Skip to main content

Full text of "Interacting electrons in disordered wires: Anderson localization and low-temperature transport"

See other formats


Interacting electrons in disordered wires: Anderson localization and low-T transport 



o 
o 

(N 



I.V. Gornyii'*, A.D. Mirlini-^'t, and D.G. Polyakovi^* 
^ Institut fiir Nanotechnologie, Forschungszentrum Karlsruhe, 76021 Karlsruhe, Germany 
^Institut fiir Theorie der kondensierten Materie, Universitdt Karlsruhe, 76128 Karlsruhe, Germany 

(Dated: May 20, 2005) 

We study transport of interacting electrons in a low-dimensional disordered system at low tem- 
perature T. In view of localization by disorder, the conductivity (j{T) may only be non-zero due 
to electron-electron scattering. For weak interactions, the weak-localization regime crosses over 
with lowering T into a dephasing-induced "power-law hopping". As T is further decreased, the 
Anderson localization in Fock space crucially affects <j{T), inducing a transition at T = Tc, so that 
a{T < T^) = 0. The critical behavior of a(r) above is Ino-(r) oc -(T-Tc)"^/^. The mechanism 
of transport in the critical regime is many-particle transitions between distant states in Fock space. 

PACS numbers: 72.20.-1, 72.15.Rn, 71.30.-fh, 73.63.-b 



I 



I 

o 
o 



o 
o 



o 
o 



X 
J3 



In a pathbreaking paper Q Anderson demonstrated 
that a quantum particle may become localized by a ran- 
dom potential. In particular, in non-interacting systems 
of one-dimensional (ID) or two-dimensional (2D) geom- 
etry even weak disorder localizes all electronic states 
thus leading to the exactly zero conductivity, cr(T) = 0, 
whatever temperature T . A non-zero cr(T) in such sys- 
tems may only occur due to inelastic scattering processes 
leading to dephasing of electrons. Two qualitatively dif- 
ferent sources of dephasing are possible: (i) scattering of 
electrons by external excitations (in practice, phonons) 
and (ii) electron-electron (e-e) scattering. In either case, 
at sufficiently high temperatures, the dephasing rate 
is high, so that the localization effects are reduced to a 
weak-localization (WL) correction to the Drude conduc- 

1/2 

tivity. This correction behaves as In in 2D and as 
in quasi-lD (many-channel wire) systems and thus 
diverges with lowering T , signaling the occurrence of the 
strong localization (SL) regime. This prompts a question 
as to how the system conducts at low T. 

For the case of electron-phonon scattering the an- 
swer is well known. The conductivity is then governed 
by Mott's variable-range hopping (VRH) [3], yielding 
cr(T) (X exp{-(To/r)^} with /.t = l/{d+l), where d is the 
spatial dimensionality. In the presence of a long-range 
Coulomb interaction, the Coulomb gap in the tunneling 
density of states modifies the VRH exponent, /i = ^ 0]. 

But what is the low-T behavior of (j{T) if the electron- 
phonon coupling is negligibly weak and the only source 
of the inelastic scattering is the e-e interaction? Our pur- 
pose here is to solve this long-standing fundamental prob- 
lem, which is also of direct experimental relevance; see, 
e.g., Refs. and 0,05 where the crossover from WL to 
SL with lowering T was studied for ID and 2D systems, 
respectively. For definiteness, we concentrate on the case 
of a many-channel ID system with a short-range interac- 
tion. Our results are, however, more general (including 
single-channel wires, 2D systems. Coulomb interaction), 
as we discuss in the end of the paper. 

It was proposed in |^ that the e-e interaction by it- 



self is sufficient to induce VRH at low T. This idea was 
widely used for interpretation of experimental [1,01 and 
numerical Tl'l results on 2D systems. Further, Ref. 
used bosonization to study the problem in ID and con- 
cluded that transport is of VRH character. These results 
are, however, in conflict with the argument [l^ - sup- 
ported by our analysis - that elementary hops in the 
low-T limit are forbidden for d < 3 even for the case of 
long-range (1/r) Coulomb interaction, since energy con- 
servation cannot be respected when an electron attempts 
a real transition by exciting an electron- hole pair 
The situation is particularly interesting in ID and 2D, 
where no mobility edge exists, activation to which other- 
wise might give cr{T) ^ 0. If neither VRH nor activation, 
then what? 

Let us now specify the model. We consider a many- 
channel weakly disordered wire, so that the relevant 
length scales satisfy kp"^ <^ I <^ ^, where kp is the Fermi 
momentum, I the mean free path, and ^ ~ nvD the local- 
ization length (ly is the density of states per unit length 
and D the diffusion constant) 0, . The correspond- 
ing energy scales are the Fermi energy Ep, the elastic 
scattering rate r^^, and the level spacing in the local- 
ization volume, A J — with Ep » A^. 
We will assume a short-range interaction U{y — r') be- 
tween electrons, characterized by a dimensionless cou- 
pling OL = z/[/(0), where ?7(q) is the Fourier transform 
of C/(r). We assume that a ^ 1, which yields a richer 
behavior of o'(T) and allows better understanding of un- 
derlying physics; the case a ~ 1 (as well as Coulomb 
interaction) will be discussed in the end. 

At sufhciently high T, the conductivity o-(T) ^(70 + 
AcTwL + AtJAA is close to its Drude value ctd , with quan- 
tum corrections related to the weak localization (Actwl) 
and to interplay of interaction and disorder (Altshuler- 
Aronov contribution Actaa) j2t], 

1/3 

• (1) 



|A(twl| 



CTD 



dq 



Here we used the result for the dephasing rate length 



2 



= {Dtci,)^/'^ due to e-e interaction Q, 



The WL correction grows with lowering T and finally 
becomes strong (A(Twl/co ^ 1) when reaches ^, or, 
equivalently, when r^^ ~ A^. This happens at T ~ Ti = 
a~^Aj, marking the beginning of the SL regime. The 
interaction-induced correction Actaa /cd (a^A^/T)^/^ 
remains small at T ~ Ti and thus is of no relevance in 
the present context. (For a 1, Actaa is of order ctd 
at T ~ Ti and does not lead to any qualitative changes 
either.) The subject of our interest is cr{T) for T <Ti. 

In fact, SL does not necessarily mean a{T) is expo- 
nentially small. Specifically, in the high-T part of the SL 
regime the transport mechanism - we will call it power- 
law hopping (PLH) - is analogous to the one identified 
in [l^l for the case of inelastic electron-phonon scatter- 
ing, i.e., hopping over length '--^ ^ in time ~ r^. In other 
words, the dephasing time serves in this regime as a 
lifetime of localized states, which adds an imaginary part 
i/2T^ to the single-particle energies £„. This yields 

a{T) ^ aacm\n=^/r, e''^e/r4>, (3) 

where crac(f^) is the zero-T conductivity of noninteracting 
electrons at frequency f2. The crucial point here is that 
in this SL regime can still be calculated via Fermi's 
golden rule, as we are going to show. The lowest-order 
decay process of a localized state |a) is the transition 
to a three-particle state - two electrons \(3), I7) and a 
hole I (5), all located within a distance ^ Fig. The 
corresponding matrix element of the interaction is 

Vapjs = V aA^; |e„ - e^l, je^, - e^l < Aj (4) 

and decays fast for larger energy differences (cf. results 
for a metallic sample [lalisj l with the dimensionless con- 
ductance set to g '--^ 1). Since energies of all the relevant 
single-particle states are within the window of width ~ T, 
the level spacing of three-particle states to which the orig- 
inal state I a) is coupled according to reads 

Af^Ayr. (5) 

Using |@J, (0 and the golden rule, we find 

^ \V\yAf ^ a'T. (6) 

Note that this result could also be obtained from ((SJ if 
is used as the infrared cutoff, as appropriate for the 
SL regime. The condition of validity of the golden-rule 
calculation is F ^ A^^\ or, equivalently, t^^ ^ A^'^''. 
This introduces a new temperature scale T3 = q;~^Aj, 
so that the PLH regime is restricted to the range ^ 




FIG. 1: Diagrams for the golden rule (a) and higher-order 
decay amplitude (b). 

T <C Ti. Combining ^ and we get the conductivity 
in this regime, 

cr(T) - e^z^C^a^T - e^^T/Ti. (7) 

What happens below T3? Simple hops on a distance 
~ f are then not sufhcient to delocalize electrons. In- 
creasing the distance r of a hop does not help: the matrix 
element vanishes exponentially with r/^, while the level 
spacing A^^^^ only as a power law. We thus have to ana- 
lyze higher-order processes by exploring the structure of 
the theory in the many-body Fock space, similarly to the 
ideas developed for the problem of a quasiparticle decay 
in quantum dots 18, 19j,20i]- The process of n-th order 
represents a transition with excitation of n electron-hole 
pairs, \a) — > = |/3o/3i/3i • ■ . /?„/3„) [with the en- 

ergy difference < A^ for each pair in view of Q], see 
Fig. 1213. Let us estimate the dimensionless coupling of 
ri-th order (ratio of the matrix element to the level spac- 
ing of final states), I/(")/A(^"+^\ which is the n > 1 
generalization of the ratio V/A^ considered above. In 
this estimate, it will be sufficient for us to keep track of 
factors of the type n" (or nl) and {T/T^)^. Factors of 
the type c", where c ^ 1, will be unessential and thus 
neglected. 

Let us assume that the set of states _B^"^ is spread over 
the length m^, with ^ n/m pairs in each box of length 
^; later we optimize with respect to m (for I m < n). 
The corresponding level spacing can be estimated as 

Ag+^'/Ae - [(n/m)Ajr]". (8) 
The matrix element T4;,3i,...,/3„+i is given by 

diagrams 71 ,■■ -7^- 1 i—1 * "^^ 

where the matrix element Vi corresponds to the i-th in- 
teraction line, ji are the virtual states corresponding to 
internal lines, and Ei the corresponding energy variable 
which can be expressed as a linear superposition of 
using the energy conservation. We need to take into ac- 
count only those contributions where all states forming 
each matrix element are within a distance ~ ^ from each 
other, so that Vi ~ aA{. Therefore, the summation over 
each 7i is effectively taken over a single localization do- 
main, and we can replace it by taking the "optimal" 7;, 



3 



o 



o 



o 



FIG. 2: Scheme of a high-order balhstic process: the string of 
particle-hole pairs (black and white circles). The initial state 
and intermediate virtual states (7^) are shown in grey. 



with \Ei — ey. \ ^ A^. We thus get 

y(")/A5 ^ [M^")]i/2(aA^)", (10) 

(n) 

where Mm is the number of diagrams contributing to 
the amplitude of the transition \a) — > These con- 

tributions have random signs, hence the factor [Mm^]^/"^. 

(n) 

To find Mm , we first calculate the number of topo- 
logically different diagrams, D'-"-\ which satisfies the re- 
cursion relation 

= £)("l)£)(«2)£,("3)^ (11) 

ni-t-n2-t-n3— n — 1; ni,2,3>0 

with the initial condition L*'"^ = 1. It is easy to show 
that its solution increases only as Z?^") ~ a", with a ~ 1, 
so that with the required accuracy it can be replaced 



by unity. Thus M, 



(") 



Am\ the number of allowed 



permutations of the set B*^"' over the final-state lines. 
To estimate Am'' , we notice that only electron-hole pairs 
within the same (or nearby) localization volume can be 
interchanged, which yields Am^ ~ [(rt/m)!]™ ~ [n/m)"'. 
Combining this with (jHJ, we finally get 



(12) 



Let us now analyze the result. First of all, for given n 
the most favorable case is m ^ n, which corresponds to 
"ballistic" paths. In such a process, an electron makes a 
many-body transition over the distance n^, leaving be- 
hind a string of n particle-hole pairs, as illustrated in 
Fig- 13 Second, y(")/A(^"+-^) increases with n at suffi- 
ciently high T and decreases with n (thus remaining small 
for all n) at low T. Therefore, at low T the higher-order 
processes do not help a localized single-particle state to 
decay, so that cr(T) is exactly zero. In contrast, at high 
T, the increase of the coupling (|12|l with n guarantees 
that the golden-rule calculation performed for the WL 
and PLH regimes is not spoiled by the higher-order ef- 
fects. The temperature Tc of the transition into the zero- 
conductivity regime can be immediately estimated from 
(fT^ . Tc ~ A^/a. In fact, (fT^ misses a Ina"^ factor 
recovered in a more accurate treatment below, Eq. (|14|l . 

What helps us to analyze the critical behavior at the 
transition is that the structure of the theory, when re- 
stricted to the optimal ("balhstic") paths, reduces essen- 
tially to that of the Anderson model on the Bethe lattice 
- a tree with a fixed branching number. Indeed, consider 



the process shown in Fig.|21 an electron hops to an adja- 
cent localization volume, creating an electron-hole pair, 
then to the next one and so forth. Clearly, the density of 
final states increases at each step by the same factor 



(13) 



which is the branching number of the Bethe lattice. The 
Bethe-lattice character of the problem can also be in- 
ferred from the exponential dependence of the coupling 
(|12|l on n at m = n. This should be contrasted with 
the opposite limit, m — 1, corresponding to the case of 
a quantum dot, when (|12ll contains an additional ri~"/^ 
factor. The latter is related to a decrease of the effective 
branching number with n in this case, as was noticed in 
[1^ Eo|. In other words, the mapping on the Bethe lat- 
tice model, oversimplified for a quantum dot (for which 
it was originally proposed in |0| ) , turns out to work per- 
fectly in the case of localized states in a wire (or, more 
generally, in a non- restricted geometry), where going to 
higher generations in Pock space can be accompanied by 
the exploration of new regions in real space. 

We can now use the results for the Anderson transition 
on the Bethe lattice that has been studied extensively 
[ilEllIllilliil. For a large branching number K the 
equation for the transition point reads 



A/V = 4lnK, 



(14) 



where V is the hopping matrix element, and A is the 
mean level spacing of states of generation n + 1 coupled 
to a given state of generation n. Using 10}, (|13|l . and 



A = A 



we find the transition temperature, 
Tc ^ A^/alna"^, 



(15) 



so that Tc ^ Ts/lna"^ The critical behavior of a{T) 
above Tc is governed by that of the decay rate , which 
translates into the imaginary part of the self-energy Im S 
for the Bethe-lattice problem. The critical behavior of 
the latter was found in 0,11^, yielding 

<7{T) oclm E oc exp{-c„[(T - T,)/Tj-i/2}, (16) 

with Ca ~ Ina^^ for a 1. Near the transition, the 
local density of states on the Bethe lattice acquires an 
increasingly more sparse "spatial" structure jjjl, 123 , so 
that the transport is governed by processes connecting 
remote states in Fock space. Fig. El which implies a glassy 
character of the system. When T — > Tc, the length of 
the particle-hole strings diverges. We term the low-T 
phase "Anderson- Fock glass" (AFG), since its physics is 
governed by the Anderson localization in Fock space. 

The found behavior of cr(T) is illustrated in Fig. |21 If 
a ~ 1, all scales become of the same order, Ti ~ Ta ~ 
Tc ~ A^, and the range of PLH disappears. In realistic 
systems, weak coupling to phonons will lead to PLH (for 



4 



a(T) Dmde 




FIG. 3; Schematic behavior of ij{T) on the log-log scale: 
the weak-localization (WL), power-law hopping (PLH), and 
Anderson-Fock glass (AFG) regimes, with a localization tran- 
sition at Tc. Dashed line: PLH and VRH contribution to (j(r) 
in the case of a weak coupling to phonons. 

a 1) and VRH below but with a small prefactor, 
so that the transition at Tc should be well observable. 
Note the peculiar character of the transition: not only 
the exponential critical behavior IjlGI) but also that the 
ordered (metallic) phase corresponds to T > Tc- An 
apparent conflict with the Mermin- Wagner theorem is 
related to the unconventional (functional) nature of the 
order parameter for Anderson localization 0, . 

Before closing the paper, let us briefly mention a few 
extensions of our results [2J|. 

(i) ID: Single channel. We have recently shown [25j 
that the notion of WL and dephasing are also applicable 
to a disordered Luttinger liquid and calculated cr{T) in 
the WL regime. The low-T results presented above can 
then be easily generalized to the single-channel case. An 
important difference is that r (and thus, is strongly 
rcnormalizcd by interaction, t[T)/t (T/Ep)". 

(ii) rf > 1. Generalization to 2D systems is straight- 
forward, with ^ depending exponentially on disorder, 
^ oc eicp{2Tr^h'D). Our consideration is also applicable to 
3D Anderson insulators; however, in this case cr(T < Tc) 
will not be exactly zero but rather determined by the 
activation above the single-particle mobility edge. 

(iii) Coulomb interaction. For ID and 2D geometry, 
the transition survives also for 1/r Coulomb interaction, 
since correlated hops of two electrons separated by a 
large distance r ^ ^ will not delocalize them. Indeed, 
the corresponding matrix elements decrease with r as 
VaiSjS OC which is not compensated by the increase 
c>c r"^ of the density of final states for d < 3 and thus 
does not help. On the other hand, in 3D such processes 
will lead to delocalization for any T the behavior of 
(t(T) in this case requires a separate study. 

(iv) Creep. Our approach can be used to analyze the 
non-linear conductivity cr(i?) at weak electric field E. 



In conclusion, we have studied the conductivity of in- 
teracting electrons in a disordered quantum wire; very 
similar results hold for a 2D system. In contrast to a 
popular belief, the e-e interaction is not sufficient to sup- 
port the VRH transport in the low-T limit. Instead, the 
system undergoes a localization transition at tempera- 
ture Tc, below which cr(T) = (assuming vanishing cou- 
pling to phonons). We have shown that the conductivity 
vanishes as In cr(T) cx -(T - Tc)"!/^ as T ^ Tc. Trans- 
port in the critical regime is governed by many-particle 
transitions between distant states in Fock space, corre- 
sponding to the formation of long strings of electron-hole 
pairs. For weak interactions, this Anderson-Fock glass 
phase is separated from the WL regime by an intermedi- 
ate temperature regime of power-law hopping. 

We thank V. Cheianov, D. Maslov, G. Minkov, T. Nat- 
termann, and B. Shklovskii for valuable discussions. We 
are particularly grateful to I. Aleiner, B. Altshuler, and 
D. Basko for very useful discussions and criticism of the 
earlier version of this work (cond-mat/0407305ifl). This 
helped us to correct an error in counting of the number 
of diagrams M^^ [which yielded a spurious double expo- 
nential tail of ct(T) in the AFG phase] , whose elimination 
gav e a true phase transition at T = Tc, as also found in 
[26|. The work was supported by SPP "Quanten-Hall- 
Systeme" and CFN of the DFG and by RFBR. 



[*] Also at A.F. loffe Physico- Technical Institute, 194021 
St. Petersburg, Russia. 

[f] Also at Petersburg Nuclear Physics Institute, 188350 
St. Petersburg, Russia. 

[1] P.W. Anderson, Phys. Rev. 109, 1492 (1958). 

[2] E. Abrahams et al, Phys. Rev. Lett. 42, 673 (1979). 

[3] B.L. Altshuler and A.G. Aronov, in Electron- Electron In- 
teractions in Disordered Systems, edited by A.L. Efros 
and M. PoUak (North-Holland, Amsterdam, 1985). 

[4] N. Mott, Metal-Insulator-Transitions (Taylor & Francis, 
London, 1974). 

[5] B.I. Shklovskii and A.L. Efros, Electronic properties of 
doped semiconductors (Springer, Berlin, 1984). 

[6] M.E. Gershenson et al, Phys. Rev. Lett. 79, 725 (1997); 
Yu.B. Khavin et al, Phys. Rev. B 58, 8009 (1998). 

[7] S.Y. Hsu, J.M. Valles, Phys. Rev. Lett. 74, 2331 (1995); 

F. W. van Keuls et al, Phys. Rev. B 56, 13263 (1997); 

G. M. Minkov et al, Phys. Rev. B 65, 235322 (2002). 
[8] S.I. Khondaker et al, Phys. Rev. B 59, 4580 (1999). 
[9] L. Fleishman et al., Phys. Rev. Lett. 40, 1340 (1978). 

[10] I. Shlimak et al. Solid State Comm. 112, 21 (1999); 

A. Ghosh et al, Phys. Stat. Sol. (b) 230, 211 (2002). 
[11] R. Berkovits and B.I. Shklovskii, J. Phys. Cond. Matter 

11, 779 (1999); P.H. Song and D.L. Shepelyansky, Phys. 

Rev. B 61, 15546 (2000). 
[12] T. Nattermann et al., Phys. Rev. Lett. 91, 056603 (2003). 
[13] L. Fleishman and P.W. Anderson, Phys. Rev. B 21, 2366 

(1980). 

[14] Recently, S.Y. Malinin et al, Phys. Rev. B TO, 235120 
(2004), argued that the VRH of Ref . H indeed requires 
a finite coupling to the phonon bath. 

[15] K.B. Efetov, Supersymmetry in Disorder and Chaos 
(Cambridge University Press, 1997). 



5 



[16] A.D. Mirlin, Phys. Rep. 326, 259 (2000). 

[17] A.A. Gogolin, V.I. Melnikov, and E.I. Rashba, Sov. Phys. 
JETP 42, 168 (1975); A.A. Gogolin and G.T. Zimanyi, 
Solid State Comm. 46, 469 (1983). A similar transport 
regime in the context of a dynamical localization was 
found in D.M. Basko, Phys. Rev. Lett. 91, 206801 (2003). 

[18] B.L. Altshuler et at, Phys. Rev. Lett. 78, 2803 (1997). 

[19] A.D. Mirlin and Y.V. Fyodorov, Phys. Rev. B 56, 13393 
(1997). 

[20] P.G. Silvestrov, Phys. Rev. Lett. 79, 3994 (1997); Phys. 
Rev. E 58, 5629 (1998). 



[21] R. Abou-Chacra et al, J. Phys. C 6, 1734 (1973). 

[22] M.R. Zirnbauer, Phys. Rev. B 34, 6394 (1986); Nucl. 

Phys. B 265, 375 (1986); A.D. Mirlin and Y.V. Fyodorov, 

Nucl. Phys. B 366, 507 (1991). 
[23] A.D. Mirlin and Y.V. Fyodorov, Phys. Rev. Lett. 72, 526 

(1994); J. Phys. I France 4, 655 (1994). 
[24] I.V.Gornyi, A.D.Mirlin, and D.G.Polyakov, unpubhshed. 
[25] I.V. Gornyi et al., cond-mat/0407305 
[26] I.L.Aleiner, B.L. Altshuler, and D.M. Basko, unpublished.