Qu emerais [114], the Wigner lattice of (ripplo)polarons melts when at least one of the two following Lindemann criteria are met: r = = R2 cms > 0 , d 2 > 0 . d (8.40) (8.41)

where and Rcms are, respectively, the relative coordinate and the center of mass coordinate of the model system (8.30): if r is the electron coordinate and R is the position coordinate of the ctitious ripplon mass M , this is Rcms = me r + M R ; = r R. me + M (8.42)

The appearance of two Lindemann criteria takes into account the composite nature of (ripplo)polarons. As follows from the physical sense of the coordinates and Rcms , the rst criterion (8.40) is related to the melting of the ripplopolaron Wigner lattice towards a ripplopolaron liquid, where the ripplopolarons move as a whole, the electron together with its dimple. The second criterion (8.41) is related to the dissociation of ripplopolarons: the electrons shed their dimple. The path-integral variational formalism allows us to calculate the expectation values
2 R2 with respect to the ground state of the variationally optimal model system. cms and

We nd R2 cms = w4 2 2 2 2 2 me [w 2 (2 1 + 2 ) 1 2 ] (1 2 ) (8.43)

2 2 4 2 (1 w ) coth( 1 /2)/1

2 2 +4 1 (w 2 ) coth( 2 /2)/2 ,

177

2 =

2 2 3 1 (w 2 ) coth ( 1 /2)

2 2 2 2 2 me (2 1 2 ) (1 w ) (w 2 )

2 2 +3 2 (1 w ) coth( 2 /2) .

(8.44)

Numerical calculation shows that for ripplopolarons in an MEB the inequality 1 w is fullled (w/1 103 to 102 ) so that the strong-coupling regime is realized. Owing to this inequality, we nd from Eqs. (8.43),(8.44) that
2 R2 . cms

(8.45)

So, the destruction of the ripplopolaron Wigner lattice in an MEB occurs through the dissociation of ripplopolarons, since the second criterion (8.41) will be fullled before the rst (8.40). The results for the melting of the ripplopolaron Wigner lattice are summarized in the phase diagram shown in Fig. 21.

178

FIG. 21: The phase diagram for the spherical 2D layer of electrons in the MEB. Above a critical pressure, a ripplopolaron solid (a Wigner lattice of electrons with dimples in the helium surface underneath them) is formed. Below the critical pressure, the ripplopolaron solid melts into an electron liquid through dissociation of ripplopolarons. (From Ref. [109].)

179

For every value of N , pressure p and temperature T in an experimentally accessible range, this gure shows whether the ripplopolaron Wigner lattice is present (points above the surface) or molten (points below the surface). Below a critical pressure (on the order of 104 Pa) the ripplopolaron solid will melt into an electron liquid. This critical pressure is nearly independent of the number of electrons (except for the smallest bubbles) and is weakly temperature dependent, up to the helium critical temperature 5.2 K. This can be understood since the typical lattice potential well in which the ripplopolaron resides has frequencies of the order of THz or larger, which correspond to 10 K. The new phase that we predict, the ripplopolaron Wigner lattice, will not be present for electrons on a at helium surface. At the values of the pressing eld necessary to obtain a strong enough electron-ripplon coupling, the at helium surface is no longer stable against long-wavelength deformations [118]. Multielectron bubbles, with their dierent ripplon dispersion and the presence of stabilizing factors such as the energy barrier against ssioning [119], allow for much larger electric elds pressing the electrons against the helium surface. The regime of N , p, T parameters suitable for the creation of a ripplopolaron Wigner lattice lies within the regime that would be achievable in recently proposed experiments aimed at stabilizing multielectron bubbles [112]. The ripplopolaron Wigner lattice and its melting transition might be detected by spectroscopic techniques [116, 120] probing for example the transverse phonon modes of the lattice [121].

Acknowledgments

I thank S. N. Klimin for discussions in the course of the preparation of the third edition of the Lectures.

180

Appendix A: Optical conductivity of a strong-coupling polaron (S. N. Klimin and J. T. Devreese to be published in Physical Review B, 2013. Figures are published in Ref. [123].) 1. Introduction

The optical conductivity of the Fr ohlich polaron model attracted attention for years [123]. In the regime of weak coupling, the optical absorption of a polaron was calculated using dierent methods, e. g., Greens function method [69], the Low-Lee-Pines formalism [71, 78], perturbation expansion of the current-current correlation function [124]. The strongcoupling polaron optical conductivity was calculated taking into account one-phonon [47] and two-phonon [73] transitions from the polaron ground state to the polaron relaxed excited state (RES). In fact the present work nalizes the project started in Ref. [47]. Using the path integral response formalism, the impedance function of an all-coupling polaron was calculated by FHIP [45] on the basis of the Feynman polaron model [41]. Developing further the FHIP approach, the optical conductivity was calculated in the path-integral formalism at zero temperature [48] and at nite temperatures [74]. In Ref. [125], the extension of the method of Ref. [48] accounting for the polaron damping (for the polaron coupling constant 8) and the asymptotic strong-coupling approach using the Franck-Condon 8) have given reasonable results for the

(FC) picture for the optical conductivity (for

polaron optical conductivity at all values of . The concept of the RES and FC polaron states played a key role in the understanding of the mechanism of the polaron optical conductivity [5, 39, 47, 48, 73, 74]. Recently, the Diagrammatic Quantum Monte Carlo (DQMC) numerical method has been developed [28, 77], which provides accurate results for the polaron characteristics in all coupling regimes. The analytic treatment [48] was intended to be valid at all coupling strengths. However, it is established in [5, 48, 73] that the linewidth of the obtained spectra [48] is unreliable for 7. Nevertheless, the position of the peak attributed to RES in

Ref. [48] is close to the maximum of the polaron optical conductivity band calculated using DQMC up to very large values of (see Fig. 1). An extension of the path-integral approach [48] performed in Ref. [125] gives a good agreement with DQMC for weak and intermediate coupling strengths. In the strong-coupling

181

limit, in Ref. [125] the adiabatic strong-coupling expansion was applied. That expansion, however, is not exact in the strong-coupling limit because of a parabolic approximation [33] for the adiabatic potential. In the present work, the strong-coupling approach of Ref. [125] is extended in order to obtain the polaron optical conductivity which is asymptotically exact in the strong-coupling limit. We develop the multiphonon strong-coupling expansion using numerically accurate in the strong-coupling limit polaron energies and wave functions and accounting for nonadiabaticity.

2.

Optical conductivity

We consider the electron-phonon system with the Hamiltonian written down in the Feynman units ( = 1, the carrier band mass mb = 1, and the LO-phonon frequency LO = 1) 1 p2 1 2 2 iq r + + bq + b+ . (A1) H= + bq bq + q e 2 2 q V q q where r, p represent the position and momentum of an electron, b+ q , bq denote the creation and annihilation operators for longitudinal optical (LO) phonons with wave vector q, and Vq describes the amplitude of the interaction between the electrons and the phonons. For the Fr ohlich electron-phonon interaction, the amplitude of the electron LO-phonon interaction 1 2 2 Vq = q V with the crystal volume V , and the electron-phonon coupling constant . is (A2)

The polaron optical conductivity describes the response of the system with the Hamiltonian (5.2) to an applied electromagnetic eld (along the z -axis) with frequency . This optical response is expressed using the Kubo formula with a dipole-dipole correlation function: n0 1 e Re ( ) = 2

eit dz (t) dz dt,

1 kB T

(A3) , n0 is the

where d = e0 r is the electric dipole moment, e0 is the unit charge, = units of e2 0 becomes Re ( ) =

electron density. In the zero-temperature limit, the optical conductivity (A3) measured in 2

eit fzz (t) dt,

(A4)

182

with the correlation function fzz (t) z (t) z (0) = 0 eitH zeitH z 0 , where |0 is the ground-state wave function of the electron-phonon system. Within the strong-coupling approach, the ground-state wave function is chosen as the product of a trial wave function of an electron 0 phonon subsystem |ph : |0 = 0 applied to the phonon vacuum |ph = U |0ph . with the unitary operator U =e
q b+ (fq bq fq q) ,

(A5)

( e)

and of a trial wave function of a

(e)

|ph .

(A6)

The phonon trial wave function is written as the strong-coupling unitary transformation (A7)

(A8)

U 1 HU takes the and the variational parameters {fq }. The transformed Hamiltonian H form =H 0 + W H with the terms
2 0 = p + H 2

(A9)

|fq |2 + Va (r ) +

b+ q bq +
q

1 2

(A10) (A11)

W =
q

+ Wq bq + Wq bq .

Here, Wq are the amplitudes of the renormalized electron-phonon interaction Wq = 2 2 iqr e q , q V (A12)

where q is the expectation value of the operator eiqr with the trial electron wave function 0
( e)

: q = 0
(e)

eiqr 0

(e)

(A13)

and Va (r ) is the self-consistent potential energy for the electron, 4 2 q eiqr . V a (r ) = 2V q q 183

(A14)

Averaging the Hamiltonian (A9) with the phonon vacuum |0 and with the trial electron wave function |0 , we arrive at the following variational expression for the ground-state energy E0 = 0 |H | 0 = 0 p2 0 2 +
q

|fq |2 (A15)

Vq fq q + Vq fq q ,

After minimization of the polaron ground-state energy (A15), the parameters fq acquire their optimal values fq = Vq q . The ground-state energy with {fq } given by Eq. (A16) takes the form E0 = 0 p2 0 2
q

(A16)

|Vq |2 |q |2 .

(A17)

With the strong-coupling Ansatz (A6) for the polaron ground-state wave function and after the application of the unitary transformation (A8), the correlation function (A5) takes the form fzz (t) = 0ph 0 eitH zeitH z 0

0ph .

(A18)

This correlation function can be expanded using a complete orthogonal set of intermediate states |j and the completeness property:
j

|j

j | = 1.

(A19)

In the present work, we use the intermediate basis of the Franck-Condon (FC) states. The FC states correspond to the equilibrium phonon conguration for the ground state. Thus the FC 0 . Further on, the FC wave wave functions are the exact eigenstates of the Hamiltonian H functions are written in the spherical-wave representation as |n,l,m = Rn,l (r ) Yl,m (, ) where Rn,l (r ) are the radial wave functions, and Yl,m (, ) are the spherical harmonics, l is the quantum number of the angular momentum, m is the z -projection of the angular momentum, and n is the radial quantum number5 . The energy levels for the eigenstates of 0 are denoted En,l . the Hamiltonian H
5

In this classication, the ground-state wave function is |0,0,0 |0 .

184

Using (A19) with that complete and orthogonal basis , we transform (A18) to the expression fzz (t) =
n,l,m, n ,l ,m , n ,l ,m

n,l,m |z | n ,l ,m

n ,l ,m |z | 0

0ph

0 eitH n,l,m

n ,l ,m eitH n ,l ,m

0ph .

(A20)

So far, the only approximation made in (A20) is the strong-coupling Ansatz for the polaron ground-state wave function. However, in order to obtain a numerically tractable expression for the polaron optical conductivity, an additional approximation valid in the strong-coupling limit must be applied to the matrix elements of the evolution operator eitH given by formula (A9). According with the Hamiltonian of the electron-phonon system H to Ref. [126], in the strong-coupling limit, the matrix elements of the Hamiltonian of the electron-phonon system between states corresponding to dierent energy levels are of order of magnitude 4 . Therefore in the strong-coupling regime these matrix elements can be neglected; this is called the adiabatic or the Born-Oppenheimer (BO) approximation [126], because of its strict analogy with the Born-Oppenheimer adiabatic approximation in the theory of molecules and crystals ([127], p. 171). Consequently, in the further treatment we neglect the matrix elements n,l,m eitH n ,l ,m for the FC states with dierent energies, En,l = En ,l . The same scheme was used in the theory of the multi-phonon optical processes for bound electrons interacting with phonons [21, 128]. Strictly speaking, the summation over the excited polaron states in Eq. (A20) must involve the transitions to both the discrete and continuous parts of the polaron spectrum. A transition to the states of the continuous spectrum means that the electron leaves the polaron potential well. Therefore these transitions can be attributed to the polaron dissociation. The transitions to the continuous spectrum are denitely beyond the adiabatic approximation. As shown in Ref. [21], the transition probability to the states of the continuous spectrum is very small compared with the transition probability between the ground and the rst excited state (which belongs to the discrete part of the polaron energy spectrum). We neglect here the contribution to the polaron optical conductivity due to the transitions to the continuous spectrum. The matrix elements neglected within the adiabatic approximation correspond to the transitions between FC states with dierent energies due to the electron-phonon interac185

tion. Hence these transitions can be called non-adiabatic. The adiabatic approximation is related to the matrix elements of the evolution operator eitH . On the contrary, the matrix elements of the transitions between dierent FC states for the electric dipole moment are, in general, not equal to zero. Moreover, these transitions can be accompanied by the emission of phonons. The electron FC wave functions constitute a complete orthogonal set. However, the corresponding phonon wave functions can be non-orthogonal because of a different shift of phonon coordinates for dierent electron states. This makes multi-phonon transitions possible [128]. It is important to note that in our treatment we neglect only the non-adiabatic transitions between the electron states with dierent energies. On the contrary, the transitions within one and the same degenerate level can be non-adiabatic. This internal non-adiabaticity (i. e., the non-adiabaticity of the transitions within one and the same degenerate level) is taken into account in the subsequent treatment. It is useful to stress the dierence between the strong-coupling Ansatz and the adiabatic approximation. The strong-coupling Ansatz consists of the choice of the trial variational ground state wave function for the electron-phonon system in the factorized form (A6). The adiabatic approximation means neglecting the matrix elements of the evolution operator between internal polaron states with dierent energies. These two approximations are not the same, but they both are valid in the strong-coupling regime and consistent with each other. The correlation function (A20) is transformed in the following way. The exponents eitH and eitH are disentangled: eitH = eitH0 T exp i eitH = eitH0 T exp i
0 t t

dsW (s) ,
0

(A21) (A22)

dsW (s)

where W (s) is the renormalized electron-phonon interaction Hamiltonian W in the interaction representation, W (s) eisH0 W eisH0 .

(A23)

186

This gives us the result fzz (t) =

n,l,m, n ,l ,m , n ,l ,m

n,l,m |z | n ,l ,m
t

n ,l ,m |z | 0 eit(E0 En ,l )

0ph

0 T exp i
0

dsW (s)
t

n,l,m n ,l,m 0ph . (A24)

n ,l ,m T exp i

dsW (s)
0

Within the adiabatic approximation, the optical conductivity is simplied. The full details of the derivation are described in the Appendix A. First, using the selection rules for and the adiabatic the dipole matrix elements, the spherical symmetry of the Hamiltonian H approximation, the correlation function (A24) is reduced to the form fzz (t) =
n t

Dn ein,0 t 0ph T exp i dsW (s)

0

n,1,0

0ph

n,1,0

(A25)

where n,0 is the FC transition frequency n,0 En,1 E0 , and Dn is the squared modulus of the dipole transition matrix element Dn = | 0 |z | n,1,0 |2 . (A27) (A26)

Within the adiabatic approximation, the partial (with the electron wave functions) averaging of the operator T-exponent in (A25) can be exactly performed (see details in Appendix A). As a result, the optical conductivity is transformed to the expression Re ( ) = 6 Dn
n

ei(n,0 )t
t

0ph Tr T exp i

dsW(n) (s)
0

0ph dt.

(A28)

The T-exponent in (A28) contains the nite-dimensional matrix W(n) (s) depending on the phonon coordinates: Wk,l,m
(n) m1 ,m2

= n,1,m1 |Wk,l,m| n,1,m2 187

(A29)

where Wk,l,m are the amplitudes of the electron-phonon interaction in the basis of spherical wave functions. Because the kinetic energy of the phonons is of order 4 compared to the leading term of the Hamiltonian [126], we neglect this kinetic energy in the present work, because the treatment is related to the strong-coupling regime. As a result, Qk,l,m commute with the 0 , so that in (A28), W(n) (s) = W(n) . Furthermore, in a nite-dimensional Hamiltonian H 0 are the same. basis {|n,l,m } for a given level (n, l), all eigenvalues of the Hamiltonian H Therefore the T-exponent entering (A28) in that nite-dimensional basis turns into a usual exponent. As a result, the strong-coupling polaron optical conductivity (A28) takes the form Re ( ) = 6 Dn
n

ei(n,0 )t 0ph Tr exp iW(n) t 0ph dt.

(A30)

The matrix interaction Hamiltonian (A29) depends on the phonon coordinates, and the matrices Wk,l,m with dierent m for one and the same degenerate energy level do not commute with each other. According to the Jahn Teller theorem [129], for a degenerate level there does not exist a unitary transformation which simultaneously diagonalizes all matrices Wk,l,m in a basis that does not depend on the phonon coordinates. The manifestations of that theorem are attributed to the Jahn Teller eect. Therefore, because we neglect the non-commutation of the matrices Wk,l,m, the Jahn Teller eect is omitted. In fact, neglecting the Jahn Teller eect is not necessary. The averaging in Eq. (A30) is performed exactly using the eective phonon modes similarly to Ref. [131] (see the details in Appendix B). As a result, we arrive at the following expression for the strong-coupling polaron optical conductivity Re ( ) = 2 3 dx0 dx1 dx2 ( n) a0 3 1 2 x2 exp x2 0 + m + ym + 2 m=1,2 j =1
n (n) (n) (n)

Dn

dy1

dy2
(n) a2

n,0

2 5 ( n) 2

j (Q2 )

a0

Here, j (Q2 ) are the eigenvalues for the matrix interaction Hamiltonian, which are explicitly
( n) ( n)

. (A31)

determined in the Appendix B by the formula (A74). The coecients a0 and a2 are given by (A65) and (A66), respectively. The polaron optical conductivity given by the expression (A31), is in fact an envelope of the multiphonon polaron optical conductivity band with 188

the correlation function (A28) provided by the phonon-assisted transitions from the polaron ground state to the polaron RES. This result is consistent with Ref. [47], where the same paradigm of the phonon-assisted transitions to the polaron RES was exploited, but the calculation was limited to the one-phonon transition. In order to reveal the signicance of the Jahn Teller eect for the polaron, we alternatively calculate 0ph Tr exp iW(n) t 0ph neglecting the non-commutation of the matrices Wk,l,m, as described in the Appendix B. 2. The resulting expression for the polaron optical conductivity is much simpler than formula (A31) and is similar to the expression (3) of Ref. [125]: Re ( ) =
n (n)

2s

D exp (n) n

( n,0 )2 2s
(n)

(A32)

with the parameter (often called the Huang-Rhys factor)

(n) s =

1 ( n) a 2 0

1 ( n) a 4 2

(A33)

The strong-coupling electron energies and wave functions in Eq. (A28) can be calculated using dierent approximations. For example, within the Landau-Pekar (LP) approximation [33], the trial wave function |0 is chosen as the ground state of a 3D oscillator. Within the Pekar approximation [21], |0 is chosen in the form |0 (r ) = Cear 1 + ar + br 2 (A34)

with the variational parameters a and b. Finally, the trial ground state wave function can be determined numerically exactly following Miyake [38] (see also [130], Chap. 5.22). Within the LP approximation, formula (A32) reproduces the polaron optical conductivity obtained in Ref. [125]. In the LP approximation, the matrix elements 0 |z | n,1,0 are dierent from zero only for n = 1, i. e. only for the 1s 2p transition. Beyond the LP approximation, also the transitions to other excited states are allowed because of the nonparabolicity of the self-consistent potential Va (r ). The use of exact strong-coupling wave functions, instead of the LP wave functions, may signicantly inuence the optical conductivity. In the present treatment we use the numerically exact electron energies and wave functions of both ground and rst excited states according to Ref. [38]. The FC transition energies n,0 to leading order of the strong-coupling approximation are determined according to (A26). In order to 189

account for the corrections of the FC energy with accuracy up to 0 , we add to n,0 the correction FC 3.8 from Ref. [125]. Because we use the numerically accurate strongcoupling wave functions and energies corresponding to Miyake [38], the formula (A24) is asymptotically exact in the strong-coupling limit, at least in its leading term in powers of 2 .

3.

Results and discussion

In Figs. 2 to 3, we have plotted the polaron optical conductivity spectra calculated for dierent values of the coupling constant . The optical conductivity spectra calculated within the present strong-coupling approach taking into account the Jahn Teller eect are shown by the solid curves. The optical conductivity derived neglecting the Jahn Teller eect is shown by the dashed curves. It is worth mentioning that there is little dierence in the optical conductivity spectra between those calculated with and without the Jahn Teller eect. The optical conductivity obtained in Ref. [125] with the Landau-Pekar (LP) adiabatic approximation is plotted with dash-dotted curves. The full dots show the numerical Diagrammatic Quantum Monte Carlo (DQMC) data [77, 125]. The FC transition frequency for the transition to the rst excited FC state 1,0 FC and the RES transition frequency RES are explicitly indicated in the gures. The polaron optical conductivity spectra calculated within the present strong-coupling approach are shifted to lower frequencies with respect to the optical conductivity spectra calculated within the LP approximation of Ref. [125]. This shift is due to the use of the numerically accurate strong coupling energy levels and wave functions of the internal polaron states, and of the numerically accurate self-consistent adiabatic polaron potential. According to the selection rules for the matrix elements of the electron-phonon interaction, there is a contribution to the polaron optical conductivity from the phonon modes with angular momentum l = 0 (s-phonons) and with angular momentum l = 2 (d-phonons). The s-phonons are fully symmetric, therefore they do not contribute to the Jahn Teller eect, while the d-phonons are active in the Jahn Teller eect. The contribution of the d-phonons to the optical conductivity spectra is not small compared to the contribution of the s-phonons. However, the distinction between the optical conductivity spectra calculated with and without the Jahn Teller eect is relatively small. 190

For = 8 and = 8.5, the maxima of the polaron optical conductivity spectra, calculated within the present strong-coupling approach are positioned to the low frequency side of the maxima of those calculated using the DQMC method. The agreement between our strongcoupling polaron optical conductivity spectra and the numerical DQMC data improves with increasing alpha. This is in accordance with the fact that the present strong-coupling approach for the polaron optical conductivity is asymptotically exact in the strong-coupling limit. The total polaron optical conductivity must satisfy the sum rule [80]
0

Re ( ) d =

. 2

(A35)

In the weak- and intermediate-coupling regimes at T = 0, there are two contributions to the left-hand side of that sum rule: (1) the contribution from the polaron optical conductivity for > LO and (2) the contribution from the central peak at = 0, which is proportional to the inverse polaron mass [80]. In the asymptotic strong-coupling regime, the inverse to the polaron mass is of order 4 , and hence the contribution from the central peak to the polaron optical conductivity is beyond the accuracy of the present approximation (where we keep the terms 2 and 0 ). As discussed above, in the present work the transitions from the ground state to the states of the continuous part of the polaron energy spectrum are neglected. Therefore the integral over the frequency [the left-hand side of (A35)] for the optical conductivity calculated within the present strong-coupling approximation can be (relatively slightly) smaller than /2. The relative contribution of the transitions to the continuous part of the polaron spectrum, c , can be therefore estimated as 2 c 1
0

Re ( ) d,

(A36)

where the right-hand side is obtained by a numerical integration of Re ( ) calculated within the present strong-coupling approach. This numeric estimation shows that for > 8, c < 0.01. Moreover, with increasing , the relative contribution of the transitions to the continuous part of the polaron spectrum falls down. This conrms the accuracy of the present strong-coupling approach. In Refs. [93, 132], the optical conductivity of a strong-coupling polaron was calculated assuming that in the strong-coupling regime the polaron optical response is provided mainly 191

by the transitions to the continuous part of the spectrum (these transitions are called there the polaron dissociation). This concept is in contradiction both with the early estimation by Pekar [21] discussed above and with the very small weight of those transitions shown in Fig. 4. The approach of Ref. [93] in fact takes into account only a small part of the strongcoupling polaron optical conductivity namely, the high-frequency tail of the optical conductivity spectrum. When comparing the polaron optical conductivity spectra calculated in the present work with the DQMC data [77, 125], we can see that the present approach, with respect to DQMC, underestimates the high-frequency part of the polaron optical conductivity. This dierence, however, gradually diminishes with increasing , in accordance with the fact that the present method is an asymptotic strong-coupling approximation. Because the optical conductivity spectra calculated in the present strong-coupling approximation using the expressions (A31) and (A32) represent the envelopes of the RES peak with the multi-phonon satellites, the separate peeks are not explicitly seen in those spectra. The FC and RES peaks are indicated in the gures by the arrows. The FC transition frequency 1,0 in the strong-coupling case is positioned close to the maximum of the polaron optical conductivity band (both calculated within the present approach and within DQMC). The RES transition frequency is positioned one LO below the onset of the LO-sidebands. Note that the strong-coupling polaron optical conductivity derived in Refs. [133] contains only the zero-phonon (RES) line and no phonon satellites at all. In contrast, in the present calculation, the maximum of the polaron optical conductivity spectrum shifts to higher frequencies with increasing , so that the multiphonon processes invoking large number of phonons become more and more important, in accordance with predictions of Refs. [47, 48]. It is worth noting the following important point: the maximum of the polaron optical conductivity band can be hardly interpreted as a broadened transition to an FC state on the following reasons. Formula (A28) describes a set of multi-phonon peaks. In the simplifying approximation which neglects the Jahn Teller eect (see Ref. [125]), those peaks are n,0 + k , where k is the number of emitted phonons and positioned at the frequencies = n,0 do not coincide with the FC is the frequency of the zero-phonon line. The frequencies transition frequencies but are determined by n,0 = n,0 (n) , s 192 (A37)

describes the energy shift due to lattice relaxation. The (n) n,0 + k physical meaning of the parameters s obviously implies that the peaks at = should be attributed to transitions to the RES with emission of k phonons. So, the so-called FC transition is realized as the envelope of a series of phonon sidebands of the polaron RES but not as a transition to the FC state. The account of the Jahn-Teller eects in general makes the multiphonon peak series non-equidistant, but it changes nothing in the concept of the internal polaron states which is discussed above.

where the Huang-Rhys factor s

(n)

4.

Conclusions

We have derived the polaron optical conductivity which is asymptotically exact in the strong-coupling limit. The strong-coupling polaron optical conductivity band is provided by the multiphonon transitions from the polaron ground state to the polaron RES and has the maximum positioned close to the FC transition frequency. With increasing the electron-phonon coupling constant , the polaron optical conductivity band shape gradually tends to that provided by the Diagrammatic Quantum Monte Carlo (DQMC) method. This agreement demonstrates the importance of the multiphonon processes for the polaron optical conductivity in the strong-coupling regime. The obtained polaron optical conductivity with a high accuracy satises the sum rule [80], what gives us an evidence of the fact that in the strong-coupling regime the dominating contribution to the polaron optical conductivity is due to the transitions to the internal polaron states, while the contribution due to the transitions to the continuum states is negligibly small. Accurate numerical results, obtained using DQMC method [77], modulo the linewidths for suciently large and the analytically exact in the strong-coupling limit polaron optical conductivity of the present work, as well as the analytical approximation of Ref. [125] conrm the essence of the mechanism for the optical absorption of Fr ohlich polarons, which were proposed in Refs. [5, 48].

193

Appendix 1. Correlation function

The dipole-dipole correlation function fzz (t) given by (A24) is further simplied within the adiabatic approximation and using the selection rules for the dipole transition matrix elements and the symmetry properties of the polaron Hamiltonian. First, according to the selection rules, the matrix element 0 |z | n,l,m is n ,l ,m |z | 0 = l ,1 m ,0 n ,1,0 |z | 0 (A38)

Second, the interaction Hamiltonian W (and hence, also the evolution operator which involves W ) is a scalar of the rotation symmetry group. The matrix elements n,l,m |W (s)| n,l ,m for l = l and m = m are then exactly equal to zero. Therefore, , we obtain in the adiabatic approximation and due to the symmetry of the Hamiltonian H the relations
t

0 T exp i
0

dsW (s)

n,l,m
t

n,0 l,0 m,0 0 T exp i

t

dsW (s)
0

0 ,

(A39)

n ,l ,m T exp i n ,n l ,l

dsW (s)
0 t

n ,l ,m dsW (s)
0

n ,l ,m T exp i
t

n ,l ,m

(A40)

Furthermore, because the ground state 0 is non-degenerate, we nd that 0 T exp i dsW (s)
0

1,

because within the adiabatic approximation, for any n 1 the averages 0 |W n | 0 = 0. The correlation function (A24) using (A38) to (A40) takes the form fzz (t) =
n t

Dn ein,0 t 0ph T exp i dsW (s)

0

n,1,0

0ph

n,1,0

(A41)

with the squared matrix elements of the dipole transitions Dn | n,1,0 |z | 0 |2 = 1 3 194
0 2

Rn,1 (r ) R0,0 (r ) r 3 dr

(A42)

and the FC transition frequencies n,0 En,1 E0 . (A43)

Further on, the interaction Hamiltonian is expressed in terms of the complex phonon coordinates Qk : W = bk + b+ Qk = k 2

2
k

Wk Qk ,

(A44)

Here, we use the spherical-wave basis for phonon modes: k,l,m (r) (1)
m|m| 2

k,l (r ) Yl,m (, ) ,

(A45)

where the radial part of the basis function is expressed through the spherical Bessel function jl (kr ): k,l (r ) = The factor (1)
m|m| 2

2 R

1/2

k jl (kr ) , R =

3V 4

1/3

(A46)

is chosen in order to full the symmetry property k,l,m (r) = k,l,m (r) .

In the spherical-wave basis, the interaction Hamiltonian is W = with the complex phonon coordinates Qk,l,m = and with the interaction amplitudes Wk,l,m = 2 2 (k,l,m (r) k,l,m) , k,l,m 0 |k,l,m| 0 . k (A49) bk,l,m + b+ k,l,m 2 (A48) 2
k,l,m

Wk,l,m Qk,l,m,

(A47)

The dipole-dipole correlation function (A41) is then fzz (t) =

n

Dn ein,0 t n,1,0 0ph T exp i 2

t

ds
0 k,l,m

Wk,l,m (s) Qk,l,m (s)

0ph

n,1,0

. (A50)

195

The operators Wk,l,m (s) in (A50) are equivalent to the (2l + 1)-dimensional matrices Wk,l,m determined in the basis of the level (n, l). The matrix elements of these matrices are Wk,l,m
(n)

(n)

m1 ,m2

= n,1,m1 |Wk,l,m| n,1,m2 .

(A51)

In these notations, fzz (t) given by (A50) can be written down as

t

fzz (t) =
n

Dn ein,0 t

0ph

T exp i

dsW(n) (s)
0 0,0

0ph

(A52)

where W(n) is the matrix electron-phonon interaction Hamiltonian expressed through the phonon complex coordinates in the spherical-wave representation as follows: W(n) = 2
k,l,m (n) 0. Here, Wk,l,m is a (3 3) matrix in a basis of a level (n, l)l=1 of the Hamiltonian H

Wk,l,mQk,l,m .

( n)

(A53)

Because W(n) is a scalar of the rotation group, we can replace the diagonal matrix element

of the T-exponent in (A52) with the trace in the aforesaid-nite-dimensional basis. As a result, we obtain for the polaron optical conductivity (A4) with (A52) the expression Re ( ) = 6 Dn
n

ei(n,0 )t
t

0ph Tr T exp i
5. Appendix 2. Eective phonon modes

dsW(n) (s)
0

0ph dt.

(A54)

In order to perform the averaging in Eq. (A30) analytically, we introduce the eective phonon modes Q0,0 and Q2,m similarly to Ref. [131]. The Hamiltonian W(n) in terms of these eective phonon modes is expressed as W ( n) = 2
l,m

(n) Ql,m W l,m

(A55)

(n) (depending on the vibration coordinates Ql,m ) are explicitly given where the matrices W l,m by the expressions (cf. Ref. [131]),
2

W(n) = a0 IQ0,0 + a2 196

(n)

( n) m=2

Bm Q2,m

(A56)

with the matrices Bj 1 0 0 1 B0 = 0 2 0 , 2 5 0 0 1 0 0 0 1 3 + B1 = B1 = 1 0 0 , 2 5 0 1 0 0 0 0 3 + B2 = B2 = 0 0 0 . 10 1 0 0 in Eq. (A56) are 2

k (n) a2

(A57)

(A58)

(A59)

The coecients a0

( n)

and a2
(n) a0

( n)

1 k2

k,0

n,1

k,0
1/2

2 0,0

1/2

(A60)

4 2
k

1 k,2 k2

2 n,1

(A61)

Here k,l is the radial part of the basis function expressed through the spherical Bessel function jl (kr ): k,l (r ) = 2 R
1/2

k jl (kr ) , R =
n,l 0

3V 4

1/3

(A62)

V is the volume of the crystal, and f (r ) f (r )

n,l

is the average
2 f (r ) Rn,l (r ) r 2 dr.

(A63)

The normalization of the phonon wave functions corresponds to the condition

R

k,l (r ) k,l (r ) r 2 dr = k,k .

0

(A64)
( n)

After the straightforward calculation using (A64), we express the coecients a0 through the integrals with the radial wave functions:
( n) a0

and a2

( n)

= =

2 2 8 2 5
0

dr
0

dr r (r )
r

1/2 2 Rn, 1 (r )

2 R0 ,0

(r )
1/2

2 Rn, 1

(r )

2 R0 ,0

(r )

, (A65) (A66)

( n) a2

dr
0

dr

(r

) 2 2 Rn,1 (r ) Rn, 1 (r ) r 197

a.

Exact averaging

Let us substitute the matrix interaction Hamiltonian (A56) to the dipole-dipole correlation function (A30), what gives us the result fzz (t) = 1 3 Dn ein,0 t
n

a2 ( n) 0ph exp ita0 Q0 Tr exp it V (Q2 ) 2 5

( n)

0ph

. (A67)

Here, we use the matrix depending on the phonon coordinates, V (Q2 ) 2 5 whose explicit form is Q 2,0 V (Q2 ) = 3Q2,1 6Q2,2 3Q2,1 6Q2,2 2Q2,0 3Q2,1 . 3Q2,1 Q2,0
2

Bm Q2m ,
m=2

(A68)

(A69)

The matrix V (Q2 ) is analytically diagonalized. The equation for the eigenvectors | (Q2 ) and eigenvalues (Q2 ) of V (Q2 ) is V (Q2 ) | (Q2 ) = (Q2 ) | (Q2 ) . The eigenvalues are found from the equation det (V (Q2 ) (Q2 ) I) = 0. We make the transformation to the real phonon coordinates, Q2,0 x0 , Q2,m (A71) (A70)

xm + iym , 2

Q2,m = Q 2,m =

xm iym . 2

Five variables x0 , x1 , x2 , y1 , y2 are the independent real phonon coordinates. The l.h.s. of Eq. (A71) is expressed in terms of these coordinates as det (V (Q2 ) (Q2 ) I) = 3 + 3p + 2q with the coecients
2 2 2 2 p = x2 0 + x1 + x2 + y1 + y2 , 3 3 3 2 2 2 2 2 3 x2 x2 q = x0 + x0 x1 + y1 + 1 y1 3x0 x2 + y2 + 3 3x1 y1 y2 . 2 2

(A72)

198

So, we have the cubic equation for : 3 3p 2q = 0. Because the matrix V (Q2 ) is Hermitian, all its eigenvalues are real. Therefore, (otherwise, sin (3) is not real). Herefrom, we have three explicit eigenvalues: q 1 1 (Q2 ) = 2 p sin + arcsin 3/2 , 3 3 p q 1 arcsin , 2 (Q2 ) = 2 p sin 3 p3/2 q 1 3 (Q2 ) = 2 p sin . arcsin 3 3 3 p /2 (A73)
|q | p3/2

(A74)

The trace in (A67) is invariant with respect to the choice of the basis. Consequently, after the diagonalization fzz (t) takes the form 1 fzz (t) = 3
3

Dn e
n

in,0 t j =1

0ph exp it

(n) a0 Q0

a2 + j (Q2 ) 2 5

(n)

0ph

(A75)

After inserting fzz (t) given by (A75) into (A4), the integration over time gives the delta function multiplied by 2 , and we arrive at the result Re ( ) = 3
3

Dn
n j =1

0ph

n,0

( n) a0 Q0

a2 j (Q2 ) 2 5

(n)

0ph

(A76)

The ground-state wave function for the eective phonon modes is |0ph 0 (Q) = 0 (Q0 ) 0 (Q2 ) . 0 (Q0 ) is the one-oscillator ground-state wave function: 0 (Q0 ) = 1/4 exp
(0) (0) (0) (2)

(A77)

Q2 0 2

(A78)

The ground-state wave function of phonons with l = 2 is: 0 (Q2 ) = 5/4 exp
(2)

1 2

x2 0 +
m=1,2

2 x2 m + ym

(A79)

The phonon ground-state wave function (A77) is then 0 (Q) = 1 3/2 exp 1 2 x2 0 +
m=1,2 2 2 x2 m + ym + Q0

(A80)

199

With these phonon wave functions, Eq. (A76) results in the following expression for the polaron optical conductivity Re ( ) = 2 3 dx0 dx1 dx2 ( n) a0 3 1 2 x2 exp x2 0 + m + ym + 2 m=1,2 j =1
n 2 5 ( n) 2
(n) a2

Dn

dy1

dy2
2

n,0

j (Q2 )

a0

b.

Averaging neglecting the Jahn-Teller eect

. (A81)

In order to perform the phonon averaging explicitly, we disentangle the exponent (n) Ql,m as follows. exp it 2 l,m W l,m exp it 2 T exp i (n) Ql,m W l,m
l,m t

= exp it
l ,m

l,m

(n) b+ W l,m l,m

l ,m

ds
0 l,m

is

(n) W b+ l ,m l ,m

(n) bl,m eis W l,m

(n) W b+ l ,m l ,m

(A82)

(n) we nd that Neglecting non-commutation of matrices W l,m e

l,m is
l ,m

(n) W b+ l ,m l ,m

(n) bl,m eis W l,m (n) . (n) W W l,m l,m

l ,m

(n) W b+ l ,m l ,m

=
l,m

(n) bl,m is W l,m

(A83)

l,m

(n) W (n) in the basis (l, m) for a denite n is proportional to the unity W l,m l,m (n) Ql,m is matrix. Therefore, exp it 2 l,m W l,m The sum
l,m

eit

l,m

(n) Ql,m W l,m (n) b+ it W l,m l,m

= eit that gives us the result

l,m

l,m

(n) bl,m t2 W l,m 2

l,m

(n) W (n) W l,m l,m

(A84)

0ph eit

l,m

(n) Ql,m W l,m

0ph = e 2

t2

l,m

(n) W (n) W l,m l,m

(A85)

(n) , the matrix sum takes the form Using the explicit formulae for the matrices W l,m (n) W (n) = (n) I W s l,m l,m 200 (A86)

l,m

with the parameter

(n) s =

1 ( n) a 2 0

1 ( n) a 4 2

(A87)

Using (A86), the optical conductivity (A30) is transformed to the expression Re ( ) =

n

( n,0 )2 Dn exp 2 Sn 2 Sn

(A88)

201

Figures to Appendix A

30 25 20

Peak positions from DQMC Peak positions from DSG

peak/LO

15 10 5 0 0 3 6 9 12 15

Fig. 1
Frequency of the main peak in the optical conductivity spectra calculated within the model of Ref. [48] (red dots) and the main-peak energy extracted from the DQMC data [77, 125] (black squares).

202

Re () [in units n0e /(mbLO)]

0.4

=8

Present approach Neglecting JT LP DQMC

0.3

0.2

0.1

0.0 0 0.4 3 6 9 12 15 18

Re () [in units n0e /(mbLO)]

= 8.5
0.3

Present approach Neglecting JT LP DQMC

0.2

0.1

0.0 0 3 6 9 12 15 18

(in units LO)

Fig. 2

The strong-coupling polaron optical conductivity calculated within the rigorous strong-coupling approach of the present work (black solid curves), within the present approach but neglecting the dynamic Jahn-Teller eect (red dashed curves), within the adiabatic approximation of Ref. [125] (blue dot-dashed curves), and the numerical Diagrammatic Monte Carlo data (full dots) for = 8 and 8.5.

203

Re () [in units n0e /(mbLO)]

0.3

Present approach Neglecting JT LP DQMC

=9

0.2

0.1

0.0 0 3 6 9 12 15 18 Present approach Neglecting JT LP DQMC

= 13

Re () [in units n0e /(mbLO)]

0.20

0.15

0.10

0.05

0.00 10 0.20

15 Present approach Neglecting JT LP DQMC

20

25

30

= 15

Re () [in units n0e /(mbLO)]

0.15

0.10

0.05

0.00 15

20

25

30

35

40

(in units LO)

Fig. 3

The strong-coupling polaron optical conductivity calculated within the rigorous strong-coupling approach of the present work (black solid curves), within the present approach but neglecting the dynamic Jahn-Teller eect (red dashed curves), within the adiabatic approximation of Ref. [125] (blue dot-dashed curves), and the numerical Diagrammatic Monte Carlo data (full dots) for = 9, 13 and 15.

204

0.012

0.008

c
0.004 0.000

10

12

14

Fig. 4
Relative contribution of the transitions to the continuum polaron states to the zeroth frequency moment of the strong-coupling polaron optical conductivity as a function of the coupling constant . .

Appendix B: Feynmans path-integral polaron treatment approached using timeordered operator calculus [S. N. Klimin and J. T. Devreese, Solid State Communications 151, 144 (2011)]

Several studies have been devoted to the search of a Hamiltonian formalism equivalent to Feynmans path integral approximation to polaron theory. Bogolubov [136] reproduced the Feynman result for the polaron free energy [41] using time-ordering T-products . Yamazaki [137] introduced two kinds of auxiliary vector elds to derive Feynmans ground state polaron energy expression with the operator technique, however he found no proof of the variational nature of this result. Cataudella et al. [138] formally re-obtained Feynmans polaron ground205

state energy expression by introducing additional degrees of freedom, but again their result could not be proved to constitute an upper bound for the polaron ground state energy. The study of the excited polaron states is of interest i. a. for its application to the polaron response properties. In [45] a path-integral based response-formalism was introduced that was applied to derive polaron optical absorption spectra in [5]. The results for the polaron response obtained in [5] were re-derived with a Hamiltonian technique (Mori- formalism) in [74]. To the best of our knowledge, no explicit description of the polaron excited states has been derived within the all coupling- Feynman approach. Only for the limiting cases of weak and strong coupling approximations (and for a 1D-model system) such excitation spectra were derived [37, 39, 123]. In principle, the spectrum of the polaron excited states can be derived indirectly using a Laplace transform of the nite-temperature partition function. However, it is not clear how to realize this program in practice. The polaron excitation spectrum is interesting by itself. E.g. the existence and the nature of relaxed excited states, Franck-Condon states, scattering states is understood from the mathematical structure of corresponding eigenstates. In the present letter we rst present a re-derivation of the original Feynman variational path integral polaron model [41] for the ground state, using a Hamiltonian formalism, and we do provide a proof of the upper bound nature of the obtained ground state energy. Furthermore, using Feynmans (Hamiltonian-) time-ordered operator calculus (and an ad hoc unitary transformation) we obtain explicitly and for the rst time the excited polaron states that correspond to the Feynman polaron model. The novelty of the present approach consists (a) in the direct calculation of the energies and the lifetimes of the excited polaron states (within a Hamiltonian all-coupling approach developed in this work equivalent to the Feynman path integral polaron model) and (b) in the extension of the Feynman variational technique to non-parabolic trial potentials. Although the time-ordered operator calculus is formally equivalent to the path-integral formalism, it is not obvious how to directly calculate the excited polaron states using path integrals. The present work, formulated with the (Hamiltonian) time ordered operator calculus, thus provides an (equivalent) tool complementary with respect to the Feynman path integral 206

approach to the polaron, to study the polaron problem. Additionally we directly study the excited polaron states. Consider an electron-phonon system with the Fr ohlich Hamiltonian

H= Hph Heph

p2 + Hph + Heph , 2 1 = a+ , q aq + 2 q 1 2 2 iq r = aq + a+ . q e q V q

(B1) (B2) (B3)

Here, the Feynman units are used: LO = 1.

= 1, the band mass mb = 1, the LO-phonon frequency

The polaron partition function after exact averaging over phonon states is

Zpol = Tr T exp where =

1 kB T

p2 [r ] d + 2

(B4)

[r ] in the the time-ordered operator . The inuence phase of the phonons

calculus has the same form as in the path-integral representation. The polaron free energy is determined as 1 Fpol = ln Zpol . mass mf through an attractive potential Vf : Htr = p2 p2 f + + V f (r r f ) . 2 2mf (B6) (B5)

The trial Hamiltonian describes the electron interacting with a ctitious particle of the

The trial potential Vf is, in general, non-parabolic. The parabolic potential with frequency parameter w corresponds to the Feynman polaron model. Consider the extended partition function of the electron-phonon system Zext = Zf Zpol where Zf is the partition function of a ctitious particle,

(B7)

Zf Tr T exp 207

d Hf,
0

(B8)

with Hamiltonian Hf =

p2 f + V f (r f ) . 2mf

(B9)

The polaron free energy is expressed as the dierence Fpol = Fext Ff , (B10)

where Ff is the free energy of the ctitious particle conned to the potential V (rf ). The free energies Fext and Ff are determined similarly to (B5), with corresponding partition functions. In the zero-temperature limit, the free energies Fpol , Fext and Ff become, respectively, the
0 0 0 ground-state energies Epol , Eext and Ef .

The key element of the present approach is the unitary transformation U = eipf r . (B11)

Application of this canonical transformation results in the transformed extended Hamil tonian Hext = UHext U 1 , Hext =

p2 (p + pf )2 f + + V f (r f r ) 2 2mf (B12)

+ Hph + Heph . This Hamiltonian can be represented as a sum of an unperturbed Hamiltonian H0 Htr + Hph and an interaction term 1 + p pf + Heph . V p2 2 f

(B13)

(B14)

Further we use the variational principle for the ground-state energy in terms of the timeordered operators following Ref. [139]. The exact ground state |0 of the system with the Hamiltonian (B12) can be written in the interaction representation starting from the unperturbed ground state | : |0 = U (, ) | where U (, ) is the time-evolution operator,
t2

(B15)

U (t2 , t1 ) = T exp i

e|t| eiH0 t V eiH0 t .

(B16)

t1

208

Here, +0 and T denotes time ordering.

0 In the exact expectation value for the ground state energy Eext 0 |Hext | 0 , the phonons

are eliminated using the time ordered-operator calculus as in Ref. [139]. The average of the interaction term becomes then 0 |Heph | 0 2 = i dtei|t||t| V 1 T U (, ) eiq[r(t)r(0)] 2 q q

(B17)

This means that the polaron ground state energy is exactly described using a retarded potential in the interaction representation, cf. Eq. (2.16) of Ref. [139]. The ground state energy satises the Ritz variational principle with a trial state. Choosing the trial state as the ground state of the Hamiltonian (B13), the variational principle can be written as [139]
0 0 Eext Etr + |T {Utr (, ) [Hext (0) H0 (0)]}| ,

(B18)

where Utr (, ) is the time-evolution operator corresponding to the trial Hamiltonian (B6). The exact polaron ground state energy is denoted here as E 0 (k), where k is the polaron translation momentum. We nd an upper bound for E 0 (k) substituting (B17) in (B18) and using the exact wave functions and energy levels of the trial Hamiltonian. The trial Hamiltonian (B6) can be rewritten in terms of the coordinates (R, ) and momenta (P, ) of the center-of-mass and relative (internal) motions of the trial system with the masses M = 1 + mf and = mf / (1 + mf ) using the frequency v = wM . The energy spectrum of the trial system is the sum of the translation- and oscillation contributions, Ek,n = 3 k2 . + n , n = v n + 2M 2 (B19)

The eigenfunctions of the Hamiltonian (B6) are products of translational- and oscillatory wave functions: 1 k;l,n,m (R, ) = eikR l,n,m () , V 209 (B20)

where l,n,m () is the 3D harmonic-oscillator wave function with a given angular momentum. The result is
(0,0) E 0 (k) Ep (k) , (0,0) Ep

(B21)

3 (v w )2 (k) = 4 v 1 1 1 + 2 (1 + mf )2

2 4 2

dq q2
2

k ,l ,n ,m

| k;0,0,0 |eiqr | k ;l ,n ,m | , 1 )2 k2 + vn + 1 ( k 2(mf +1)

(B22)

where v > w are the Feynman variational frequencies. The functional (B22) can be reduced to the known Feynman result for the polaron ground-state energy. In the r.h.s. of (B22 at the polaron momentum k = 0, we introduce the integral over the Euclidean time: 1
(k )2 2(mf +1)

+ vn + 1

=
0

(k )2
2 mf +1

+vn +1

d.

(B23)

After this, the summations and integrations in (B25) are performed analytically, and we arrive at the Feynman variational expression for the polaron ground-state energy: E 0 (k) 3 (v w )2 k=0 4 v v e d. 2 w2 0 v ) (1 e w2 + v v

(B24)

The electron-phonon contribution in (B22) is structurally similar to the second-order perturbation correction to the polaron ground-state energy due to the electron-phonon interaction (using states of the Feynman model k;l,n,m as the zero-order approximation). Therefore we can estimate the energies of the excited polaron states when averaging the dierence between exact and unperturbed Hamiltonians in (B18) with an excited trial state. We then arrive at the following extension for the r.h.s. of (B22):
(l,n) Ep (k) =

v2 + w2 2v

n+

3 2

3 w 2 k2

1 + 2

1 1 (1 + mf )2

2 4 2

dq q2
2

k ,l ,n ,m

| k;l,n,m |eiqr | k ;l ,n ,m | . 1 )2 k2 + v (n n) + 1 ( k 2(mf +1) 210

(B25)

In the same approach, we obtain the inverse lifetimes for the excited states of the polaron: 1 2 l,n (k) = dq 2 4 q
k ,l ,n ,m

k;l,n,m eiqr k ;l ,n ,m

q2 + v (n n) + 1 . 2 (mf + 1)

(B26)

The broadening of the excited polaron non-scattering states must be taken into account for an analytical study of the polaron optical conductivity. Using the above expressions, we determine the transition energies for the transitions between the ground and the rst excited state consider the transition energies in which Ep
(0) (1exc)

01exc Ep

(1exc)

Ep . Let us rst

(0)

are calculated using optimal values of the

parameters of the Feynman model obtained from the minimization of the variational groundstate energy Ep . This method formally leads to the Franck-Condon (FC) excited states, with the frozen phonon conguration corresponding to the ground state of the polaron. Note that the existence of Franck-Condon states as eigenstates of the Fr ohlich polaron Hamiltonian has not been proved: Ref [39] suggests their non-existence as eigenstates for a simplied polaron model. Nevertheless the Franck-Condon concept can be signicant, e. g. for approximate treatments using a basis of Franck-Condon states, as indicative for the frequency of the maxima of phonon-sidebands, etc.

Fig. 1. Franck-Condon transition energies as a function of the coupling constant compared to the lowest-energy peak position of the polaron optical conductivity from Ref. [5] and the maximum of the polaron optical conductivity band from Ref. [77]. 211

In Fig. 1, the FC transition energies calculated with the approach introduced in the present work for polaron momentum k = 0 are plotted as a function of the coupling constant . They are compared with the peak energies of the polaron optical conductivity calculated using the diagrammatic Monte Carlo method (DQMC) [77, 125] and with the peak energies attributed to polaron relaxed excited states (RES) in Ref. [5] (DSG). The DQMC and DSG main-peak energies are close to each other in the whole range of the coupling strength. In the range 4 10, the present result for the transition energy is close to the DQMC and the DSG peak energies. Furthermore, in this range of , the non-monotonous behavior of the curvature is remarkably the same for the DQMC and DSG peak energies and for the present result. There is a remarkable agreement between the peaks attributed to the RES in Ref. [5], the peak positions obtained within the strong-coupling approach, Eq. (3) of Ref. [125], and the positions of the maximum of the optical conductivity band calculated in Ref. [77] using DQMC. It is reasonable that the three aforesaid peaks must be interpreted in one and the same way. In order to clarify this, we can refer to Ref. [77]. In the strong-coupling regime, the dominant broad peak of the polaron optical conductivity spectrum can be considered as a Franck-Condon sideband of the groundstate to RES-transition, even if this latter transition can have a negligible oscillator strength (see also [47]). The optical conductivity spectra of Ref. [125] in the strong-coupling approximation have been calculated taking into account the polaronic shift of the energy levels. The polaronic shift in Ref. [125] has been calculated with the Franck-Condon wave functions (i. e., with the strong-coupling wave functions corresponding to the frozen lattice conguration for the ground state). Note that the exact excitation spectrum of the Fr ohlich-Hamiltonian might be devoid of Franck-Condon eigenstates, cf. Ref. [39]). It should be remarked that the maxima of the FC-sideband structures of Ref. [5] are positioned at the frequency = v , i. e., at the transition frequency for the model system without the polaron shift. The Franck-Condon peak energies calculated in the present work also take into account the polaron shift. As follows from the above analysis, in the strong-coupling limit they must correspond to the Franck-Condon peak energies of the strong-coupling expansion of Ref. [125]. The agreement of the position of the maxima of these peaks with those attributed to transitions to the RES in Ref. [5] shows that in the strong-coupling range of , the latter should be associated to the Franck-Condon sidebands rather than to the RES. 212

Another approach, in which the parameters of the rst excited state are determined selfconsistently (Ref. [47]), was used i. a. to calculate (in the strong-coupling case) the (lowest) energy level of the relaxed excited state (RES). The transitions from the polaron ground state to the RES correspond to a zero-phonon peak in the optical conductivity. For the study of the energies of excited states of the polaron, a variational approach requires special care, because the excited states of the polaron are not stable. A variational approach, strictly speaking, is only valid for excited states when the variational wave function of the excited state is orthogonal to the exact ground-state wave function. For the estimation of the energy of the rst RES with our present formalism, we determine a minimum of the expression (B25) in a physically reasonable range of the variational parameters. In order to determine that range, we refer to Ref. [140], where the energy of the polaron RES is calculated variationally within the Greens function formalism. The expression for the RES energy in Ref. [140] contains the electron-phonon contribution corresponding to the second-order perturbation formula. It diers, however, from the weakcoupling second-order perturbation expression by the choice of the unperturbed states: in Ref.[140] those are variational states rather than free-electron states. There exists some analogy between our approach and that of Ref. [140]. The latter, however, does not take into account the translation invariance of the polaron problem. In Ref. [140], the energy of the polaron RES is calculated variationally. The unperturbed wave function of the RES is chosen orthogonal (due to symmetry) to the unperturbed ground state wave function. In the present approach, this orthogonality is also exactly satised because of symmetry. The expressions for the polaron RES energy of Ref. [140] contain singularities, which occur when the energies of the unperturbed ground state and that of the rst excited states are in resonant with the LO-phonon energy. These singularities are related to the instability of the excited polaron with respect to the emission of LO-phonons. Using the same reasoning as in Ref. [140] we search for a local minimum of the polaron RES energy in the range where the connement frequency v of the Feynman model satises the inequality v > 1. The instability of the excited polaron state is then avoided. The resulting numerical values of the transition energy to the rst RES as a function of are shown in Fig. 2. They are compared with the numerical-DQMC peak energies of the polaron optical conductivity band [77, 125], with the FC transition energies obtained in the 213

present work, and with the leading term of the strong-coupling approximation for the RES transition energy from Ref. [47].

Fig. 2. The transition energy for the transition from the polaron ground state to the rst RES(solid black curve ) and to the rst excited FC state (dashed red curve ) as a function of obtained in the present work, compared with the maximum of the polaron optical conductivity band from numerical DQMC (black squares, Ref. [77]). The dashed-dot green curve : the strong-coupling result for this transition energy as given in Ref. [47]. For 2.5, there exists no minimum of Ep
(1exc)

in the range v > 1. We can interpret

this result as a manifestation of the fact that for decreasing coupling strength, the RES is suppressed at suciently weak coupling. We see that for suciently small ( 6), the RES transition energies show good agreement with the DQMC peak energies, what conrms the concept of RES developed in Refs. [5, 47]. For higher coupling strengths, the DQMC data appear to be closer to the FC (rather than to RES) transition energies. This result can be an indication of the fact that with increasing , the mechanism of the polaron optical absorption changes its nature as suggested in Ref. [125], from a regime with dynamic lattice relaxation (for which the RES are relevant) at weak and intermediate coupling to the Franck-Condon (LO-phonon sidebands-) regime at strong coupling. In summary, we have re-formulated the Feynman all-coupling path integral method for the polaron problem within a Hamiltonian formalism using time-ordered operator calculus. This reformulation allows us to describe not only the free energy and the ground state, but also to directly determine for the rst time the excited polaron states that correspond to the Feynman all-coupling polaron model. A variational procedure for the polaron RES 214

energy has been developed, within the formalism presented in this work, which provides results i.a. in agreement with the strong-coupling limit of Ref. [47]. The present treatment oers the prospect of further elucidation of the nature of the polaron resonances (relaxed excited states versus Franck-Condon sidebands [125]) at intermediate coupling.

Appendix C: Notes on the polaron mobility

Here, the discussion on the polaron mobility in the weak-coupling regime is based on Ref. [141]. Several theoretical methods have been applied to study the transport properties of the Fr ohlich polaron. A challenging diculty is that, even for weak coupling and in the ohmic regime, there is a remarkable dierence in the mobility as obtained via a relaxation-time approximation [6063, 142], and as obtained via the path-integral formalism, worked out by Thornber and Feynman [46], and which is based on the Feynman polaron model [41]. At weak coupling and small electric eld, the relaxation time result for the mobility [64] seems more reliable than the Thornber-Feynman result. This might be partly due to the deviation of the electron velocity distribution from a drifted Maxwellian as shown analytically [143] from the Boltzmann equation at weak electron-phonon coupling and low temperature in the steady state regime. Because the Boltzmann equation is valid at weak electron-phonon coupling, and because of its intuitively transparent structure, this equation is an important tool to study transport properties of polarons for weak coupling. In Ref. [141], its solution is discussed in the ohmic regime and for the steady state. The mobility in the zero temperature limit from Ref. [141] is given by: |T 0 e 1 N , 2 (C1)

is the average number of phonons. This is equivalent to the result from the where N relaxation-time approximation [63], which therefore holds in the zero-temperature limit. An analytical solution of the Boltzmann equation at T = 0 was obtained in Ref. [144]. In Fig. ??, the mobility of a polaron in the weak-coupling regime, calculated using the exact solution of the Boltzmann equation [144] is calculated for InSb at T = 77 K and compared to the mobility from Ref. [46]. For weak electric elds the result of Ref. [144] is quite close to that of the polaron theory with a relaxation time but diers by the factor
3 kB T 1 2 0 2.5

from [46]. These results seem to conrm the validity of a relaxation time approach 215

for the electric eld E 0 (at least for InSb at 77 K). Nevertheless, as pointed out in [145], a system of non-interacting polarons is not ergodic and this point should be examined carefully before denite conclusions can be drawn when E 0. In Ref. [146], arbitrary temperature and electric eld are considered, and an exact recursion relation is obtained for the time-dependent expansion coecients of the electron distribution function in terms of Legendre polynomials.

FIG. 22: Mobility of weak-coupling polarons, obtained from the exact solution of the Boltzmann equation [144] (solid line) and from [46] (dashed line). (After Ref. [145].)

The DC mobility of a polaron in the strong-coupling regime was investigated by Volovik et al. [147]. They showed that the interaction Hamiltonian corresponding to scattering of a phonon by a polaron can be separated in the strong-coupling limit with the aid of the transformations of Bogoliubov and Tyablikov in conjunction with the LLP canonical transformation. In the leading order in powers of the inverse coupling constant 1 , the principal role in the scattering is played by two-phonon processes. In the system of units with = 1, the following result for the strong-coupling polaron mobility has been obtained: = mLO 2 T 0 /T e , 0 (C2)

with a numerical temperature-independent coecient 1. 216

The relaxation-time approximation used in Refs. [6063, 142] consists in disregarding the contribution of the so-called re-population term [87] in the collision integral. In the work by V. F. Los [148], the polaron mobility was calculated on the basis of Kubos formula using a Greens superoperator technique. As stated in Ref. [148], the relaxation-time approximation [63] does not take into account the change in the electron velocity in all the electron-phonon scattering processes allowed by the energy and momentum conservation laws. The polaron mobility obtained in Ref. [148] gives the correct temperature dependence of the polaron mobility but exceeds the expression obtained by Kadano exactly by a factor 3. Recently, an approach to the polaron mobility has been proposed [149], based on the dynamics of the Wigner distribution function, using the kinetic equations derived in Refs. [150, 151]. In these works, the re-population term has been explicitly taken into account. Consequently, the result of Ref. [148] for the polaron mobility is retrieved.

Appendix D: Publications on the polarons in Physical Review Letters (since 2005)

1. Method for Analyzing Second-Order Phase Transitions: Application to the Ferromagnetic Transition of a Polaronic System, J. A. Souza, Yi-Kuo Yu, J. J. Neumeier, H. Terashita, and R. F. Jardim, Phys. Rev. Lett. 94, 207209 (2005).
Abstract A new method for analyzing second-order phase transitions is presented and applied to the polaronic system La0.7 Ca0.3 MnO3 . It utilizes heat capacity and thermal expansion data simultaneously to correctly predict the critical temperatures pressure dependence. Analysis of the critical phenomena reveals second-order behavior and an unusually large heat capacity exponent.

2. Validity of the Franck-Condon Principle in the Optical Spectroscopy: Optical Conductivity of the Fr ohlich Polaron, G. De Filippis, V. Cataudella, A. S. Mishchenko, C. A. Perroni, and J. T. Devreese, Phys. Rev. Lett. 96, 136405 (2006).
Abstract The optical absorption of the Fr ohlich polaron model is obtained by an approximation-free diagrammatic Monte Carlo method and compared with two new approximate approaches that treat lattice relaxation eects in dierent ways. We show that: (i) a strong coupling expansion, based on the Franck-Condon principle, well describes the optical conductivity for

217

large coupling strengths ( > 10); (ii) a memory function formalism with phonon broadened levels reproduces the optical response for weak coupling strengths ( < 6) taking the dynamic lattice relaxation into account. In the coupling regime 6 < < 10, the optical conductivity is a rapidly changing superposition of both Franck-Condon and dynamic contributions.

3. Remanent Zero Field Spin Splitting of Self-Assembled Quantum Dots in a Paramagnetic Host, C. Gould, A. Slobodskyy, D. Supp, T. Slobodskyy, P. Grabs, P. Hawrylak, F. Qu, G. Schmidt, and L. W. Molenkamp, Phys. Rev. Lett. 97, 017202 (2006). 4. Quantum Transport of Slow Charge Carriers in Quasicrystals and Correlated Systems, Guy Trambly de Laissardi` ere, Jean-Pierre Julien, and Didier Mayou, Phys. Rev. Lett. 97, 026601 (2006).
Abstract We show that the semiclassical model of conduction breaks down if the mean free path of charge carriers is smaller than a typical extension of their wave function. This situation is realized for suciently slow charge carriers and leads to a transition from a metalliclike to an insulatinglike regime when scattering by defects increases. This explains the unconventional conduction properties of quasicrystals and related alloys. The conduction properties of some heavy fermions or polaronic systems, where charge carriers are also slow, present a deep analogy.

5. Occurrence of Intersubband Polaronic Repellons in a Two-Dimensional Electron Ga s, Stefan Butscher and Andreas Knorr, Phys. Rev. Lett. 97, 197401 (2006). 6. Subsecond Spin Relaxation Times in Quantum Dots at Zero Applied Magnetic Field Due to a Strong Electron-Nuclear Interaction, R. Oulton, A. Greilich, S. Yu. Verbin, R. V. Cherbunin, T. Auer, D. R. Yakovlev, M. Bayer, I. A. Merkulov, V. Stavarache, D. Reuter, and A. D. Wieck, Phys. Rev. Lett. 98, 107401 (2007). 7. Exciton Dephasing in Quantum Dots due to LO-Phonon Coupling: An Exactly Solvable Model, E. A. Muljarov and R. Zimmermann, Phys. Rev. Lett. 98, 187401 (2007) 8. Electron-Phonon Interaction and Charge Carrier Mass Enhancement in SrTiO 3 , J. L. M. van Mechelen, D. van der Marel, C. Grimaldi, A. B. Kuzmenko, N. P. Armitage, 218

N. Reyren, H. Hagemann, and I. I. Mazin, Phys. Rev. Lett. 100, 226403 (2008).
Abstract We report a comprehensive THz, infrared and optical study of Nb-doped SrTiO3 as well as dc conductivity and Hall eect measurements. Our THz spectra at 7 K show the presence of an unusually narrow (< 2meV) Drude peak. For all carrier concentrations the Drude spectral weight shows a factor of three mass enhancement relative to the eective mass in the local density approximation, whereas the spectral weight contained in the incoherent midinfrared response indicates that the mass enhancement is at least a factor two. We nd no evidence of a particularly large electron-phonon coupling that would result in small polaron formation.

9. Orbital and Charge-Resolved Polaron States in CdSe Dots and Rods Probed by Scanning Tunneling Spectroscopy, Zhixiang Sun, Ingmar Swart, Christophe Delerue, Dani el Vanmaekelbergh, and Peter Liljeroth, Phys. Rev. Lett. 102, 196401 (2009). 10. Dynamical Response and Connement of the Electrons at the LaAlO 3 /SrTiO 3 Interface, A. Dubroka, M. R ossle, K. W. Kim, V. K. Malik, L. Schultz, S. Thiel, C. W. Schneider, J. Mannhart, G. Herranz, O. Copie, M. Bibes, A. Barth el emy, and C. Bernhard, Phys. Rev. Lett. 104, 156807 (2010).
Abstract With infrared ellipsometry and transport measurements we investigated the electrons at the interface between LaAlO3 and SrTiO3 . We obtained a sheet carrier concentration of

Ns 5 9 1013 cm2 , an eective mass of m = 3.2 0.4me , and a strongly frequency

dependent mobility. The latter are similar as in bulk SrTi1x Nbx O3 and therefore suggestive of polaronic correlations. We also determined the vertical concentration prole which has a strongly asymmetric shape with a rapid initial decay over the rst 2 nm and a pronounced tail that extends to about 11 nm.

11. Bipolaron and N-Polaron Binding Energies, Rupert L. Frank, Elliott H. Lieb, Robert Seiringer, and Lawrence E. Thomas, Phys. Rev. Lett. 104, 210402 (2010).
Abstract The binding of polarons, or its absence, is an old and subtle topic. Here we prove two things rigorously. First, the transition from many-body collapse to the existence of a thermodynamic limit for N polarons occurs precisely at U = 2, where U is the electronic Coulomb repulsion and is the polaron coupling constant. Second, if U is large enough, there is no

219

multipolaron binding of any kind. Considering the known fact that there is binding for some

U > 2, these conclusions are not obvious and their proof has been an open problem for
some time.

12. Polaronic Conductivity in the Photoinduced Phase of 1T-TaS 2, N. Dean, J. C. Petersen, D. Fausti, R. I. Tobey, S. Kaiser, L. V. Gasparov, H. Berger, and A. Cavalleri, Phys. Rev. Lett. 106, 016401 (2011). 13. Spectroscopy of Single Donors at ZnO(0001) Surfaces, Hao Zheng, J org Kr oger, and Richard Berndt, Phys. Rev. Lett. 108, 076801 (2012) 14. Polarons in Suspended Carbon Nanotubes, I. Snyman, and Yu. V. Nazarov, Phys. Rev. Lett. 108, 076805 (2012) 15. Two-Dimensional Polaronic Behavior in the Binary Oxides m-HfO 2 and m-ZrO 2 , K. P. McKenna, M. J. Wolf, A. L. Shluger, S. Lany, and A. Zunger, Phys. Rev. Lett. 108, 116403 (2012) 16. Polaron-to-Polaron Transitions in the Radio-Frequency Spectrum of a Quasi-TwoDimensional Fermi Gas, Y. Zhang, W. Ong, I. Arakelyan, and J. E. Thomas, Phys. Rev. Lett. 108, 235302 (2012)
Abstract We measure radio-frequency spectra for a two-component mixture of a 6 Li atomic Fermi gas in a quasi-two-dimensional regime with the Fermi energy comparable to the energy level spacing in the tightly conning potential. Near the Feshbach resonance, we nd that the observed resonances do not correspond to transitions between connement-induced dimers. The spectral shifts can be t by assuming transitions between noninteracting polaron states in two dimensions.

17. Model of the Electron-Phonon Interaction and Optical Conductivity of Ba 1x K x BiO 3 , R. Nourafkan, F. Marsiglio, and G. Kotliar, Phys. Rev. Lett. 109, 017001 (2012) 18. p-Wave Polaron, Jesper Levinsen, Pietro Massignan, Fr ed eric Chevy, and Carlos Lobo, Phys. Rev. Lett. 109, 075302 (2012)

220

19. Eect of Electron-Phonon Interaction Range for a Half-Filled Band in One Dimension, Martin Hohenadler, Fakher F. Assaad, and Holger Fehske, Phys. Rev. Lett. 109, 116407 (2012) 20. Digital Quantum Simulation of the Holstein Model in Trapped Ions, A. Mezzacapo, J. Casanova, L. Lamata, and E. Solano, Phys. Rev. Lett. 109, 200501 (2012) 21. Bilayers of Rydberg Atoms as a Quantum Simulator for Unconventional Superconductors, J. P. Hague and C. MacCormick, Phys. Rev. Lett. 109, 223001 (2012) 22. Relaxation Dynamics of the Holstein Polaron, Denis Gole z, Janez Bon ca, Lev Vidmar, and Stuart A. Trugman, Phys. Rev. Lett. 109, 236402 (2012) 23. Quantum Simulation of Small-Polaron Formation with Trapped Ions, Vladimir M. Stojanovi c, Tao Shi, C. Bruder, and J. Ignacio Cirac, Phys. Rev. Lett. 109, 250501 (2012) 24. Quantum Breathing of an Impurity in a One-Dimensional Bath of Interacting Bosons, Sebastiano Peotta, Davide Rossini, Marco Polini, Francesco Minardi, and Rosario Fazio, Phys. Rev. Lett. 110, 015302 (2013)
Abstract By means of the time-dependent density-matrix renormalization-group (TDMRG) method we are able to follow the real-time dynamics of a single impurity embedded in a onedimensional bath of interacting bosons. We focus on the impurity breathing mode, which is found to be well described by a single oscillation frequency and a damping rate. If the impurity is very weakly coupled to the bath, a Luttinger-liquid description is valid and the impurity suers an Abraham-Lorentz radiation-reaction friction. For a large portion of the explored parameter space, the TDMRG results fall well beyond the Luttinger-liquid paradigm.

25. Measurement of Coherent Polarons in the Strongly Coupled Antiferromagnetically Ordered Iron-Chalcogenide Fe 1.02 Te using Angle-Resolved Photoemission Spectroscopy, Z. K. Liu, R.-H. He, D. H. Lu, M. Yi, Y. L. Chen, M. Hashimoto, R. G. Moore, S.-K. Mo, E. A. Nowadnick, J. Hu, T. J. Liu, Z. Q. Mao, T. P. Devereaux, Z. Hussain, and Z.-X. Shen, Phys. Rev. Lett. 110, 037003 (2013) 221

26. Decoherence of a Single-Ion Qubit Immersed in a Spin-Polarized Atomic Bath, L. Ratschbacher, C. Sias, L. Carcagni, J. M. Silver, C. Zipkes, and M. K ohl, Phys. Rev. Lett. 110, 160402 (2013) 27. Tunable Polaronic Conduction in Anatase TiO 2 , S. Moser, L. Moreschini, J. Ja cimovi c, O. S. Bari si c, H. Berger, A. Magrez, Y. J. Chang, K. S. Kim, A. Bostwick, E. Rotenberg, L. Forr o, and M. Grioni, Phys. Rev. Lett. 110, 196403 (2013) 28. Investigating Polaron Transitions with Polar Molecules, Felipe Herrera, Kirk W. Madison, Roman V. Krems, and Mona Berciu, Phys. Rev. Lett. 110, 223002 (2013)
Abstract We determine the phase diagram of a polaron model with mixed breathing-mode and SuSchrieer-Heeger couplings and show that it has two sharp transitions, in contrast to pure models which exhibit one (for Su-Schrieer-Heeger coupling) or no (for breathing-mode coupling) transition. We then show that ultracold molecules trapped in optical lattices can be used as a quantum simulator to study precisely this mixed Hamiltonian, and that the relative contributions of the two couplings can be tuned with external electric elds. The parameters of current experiments place them in the region where one of the transitions occurs. We also propose a scheme to measure the polaron dispersion using stimulated Raman spectroscopy.

222

[1] L. D. Landau, Phys. Z. Sowjetunion 3, 664 (1933) [English translation in Collected Papers, Gordon and Breach, New York, 1965, pp. 67-68]. [2] J. T. Devreese, in Lectures on the Physics of Highly Correlated Electron Systems VII, edited A. Avella and F. Mancini Proceedings of the 7th Training Course in the Physics of Correlated Electron Systems&High-Tc Superconductors, Vietri sul Mare, Italy, October 14-16, 2002, AIP, Melville (2003), pp. 3 - 56. [3] H. Fr ohlich, Adv. Phys. 3, 325 (1954). [4] R. P. Feynman, R. B. Leiton and M. Sands, The Feynman Lectures on Physics (AddisonWesley, 1972), Vol. II. [5] E. Kartheuser, in Polarons in Ionic Crystals and Polar Semiconductors edited J. T. Devreese, North-Holland, Amsterdam (1972), pp. 717 - 733. [6] M. Grynberg, S. Huant, G. Martinez, J. Kossut, T. Wojtowicz, G. Karczewski, J. M. Shi, F. M. Peeters, J. T. Devreese, Phys. Rev. B 54, 1467 (1996). [7] The spectra of the infrared-active LO (and TO) phonons in -Al2 O3 (sapphire) contain six modes. The values of the LO and TO phonon frequencies and of the high-frequency dielectric constants = 3.072, = 3.077 are taken from Ref. [8]. Using these parameters and the electron band mass mb = 0.25me as estimated in Ref. [9], the eective value of the electron-phonon coupling constant in Al2 O3 has been calculated as =
j

ej

k k

where ej is the polarization vector of the j -th LO-phonon branch, k is the phonon wave vector, denote the angular averaging, and the coupling constants j for each branch are

obtained using the method [10]. The resulting value of the polaron coupling constant is 1.25 [8] M. Schubert, T. E. Tiwald, and C. M. Herzinger, Phys. Rev. B 61, 8187 (2000). [9] J. Shan, F. Wang, E. Knoesel, M. Bonn, and T. F. Heinz, Phys. Rev. Lett. 90, 247401 (2003). [10] S. N. Klimin, V. M. Fomin, and J. T. Devreese, to be published. [11] J. W. Hodby, G. P. Russell, F. Peeters, J. T. Devreese, and D. M. Larsen, Phys. Rev. Lett.

223

58, 1471 (1987). [12] The spectra of the LO (and TO) phonons in -SiO2 contain ten modes. The values of the LO and TO phonon frequencies are taken from Refs. [13, 14]. Using these frequencies and the value = 2.40 from Ref. [15] for the high-frequency dielectric constant, the eective value of the electron-phonon coupling constant in SiO2 has been calculated using the method of Ref. [10] as indicated in Ref. [7]. We use the estimated value of the electron band mass mb = 0.5me as in Refs. [1618]. The resulting value of the polaron coupling constant is 1.59. [13] F. Gervais and B. Piriou, Phys. Rev. B 11, 3944 (1975). [14] J. L. Duarte, J. A. Sanjurjo, and R. S. Katiyar, Phys. Rev. B 36, 3368 (1987). [15] S. T. Pantelides and W. A. Harrison, Phys. Rev. B 13, 2667 (1976). [16] M. V. Fischetti, D. J. DiMaria, L. Dori, J. Batey, E. Tierney, and J. Stasiak, Phys. Rev. B 35, 4404 (1987). [17] D. Arnold, E. Cartier, and D. J. DiMaria, Phys. Rev. B 49, 10278 (1994). [18] P. Martin, S. Guizard, Ph. Daguzan, G. Petite, P. DOliveira, P. Meynadier, and M. Perdrix, Phys. Rev. B 55, 5799 (1997). [19] I. Biaggio, R. W. Hellwarth, and J. P. Partanen, Phys. Rev. Lett. 78, 891 (1997) (for mb /me = 2). [20] G. Verbist, F. M. Peeters, and J. T. Devreese, Ferroelectrics, 130, 27 (1992) (for mb /me = 2.6). [21] S. I. Pekar, Issledovanija po Ekektronnoj Teorii Kristallov, Gostekhizdat, Moskva, 1951 (in Russian) [German translation: Untersuchungen u ber die Elektronentheorie der Kristalle, Akademie Verlag, Berlin, 1951]. [22] G. C. Kuper and G. D. Whiteld (eds.), Polarons and Excitons, Oliver and Boyd, Edinburgh, 1963. [23] J. Appel, in Solid State Physics, edited by F. Seitz, D. Turnbull, and H. Ehrenreich, Academic Press, New York, 1968, vol. 21, pp. 193-391. [24] J. T. Devreese (ed.), Polarons in Ionic Crystals and Polar Semiconductors, North-Holland, Amsterdam, 1972. [25] T. K. Mitra, A. Chatterjee, and S. Mukhopadhyay, Phys. Rep. 153, 91 (1987). [26] J. T. Devreese, in Encyclopedia of Applied Physics, edited by G. L. Trigg, VCH, Weinheim,

224

1996, vol. 14, pp. 383 - 413. [27] A. S. Alexandrov and Sir Nevill Mott, Polarons and Bipolarons, World Scientic, Singapore, 1996. [28] A. S. Mishchenko, N. V. Prokofev, A. Sakamoto, and B. V. Svistunov, Phys. Rev. B 62, 6317 (2000). [29] T. D. Lee, F. E. Low, and D. Pines, Phys. Rev. 90, 297 (1953). [30] J. R oseler, Phys. Stat. Sol.(b) 25, 311 (1968) [31] M. A. Smondyrev, Teor. Math. Fiz. 68, 29 (1986) [English translation: Theor. Math. Phys. 68, 653 (1986)] [32] O. V. Selyugin and M. A. Smondyrev, Phys. Stat. Sol.(b) [33] L. D. Landau and S. I. Pekar, Zh. Eksper. Teor. Fiz. 18, 419 (1948) [34] N. N. Bogolubov and S. V. Tyablikov, Zh. Eksp. i Teor. Fiz. 19, 256 (1949) [35] N. N. Bogolubov, Ukr. Matem. Zh. 2, 3 (1950) [36] S. V. Tyablikov, Zh. Eksp. i Theor. Phys. 21, 377 (1951) [37] R. Evrard, Phys. Letters 14, 295 (1965) [38] S. J. Miyake, J. Phys. Soc. Japan 38, 181 (1975) [39] J. T. Devreese and R. Evrard, Phys. Letters 11, 278 (1964). [40] E. Kartheuser, R. Evrard, and J. Devreese, in Optical Properties of Solids, edited by E. D. Haidemenakis, Gordon and Breach, New York, 1970, pp. 433-459 (Table 1). [41] R. P. Feynman, Phys. Rev. 97, 660 (1955). [42] J. T. Devreese and R. Evrard, in Proceedings of the British Ceramic Society 10, 151 (1968). [43] J. T. Titantah, C. Pierleoni, and S. Ciuchi, Phys. Rev. Lett. 87, 206406 (2001). [44] G. De Filippis, V. Cataudella, V. Marigliano Ramaglia, C. A. Perroni, and D. Bercioux, Eur. Phys. J. B 36, 65 (2003). [45] R. P. Feynman, R. W. Hellwarth, C. K. Iddings, and P. M. Platzman, Phys. Rev. 127, 1004 (1962). [46] K. K. Thornber and R. P. Feynman, Phys. Rev. B 1, 4099 (1970) [47] E. Kartheuser, R. Evrard, and J. Devreese, Phys. Rev. Lett.22, 94 (1969). [48] J. T. Devreese, J. De Sitter, and M. Goovaerts, Phys. Rev. B 5, 2367 (1972) [49] F. M. Peeters and J. T. Devreese Phys. Rev. B 34, 7246 (1986). [50] V. Cataudella, G. De Filippis, and G. Iadonisi, Eur. Phys. J. B 12, 17 (1999).

225

[51] J. Tempere and J. T. Devreese, Phys. Rev. B 64, 104504 (2001). [52] J. Devreese, R. Evrard, and E. Kartheuser, Phys. Rev. B 12 3353 (1975). [53] W. Becker, B. Gerlach, H. Schlike, Phys. Rev. B 28, 5735 (1983). [54] F. M. Peeters, J. T. Devreese, Phys. Rev. B 31, 6826 (1985). [55] Y. Osaka, Prog. Theor. Phys. 22, 437 (1959). [56] J. T. Devreese, Contribution to the polaron theory, Ph.D. Thesis, KU Leuven, 1964. [57] J. T. Devreese and R. Evrard, in Proceedings of the British Ceramic Society 10, 151 (1968). Reprinted in: Path Integrals and Their Applications in Quantum, Statistical, and Solid State Physics, edited by G. J. Papadopoulos and J. T. Devreese, NATO ASI Series B, Physics, vol. 34, Plenum, New York, 1977, pp. 344-357. [58] F. M. Peeters and J. T. Devreese, in Solid State Physics, edited by F. Seitz and D. Turnbull, Academic Press, New York, 1984, vol. 38, pp. 81 - 133. [59] H. Fr ohlich, Proc. R. Soc. (London) Ser. A 160, 230 (1937). [60] D. J. Howarth and E. H. Sondheimer, Proc. R. Soc. (London) Ser. A 219, 53 (1953). [61] Y. Osaka, Progr. Theoret. Phys. 25, 517 (1961). [62] F. E. Low and D. Pines, Phys. Rev. 98, 414 (1955). [63] L. P. Kadano, Phys. Rev. 130, 1364 (1963). [64] D. C. Langreth and L. P. Kadano, Phys. Rev. 133, A1070 (1964). [65] F.M. Peeters, J.T. Devreese, Phys. Stat. Sol. (b) 115, 539 (1983). [66] R. W. Hellwarth and I. Biaggio, Phys. Rev. B 60, 299 (1999). [67] F. C. Brown, in Point Defects in Solids, edited by J. H. Crawford and L. M. Slifkin, Plenum, New York, 1972, vol. 1, p. 537. [68] E. Hendry, F. Wang, J. Shan, T. F. Heinz, and M. Bonn, Phys. Rev. B 69, 081101(R) (2004). [69] V. L. Gurevich, I. G. Lang, and Yu. A. Firsov, Fiz. Tverd. Tela 4, 1252 (1962) [English translation: Sov. Phys. Solid St. 4, 918 (1962)]. [70] G. D. Mahan, Many-Particle Physics, Kluwer/Plenum, New York, 2000. [71] J. Devreese, W. Huybrechts, and L. Lemmens, Phys. Stat. Sol. (b) 48, 77 (1971). [72] H. Finkenrath, N. Uhle, and W. Waidelich, Solid State Commun. 7, 11 (1969). [73] M. J. Goovaerts, J. De Sitter, and J. T. Devreese, Phys. Rev. 7, 2639 (1973). [74] F. M. Peeters and J. T. Devreese, Phys. Rev. B 28, 6051 (1983). [75] D. Forster, Hydrodynamic Fluctuations, Broken Symmetry and Correlation Functions, Ben-

226

jamin, New York, 1975. [76] F. M. Peeters and J. T. Devreese, Phys. Rev. B 23, 1936 (1981). [77] A. S. Mishchenko, N. Nagaosa, N. V. Prokofev, A. Sakamoto, and B. V. Svistunov, Phys. Rev. Lett. 91, 236401 (2003). [78] W. Huybrechts and J.T. Devreese, Phys. Rev. B 8, 5754 (1973). [79] D. M. Eagles, R. P. S. M. Lobo, and F. Gervais, Phys. Rev. B 52, 6440 (1995). [80] J. T. Devreese, L. Lemmens, and J. Van Royen, Phys. Rev. B 15, 1212 (1977). [81] L. F. Lemmens, J. De Sitter, and J. T. Devreese, Phys. Rev. B 8, 2717 (1973). [82] F. M. Peeters, Wu Xiaoguang, J. T. Devreese, Phys. Rev. B 33, 3926 (1986). [83] Wu Xiaoguang, F. M. Peeters, J. T. Devreese, Phys. Rev. B 31, 3420 (1985). [84] F. M. Peeters, J. T. Devreese, Phys. Rev. B 36, 4442 (1987). [85] F. M. Peeters, J. T. Devreese, Phys. Rev. B 28, 6051 (1983). [86] J. T. Devreese, L. F. Lemmens, J. Van Royen, Phys. Rev. B 2, 1212 (1977). [87] F. M. Peeters, J. T. Devreese, Solid State Phys. 38, 81 (1984). [88] F. Brosens, S. N. Klimin, and J. T. Devreese, Phys. Rev. B 77, 085308 (2008). [89] L. F. Lemmens, J. T. Devreese, and F. Brosens, Phys. Stat. Sol. (b) 82, 439 (1977). [90] S. Lupi, P. Maselli, M. Capizzi, P. Calvani, P. Giura and P. Roy, Phys. Rev. Lett. 83, 4852 (1999). [91] Ch. Hartinger, F. Mayr, J. Deisenhofer, A. Loidl, and T. Kopp, Phys. Rev. B 69, 100403(R) (2004). [92] Ch. Hartinger, F. Mayr, A. Loidl, and T. Kopp, cond-mat/0406123. [93] D. Emin, Phys. Rev. B 48, 13691 (1993). [94] G. P. Zhang, T. A. Callcott, G. T. Woods, L. Lin, B. Sales, D. Mandrus, and J. He, Phys. Rev. Lett. 88, 077401 (2002). [95] T. Hotta, Phys. Rev. B 67, 104428 (2003). [96] S. N. Klimin, V. M. Fomin, F. Brosens, and J. T. Devreese, Phys. Rev. B 69, 235324 (2004). [97] L. F. Lemmens, F. Brosens, and J. T. Devreese, Phys. Rev. E 53, 4467 (1996). [98] J. T. Devreese, S. N. Klimin, V. M. Fomin, and F. Brosens, Solid State Communications 114, 305 (2000). [99] F. Brosens, J. T. Devreese, and L. F. Lemmens, Phys. Rev. E 55, 227 (1997); 55, 6795 (1997); 58, 1634 (1998).

227

[100] L. F. Lemmens, F. Brosens, and J. T. Devreese, Solid State Communications 109, 615 (1999). [101] G. De Filippis, V. Cataudella and G. Iadonisi, Eur. Phys. J. B 8, 339 (1999). [102] W.B. da Costa, N. Studart, Phys. Rev. B 47, 6356 (1993). [103] F. Brosens, S. N. Klimin, and J. T. Devreese, Phys. Rev. B 71, 214301 (2005). [104] S. N. Klimin, V. M. Fomin, F. Brosens, and J. T. Devreese, Physica E 22, 494 (2004). [105] J. T. Devreese, in: Fluctuating Paths and Fields (World Scientic, Singapore, 2001), pp. 289-304. [106] G. Verbist, F. M. Peeters and J. T. Devreese, Phys. Rev. B 43, 2712 (1991). [107] M. A. Smondyrev, G. Verbist, F. M. Peeters, and J. T. Devreese, Phys. Rev. B 47, 2596 (1993). [108] N. I. Kashirina, V. D. Lakhno, and V. V. Sychyov, Phys. Stat. Sol. (b) 239, 174 (2003). [109] J. Tempere, S. N. Klimin, I. F. Silvera, J. T. Devreese, Eur. Phys. J. 32, 329 (2003). [110] A.P. Volodin, M.S. Khaikin, and V.S. Edelman, JETP Lett. 26, 543 (1977); U. Albrecht and P. Leiderer, Europhys. Lett. 3, 705 (1987). [111] V.B. Shikin, JETP Lett. 27, 39 (1978); M.M. Salomaa and G.A. Williams, Phys. Rev. Lett. 47, 1730 (1981). [112] I.F. Silvera, Bull. Am. Phys. Soc. 46, 1016 (2001). [113] J. Tempere, I.F. Silvera, and J.T. Devreese, Phys. Rev. Lett. 87, 275301 (2001). [114] S. Fratini and P. Qu emerais, Eur. Phys. J. B 14, 99 (2000). [115] F. Lindemann, Z. Phys. 11, 609 (1910); C.M. Care and N.H. March, Adv. Phys. 24, 101 (1975). [116] C.C. Grimes and G. Adams, Phys. Rev. Lett. 42, 795 (1979). [117] V.M. Bedanov and F.M. Peeters, Phys. Rev. B 49, 2667 (1994). [118] L.P. Gorkov and D.M. Chernikova, Pisma Zh. Eksp. Teor. Fiz. 18, 119 (1973) [JETP Lett. 18, 68 (1973)]. [119] J. Tempere, I.F. Silvera, and J.T. Devreese, Phys. Rev. B 67, 035402 (2003). [120] D.S. Fisher, B.I. Halperin, and P.M. Platzman, Phys. Rev. Lett. 42, 798 (1979). [121] G. Deville et al., Phys. Rev. Lett. 53, 588 (1984). [122] S. N. Klimin and J. T. Devreese (to be published ). [123] J. T. Devreese and A. S. Alexandrov, Rep. Prog. Phys. 72, 066501 (2009); A. S. Alexandrov and J. T. Devreese, Advances in Polaron Physics (Springer, 2009).

228

[124] B. E. Sernelius, Phys. Rev. B 48, 7043 (1993). [125] G. De Filippis, V. Cataudella, A. S. Mishchenko, C. A. Perroni, and J. T. Devreese, Phys. Rev. Lett. 96, 136405 (2006). [126] G. R. Allcock, in Polarons and Excitons, edited by C. G. Kuper and G. D. Whiteld (Oliver and Boyd, Edinburgh, 1963), pp. 45 70. [127] M. Born and K. Huang, Dynamical theory of crystal lattices (Oxford University Press, 2007). [128] Yu. E. Perlin, Sov. Physics. Uspekhi. 6, 542 (1964). [129] H. Jahn and E. Teller, Proc. R. Soc. London A 161, 220 (1937). [130] H. Kleinert, Path Integrals in Quantum Mechanics, Statistics, Polymer Physics, and Financial Markets (5th edition, World Scientic, Singapore 2009). [131] V. M. Fomin, V. N. Gladilin, J. T. Devreese, E. P. Pokatilov, S. N. Balaban, and S. N. Klimin, Phys. Rev. B 57, 2415 (1998). [132] E. N. Myasnikov, A. E. Myasnikova, and Z. P. Mastropas, Physics of the Solid State 48, 1046 (2006). [133] H. Spohn, Phys. Rev. B 33, 8906 (1986); Ann. Phys. 175, 278 (1987). [134] G. Wellein, H. R oder, and H. Fehske, Phys. Rev. B 53, 9666 (1996). [*] This work was presented at the 10th International Conference Path Integrals 2010, July 11 16, 2010, Washington DC, USA. [136] N. N. Bogoliubov and N. N. Bogoliubov, Jr., Some Aspects of Polaron Theory. In: Lecture Notes in Physics vol. 4, World Scientic, Singapore (1988). [137] K. Yamazaki, J. Phys. A 16, 3675 (1983). [138] V. Cataudella, G. De Filippis, and C. A. Perroni, in Polarons in Advanced Materials, Springer Series in Materials Science , Vol. 103, Edited by A. S. Alexandrov (Canopus and Springer, Bath, UK, 2007), pp. 149 189. [139] J. T. Devreese and F. Brosens, Phys. Rev. B 45, 6459 (1992). [140] Y. L epine and M. Charbonneau, Phys. Status Solidi B 122, 151 (1984); Y. Frongillo and Y. L epine, Phys. Rev. B 40, 3570 (1989). [141] J. T. Devreese and F. Brosens, Phys. Stat. Sol. (b) 108, K29 (1981). [142] T. D. Schultz, Phys. Rev. 116, 596 (1959). [143] J. T. Devreese and R. Evrard, Phys. Stat. Sol. (b) 78, 85 (1976). [144] J. T. Devreese, R. Evrard, and E. Kartheuser, Phys. Stat. Sol. (b) 90, K73 (1978).

229

[145] J. T. Devreese and R. Evrard, in: Linear and Nonlinear Electron Transport in Solids (Plenum Press, New York, 1978). [146] F. Brosens and J. T. Devreese, Phys. Stat. Sol. (b) 111, 591 (1982). [147] G. E. Volovik, V. I. Melnikov, and V. M. Edelshtein, JETP Letters 18, 138 (1973). [148] V. F. Los, Theor. and Math. Phys. 60, 703 (1984). [149] D. Sels and F. Brosens, Phase-space approach to polaron response: Kadano and FHIP revisited, arXiv:1310.3041v1 (2013). [150] D. Sels, F. Brosens, and W. Magnus, Physica A 392, 326 (2013). [151] D. Sels and F. Brosens, Phys. Rev. E 88, 042101 (2013).

230