AC-Conductivity of Pinned Charge Density Wave Fluctuations in Quasi One-Dimensional Conductors

18 Pages·0.16 MB·English

by  W. Wonneberger

AC–Conductivity of Pinned Charge Density Wave
Fluctuations in Quasi One–Dimensional Conductors
9 W. Wonneberger
9
9
1
Department of Physics, University of Ulm
n
a D–89069 Ulm, Germany
J

6
2

Abstract. Quasi one–dimensional conductors which undergo a Peierls transition
1
v to a charge density wave state at a temperature T show a region of one–dimensional
P
1
fluctuations above T . The Ginzburg–Landau–Langevin theory for the frequency de-
8 P
2 pendent collective conductivity from conductive fluctuations into the charge density
1
wave state is developed. By inclusion of a phase breaking term the effect of local
0
9 pinning due to random impurities is simulated. It is found that the spectral weight
9
of the unpinned fluctuations is partly redistributed into a pinned mode around a
/
t pinning frequency in the far infrared region. In addition, selection rule breaking by
a
m the impurities makes the fluctuating amplitude mode visible in the optical response.
-
d
n
o Short title: Ac–Conductivity of Pinned CDW Fluctuations
c
:
v
i
X
PACS: 71.45.Lr, 71.55.Jv, 72.15.Nj
r
a
Phone ++49 731 502 2991/4
Fax ++49 731 502 3003
e–mail [email protected]
1
1 Introduction
Quasi one–dimensional conductors like the transition metal chalcogenides are charac-
terizedbyanestedFermisurface. ThisrendersthemunstabletothePeierlstransition
when electron–phonon backscattering between the two sheets of the Fermi surface
become relevant. We call this a Peierls–system (PS). The ensuing charge density
wave (CDW) which develops below a transition temperature T of the order of 200K
P
shows many unusual properties, especially in its electronic responses [1].
In recent years it became increasingly evident that the electronic properties of PS
in their normal phase (T > T ) are also unusual. Photoemission studies ([2, 3] and
P
earlierworkcitedtherein) pointtowardspossiblenon–Fermi–liquid(NFL)behaviour.
The microwave and optical response also deviates from the Drude predictions for a
normal metal [4, 5]. Specifically, the real part of the complex conductivity function
Reσ(ω,T ) for the chain direction and at room temperture T shows the presence
0 0
of a pseudo–gap at ω ≈ 2∆ where ∆ is the zero temperature half gap of the
0 0
corresponding CDW. In addition, a peak structure appears in the far infrared well
below the pseudo–gap. This peak resembles the pinned Fr¨ohlich mode at about the
same frequency but seen in the fully developed CDW. Similar results were also found
in PS films [6].
InprinciplethesefeaturesareknownforalongtimefromtheCDWmaterialK P (CN) Br ·
2 t 4 0.3
3(H O) (KCP) in which fluctuation effects due to intrinsic disorder are prominent.
2
In [7] the ac–conductivity of KCP is modelled by a dielectric function which takes
pinning–unpinning fluctuations into account.
A related explanation for the optical data is provided by the concept of fluctuating
CDW segments which behave similar to the fully developed CDW. These segments
are pinned by random impurities which break the phase invariance of the equation
for the fluctuating order parameter.
In view of the one–dimensional nature of fluctuations above the temperature T∗ >
T below which transverse fluctuations set in to initiate the phase transition, the
P
observed effects could also be a signature of NFL behaviour. The latter seems to be
established for some of the Bechgaard salts [8, 9, 10] where CDW fluctuations are
not important.
In the case of PS, the attractive backscattering and the softening of the Peierls
phonons make a plain Luttinger–liquid scenario unlikely. Voit [11, 12] advocates a
Luther–Emery state [13] for the electrons in blue bronze as explanation of the pho-
toemission spectra. In this case a spin gap would open before true CDW formation.
Such conclusions, however, are not undisputed. Shannon and Joyce [14] argue in
favour of a model [15, 16] for a fluctuating Peierls system (FPS). The unusual plas-
mon dispersion in quasi one–dimensional conductors can also be understood in terms
of a conventional band picture [17]. There seems to be good reason to persue the
FPS concept for one–dimensional metals with CDW fluctuations.
The FPS model has recently been extended by McKenzie [18]. He calculates the
one–particle Green’s function for Gaussian order parameter fluctuations and large
2
correlation length using Sadovskii’s exact method [19]. He points out that the elec-
tronspectral function of FPS is of NFL type asfound earlier in the same context [20].
McKenzie also calculates renormalized coefficients for Ginzburg–Landau functionals
of PS. The model which we will solve for the collective conductivity corresponds to
the FPS concept.
The present paper computes the frequency dependent conductivity from one–dimen-
sional conductive order parameter fluctuations using a modified, linear Ginzburg–
Landau–Langevin (GLL) equation for CDW including phase breaking by impurities.
In superconductivity (SC) where phase pinning does not exist, this part of the con-
ductivitywasfirstinvestigatedbyAslamazovandLarkin[21](AL).TheCDWversion
of the AL–theory was given in [22].
From SC it is known [23, 24] that there are two further contributions from order
parameter fluctuations to the dc conductivity: A resistive contribution from the
reduction of the single particle density of states which is related to the pseudo gap
and the anomalous Maki–Thompson [25, 26] term which is conductive. In CDW the
latter becomes resistive [27] and dominates the dc conductivity near the transition.
The issue of one–dimensional collective dc conductivity in CDW was strongly de-
bated in the seventies [28, 29, 22, 30, 31, 32]. In [28] the idea of paraconductivity
in one–dimensional metals with dominant electron–phonon interaction was advanced
using a GL approach. This result was criticized in [29] where pinned collective fluc-
tuations were shown to reduce the dc conductivity. These authors also studied the
corresponding results for the Hubbard and the Tomonaga–Luttinger model. Detailed
microscopic studies of FPS in [22, 30] find both resistive and conductive fluctuations
but neglect phase pinning. The dc paraconductivity of commensurate FPS was stud-
ied in [32] in a GL context. This paper served as a starting point for the present
work.
The paper is organized as follows: Sec. 2 develops the GLL approach for CDW and
derives the known results for unpinned conductive fluctuations. Sec. 3 introduces a
modified GLL–equation where phase invariance is broken in a way which simulates
local pinning by random impurities, and evaluates the basic correlation function of
the order parameter fluctuations. This result is used to calculate the frequency de-
pendent collective conductivity exactly within the model. The complicated formula
is evaluated approximately for pinning frequencies small in comparison to the fre-
quency of amplitude fluctuations in Sec. 4. Two appendices present mathematical
details.
2 Ginzburg–Landau–Langevin Approach
It is interesting to formulate the problem of fluctuation ac–conductivity in CDW
without pinning using the GLL–method. One expects to find close similarities to
the elegant formulations for SC [33, 34]. However, it turns out that one must go
beyond overdamped dynamics which in case of CDW would only give an instanta-
3
neous response in the current correlation function and thus a frequency independent
conductivity.
Our starting point is the GLL equation directly in terms of the gap fluctuations ∆ :
k
∆¨ (t)+γ ∆˙ (t)+ω2∆ (t) = Γ (t). (1)
k 0 k k k k
The parameters γ and ω are taken from [28, 31]. Static parameters below corre-
0 k
spond to the rigid lattice values in [18]. For simplicity the renormalization of the
rigid lattice (RL) values due to fluctuations as proposed in [18] is not considered. To-
gether with Hartree–Fock corrections to the linear GL–equation as in [35], it would
extend the region of applicability of the linearized approach which is marginal at
the RL–level. This and the absence of resistive fluctuations prohibit a quantitative
comparison with experiments. Below the CDW transition temperature McKenzie’s
approach [18] is neccesary to give the characteristic optical absorption which has
been measured in [36].
The damping constant γ is
0
h¯π
γ = ω2 , (2)
0 A8k T
B
where
ω2 = λω2 (3)
A Q
is the frequency of the amplitude mode of the fully developed CDW [37], ω is the
Q
bare frequency of the 2k –phonon which goes soft, and λ is the electron–phonon
F
coupling constant.
The actual frequency ω of the amplitude fluctuations in (1) is
k
ω2 = ω2 1+k2ξ2 . (4)
k 0
(cid:16) (cid:17)
with
ω2 = ω2ǫ . (5)
0 A RL
For definiteness we assume underdamping (γ < 2ω ) which is the case in the linear
0 0
fluctuation regime sufficiently above the transition temperature.
Without a form of selection rule breaking neither ω nor ω can be observed in
A 0
optical conductivity measurements. The amplitude mode ω can be seen, however,
A
in Raman scattering [38, 39]. The fluctuating amplitude mode above the transition
temperature is observed in neutron scattering studies [40, 41].
The correlation length ξ in (4) is
ξ2 = ξ2/ǫ , (6)
0 RL
with ǫ given by
RL
4
T
ǫ = ln , (7)
RL
T
RL
where T is the rigid lattice mean field transition temperature. The reference length
RL
ξ is given by [31]
0
7ζ(3)h¯2v2
ξ2 = F . (8)
0 16π2(k T)2
B
The complex Gaussian Langevin force Γ has zero mean. Its correlation function
k T
hΓk(t)Γ∗k′(0)i = h|∆k|2i02γ0ωk2δk,k′δ(t) = 2γ0ω02 fB δk,k′δ(t) ≡ Aδk,k′δ(t). (9)
0
is constructed in such a way as to give the fluctuation intensity
k T
h|∆ |2i = B . (10)
k 0
a
k
The latter follows from the linear free energy functional
F = a |∆ |2, (11)
0 k k
k
X
with
Lǫ
a = f 1+k2ξ2 , f = RL. (12)
k 0 0
πh¯v
F
(cid:16) (cid:17)
Note that (1) implies a spatial correlation of the order parameter according to
L k Tπh¯v
h∆(x,0)∆∗(0,0)i = dkeikxh|∆ |2i = B F e−|x|/ξ ≡ ψ2 e−|x|/ξ. (13)
2π Z k 0 2ξǫRL ! RL
In the next step the one–dimensional conductivity is computed from the classical
Kubo formula
L ∞
σ(ω) = dteiωthJ(t)J(0)i, (14)
k T
B Z0
where L is the sample length.
The collective current density was calculated in [32] 1 and reads:
b
j(x,t) = i (∆˙ (x,t)∆∗(x,t)−∆(x,t)∆˙ ∗(x,t)). (15)
2
1S.N. Artemenko informed the author that he obtained the same expression for the collective
current density by the Keldysh approach to CDW dynamics.
5
The collective current is proportional to the time derivative ϕ˙ of the order parameter
phase as for the fully developed CDW but the prefactor is different. The coefficient
b in (15) is [31]
e 2
b2 = 0 , (16)
2k Th¯ν
(cid:18) B b(cid:19)
involving the backward scattering rate ν due to random static scattering centers.
b
This formula holds in the pure limit when the electron scattering rate obeys h¯ν ≡
h¯(ν +ν /2) < 2πk T.
f b B
The homogeneous current density J in (15) is related to j by
1 L
J(t) = dxj(x,t) = j (t). (17)
k=0
L
Z0
In the linear setting of (1) not only Γ obeys Gaussian statistics but also ∆ and exact
Gaussian decoupling gives
b2
hJ(t)J(0)i = C˙(k,t)2 −C(k,t)C¨(k,t) , (18)
2
Xk h i
provided the correlation function
C(k,t) ≡ h∆ (t)∆∗(0)i (19)
k k
is real and even in t. C(k,t) is evaluated from (1) and explicitly given by
γ γ
C(k,t) = h|∆ |2i exp(− 0|t|) cosD t+ 0 sinD |t| , (20)
k 0 k k
2 2D
(cid:20) k (cid:21)
with
γ2
D = ω2 − 0. (21)
k s k 4
This leads to
C˙2 −CC¨ = h|∆ |2i2 ω2 exp(−γ |t|). (22)
k 0 k 0
Note that all oscillating terms in the correlation functions cancel out leaving a purely
relaxational response. If one were to use the correlation function
C(k,t) = h|∆ |2i exp(−γ |t|), (23)
k 0 k
for the overdamped version of (1) with γ = ω2/γ one would get an instantaneous
k k 0
response hJ(t)J(0)i ∝ δ(t) and hence a frequency independent conductivity.
6
Calculation of the conductivity using (22) gives, however, the correct result given in
[35]
γ2
Reσ(ω) = σ 0 , (24)
F γ2 +ω2
0
irrespective of the relation between γ and ω . The scale–value σ of the fluctuation
0 0 F
conductivity is
L2A4b2
σ = , (25)
F 16k Tω2ξγ3
B 0 0
and coincides with the result [31]. Explicitly σ reads
F
2π2e2k Tv
σ = 0 B F . (26)
F
7ζ(3)ǫ (h¯ν )2
RL b
q
−1/2
The conductivity shows the mean–field critical behaviour σ ∝ ǫ . In the picture
F RL
of a metal with order parameter fluctuations the collective conductivity adds to the
normal state conductivity σ = 8e2v /(4πh¯ν ).
N 0 F b
Formal calculations in higher spatial dimensions require a momentum cut–off in con-
trast to SC. This is related to the form of the collective current density (15).
3 Breaking of Phase Invariance
The space–time version of (1) is
∂2
∆¨(x,t)+γ ∆˙ (x,t)+ω2 1−ξ2 ∆(x,t) = Γ(x,t). (27)
0 0 ∂x2!
The simplest way to break the phase invariance of this equation is to add a pinning
term
2ω2|∆(x,t)| cosϕ(x,t), (28)
i
to the left hand side which is a simple local coupling. This is clearly not the general
startingpointtotreatpinning byrandomimpurities [42,43]. However, itwillbecome
evident later that this approach simulates local pinning because the final pinning
frequency is proportional to the impurity concentration.
To arrive at (28) we start from the more general form
ω2 h(x−x )(∆(x ,t)+∆∗(x ,t)). (29)
s i i i
i
X
Theimpuritiesarelocallycoupledtotheorderparameter. Therealstructurefunction
h(x)transmitstheeffecttotheorderparameteratx. Theterms∆∗(x ,t)breakphase
i
7
invariance by modelling backward scattering. The scale frequencies ω and ω are
s i
different from the final pinning frequency.
We make two further asumptions: The function h is a contact interaction h(x) =
l δ(x) with a scattering length l . The crudest assumption is, however,
i i
→ n dx , (30)
i i
i Z
X
where n is the density of impurities. This requires n ξ > 1 and amounts to an early
i i
impurity average. Introducing the scale frequency
ω2 ≡ ω2n l (31)
i s i i
then leads to (28).
This admittedly crude model has the advantage to allow for an exact solution.
The complete GLL–equation which replaces (1) can be written:
∆¨ (t)+γ ∆˙ (t)+ω2∆ (t)+ω2(∆ (t)+∆∗ (t)) = Γ (t). (32)
k 0 k k k i k −k k
The complex order parameter ∆(x,t) is decomposed into real and imaginary parts
U(x,t) = Re∆(x,t) and V(x,t) = Im∆(x,t) giving
∆ = U +iV , (33)
k k k
with complex U and V which satisfy the usual reality conditions
k k
U = U∗, V = V∗. (34)
−k k −k k
The Langevin equation splits into two equations for U and V :
k k
U¨ (t)+γ U˙ (t)+ω2U (t)+2ω2U (t) = Γ (t), (35)
k 0 k k k i k Uk
V¨ (t)+γ V˙ (t)+ω2V (t) = Γ (t).
k 0 k k k Vk
Here Γ (t) and Γ (t) are the Fourier transforms of the real and imaginary part of
Uk Vk
Γ(x,t), respectively. It is possible that the impurities modify the thermal random
force Γ(x,t). In our model we assume that this is not the case. This assumption
is reasonable for ω ≪ ω . The random forces Γ and Γ are then independent
i 0 Uk Vk
Langevin forces with the same statistical properties as Γ (t) (cf. (9)) but only half
k
its strength. The two equations (35) become independent and are both isomorphic
with (1). However, the frequencies for the U–modes are modified and change their
fluctuation intensities:
ω2 k T k T
h|U |2i = (1+2 i )−1 B , h|V |2i = B . (36)
k ω2 2a k 2a
k k k
8
Hence the intensity of the fluctuating order parameter is reduced:
1+(ω /ω )2
h|∆ |2i = h|∆ |2i i k . (37)
k k 01+2(ω /ω )2
i k
A thermodynamic derivation for (36) is given in Appendix A. In view of the real-
istic condition ω ≪ ω the renormalization of the mean square order parameter is
i k
irrelevant.
The order parameter correlation function becomes
C(k,t) = p C (k,t)+p C (k,t), (38)
+ + − −
with weights
1 1
p = , p = . (39)
+ 2(1+2(ω2/ω2)) − 2
i k
and the replacement
γ2 γ2
D → D(+) = ω2 +2ω2 − 0 ≡ (ω(+))2 − 0 (40)
k k s k i 4 s k 4
in the expression (20) for C(k,t) in order to get C (k,t) while C (k,t) remains
+ −
(−) (−)
unchanged, i.e. formally D = D , ω = ω . This solves completely the GLL–
k k k k
equation (32).
4 Discussion of Fluctuation Conductivity
We define the wave–number dependent pinning frequency
ω2
ω (k) ≡ i . (41)
p
D
k
From (31) it is seen that ω (k) is proportional to the linear impurity concentration.
p
Thus our model simulates local pinning.
Using the condition ω ≪ D the result of Appendix B leads to the following expres-
i k
sion for the real part of the fluctuating conductivity
L b2 γ
Reσ(ω) = h|∆ |2i2ω2 0 (42)
kBT 4 k k 0 k("γ02 +ω2#AL
X
9
γ2 +ω2(k)+ω2
+ γ 0 p
 0 2 
γ2 +ω2 +ω2(k) −4ω2ω2(k)
0 p p
 P
 (cid:16) (cid:17) 
ω2(k)(3−γ2/ω2)(γ2 +4D2)−ω2
+ γ p 0 k 0 k .
" 0 4D2 (γ2 +ω2 +4D2)2 −16ω2D2# )
k 0 k k A
Even before the k–summation is performed three different contributions to the fluc-
tuation conductivity can be discriminated: the relic of the AL–conductivity (AL)
centered at zero frequency, a pinned mode (P) near the frequency ω (0), and a weak
p
structure (A) associated with the fluctuating amplitude modes ω . The latter results
k
from selection rule breaking by the impurities. Thus traces of the fluctuating ampli-
tude mode should be seen in the optical conductivity. A similar case regarding the
pinned Frhlich mode is found in the fully developed CDW [44, 45].
The k–summation is easily done for the AL–part and gives
1 γ2
Reσ(ω) = σ 0 . (43)
AL 2 F γ2 +ω2
0
This is exactly half the result (24). The spectral weight W according to
∞
W ≡ dωReσ(ω) (44)
Z−∞
is
π
W = γ σ . (45)
AL 0 F
2
A lenghty but exact calculation gives for the P–mode
Reσ(ω) = σ [J (ω)+2ReJ (ω)], (46)
P F 1 2
with
1 γ2 +ω2 −4ω4/γ2
J (ω) = γ2 0 i 0 , (47)
1 2 0 ω2 +[γ +2ω2/γ ]2 ω2 +[γ −2ω2/γ ]2
0 i 0 0 i 0
(cid:16) (cid:17)(cid:16) (cid:17)
and
ω ω4 1
J (ω) = 2 0 i . (48)
2 γ0 4ωi4/γ02 −(γ0 +iω)2 (4ω2−γ2)(γ +iω)2 +4ω4
0 0 0 i
q
Note the non–algebraic structure of the conductivity due to J . The spectral weight
2
associated with Reσ(ω) is independent of pinning parameters and given by
P
π
W = γ σ . (49)
P 0 F
2
10