Magneto Conductance Sign Change
Magneto Conductance Sign Change
Magneto Conductance Sign Change
15 FEBRUARY 1994-II
sign
Y. Levinson
Department
and A.
G. Aronov*
of Physics,
The distribution of the hopping probabilities in the strong localization regime is calculated via the independent-directed-path formalism, allowing for level broadening of the intermediate site energies of the hopping process. The distribution belongs to the Gaussian unitary ensemble even in the absence of a magnetic field, whereas in the absence of level broadening it belongs to the orthogonal ensemble. In the presence of the field it leads to an orbital magnetoconductance quadratic in the field, changing from positive to negative as the temperature is lowered and the number of paths increases.
I. INTRODUCTION
The study of the magnetoconductance (MC) in the weak-localization regime has shown that it is related to the interference processes in such systems. There is no of the magnetotransport similar detailed understanding in the strong localization regime. However, descriptions of the MC at low fields, which are based on the interference among the tunneling paths, seem to be successful in explaining experimental data of highly localized samples. This observation has led to investigations of the consequences ensuing modifications in the interference pattern, e.g. , due to spin-orbit scattering. This paper addresses the interference-induced MC of strongly localized electrons when level broadening, due to phonon scattering, of the on-site energies along the tunneling paths is taken into account. The low-temperature conductance in the strong localization regime is dominated by thermal hopping. When electronic correlations are not significant, it obeys Mott's variable-range-hopping (VRH) law, g, -exp[ Tp/ ( T)'~' +"], in two and three dimensions (d =2, 3) with hopping distance R typically on the order of several localization lengths (depending on the temperature and the amount of disorder). The interference description, first invoked by Nguyen, Spivak, and Shklovskii, is based upon the latter fact. They pointed out the importance of the interference among various paths associated with the hopping process, which are contained within a cylindershaped domain of length R and width (Rg)', where g is the localization length. They found that in the presence of a magnetic field the modifications in the interference pattern considerably affect the hopping probability. the logarithm of the hopping Averaging numerically probability over many random impurity realizations, they obtained under certain conditions a positive MC, linear in the field at weak fields. The picture of Nguyen, Spivak, and Shklovskii' refers to a single hopping process. Its relationship to the macroscopic conductance is based on the analysis of the conducting properties of a sample in the VRH regime in
0163-1829/94/49(8)/5165(7)/$06. 00
terms of an equivalent resistor network. ' In that description, any two sites between which the electron hops are connected by a conductance g,&
g (I= g py |/exp
r t//g + e~@/k T] ) [
where go has units of conductance, r & is the distance between the two sites, and e f2(~'E''~+~EI~+[6; E'I)), e; and e& being the initial and final site energies measured from the Fermi level. In (1), y denotes the renormalization of the effective overlap probability of the wave functions localized at the two sites, due to the interference effect. The effective overlap results from electron tunneland contains the effects of ing between the sites i and the interference among the various possible paths. The macroscopic conductance is determined by the "critical" conductances g, which connect sites whose spatial separation is R g and whose site energy difference is kT. Alternatively stated, the conducting properties are dominated by hops of elementary conductance g, which span a critical network throughout the system. ' Deep in the strong localization regime the critical conductance may be obtained by treating separately the exponential factor and the interference factor in (1), since the scale of the r,, dependence of y, , is large compared with g. (The change of the localization length, b, g, due to the interference is small compared with g. ) The analysis of the exponential factor leads to Mott's law giving the essential temperature dependence of the conductance in the VRH regime, with R -g(TQ/T)' and e-kT(TQ/T)'~. The contribution of the interference term is given' ' by exp[(lny)], i.e., by averaging over the logarithm of the overlap probability. It follows that the interference aspect of the hopping conduction is determined by the distribution of the wave-function overlaps in the highly disordered system. The latter is particularly sensitive to the effect of a constant magnetic field H. At weak-enough fields, such that the flux in the cylinder-shaped domain does not exceed pp hc /e is the quantum flux unit), the only orbital (QQ effect of the field is to multiply the overlap amplitude of
e))
49
5165
1994
5166
O. ENTIN-%'OHLMAN,
each tunneling path by a magnetic phase factor. The interference pattern, and, consequently, the distribution of the overlaps, are then modified and the logarithmic average is changed. This is the source of the orbita1 MC in the VRH regime. ' lt should be noted that this description fields at effects occurring higher disregards (H ~ H, =go/g ) such as the shrinking of the wave functions or modifications in the localization length. The distribution of the wave-function overlaps has been calculated using various numerical and analytic ap' ' In particular it has been found that proximations. it leads to positive MC in the presence and in the absence contrary to the situation in of spin-orbit scattering, the weak-localization regime. Recently, the relationship between the distribution and the random-matrix theory of transition strengths has been established. ' Here we formalism to obemploy the independent-directed-path tain the distribution of the overlaps when the on-site energies at the intermediate impurity sites along the tunneling paths may be broadened due to electron-phonon scattering. The effect is introduced by adding a small imaginary part to the site energies. We find that the resulting overlap distribution belongs to the Gaussian unitary ensemble (cf. Refs. 11 and 12) even in the absence of a magnetic field. The reason is that this ensemble describes systems in which time-reversal symmetry is broken; in our case by level broadening. We use the overlap distribution to calculate the logarithmic average and to obtain the effects of level broadening upon the MC in the strong localization regime. The result is that for a small amount of broadening the MC is positive. However, its functional dependence upon the field at weak fields changes from linear (in the absence of the broadening" ) to quadratic. Surprisingly enough, the effect of the level broadening becomes more pronounced as the number of paths increases, i.e., as the temperature decreases and the hopping distance becomes longer. Then the sign of the MC becomes negative. The formal considerations and details of the calculations are given in the next two sections. Section IV and presents the results for the magnetoconductance their discussion.
V= I' ll
V
k
V W
ek)/W (e,
'
' '
'"
where the product is over the L impurity sites along the path. Thus the random part of the amplitude is represented by the last product in Eq. (2). As L is the length of the path, the prefactor in (2) gives rise to the term exp( r,// in the elementary conductance. To 2g) a'ccount for level broadening of the intermediate site energies, we add to the energies ek a small imaginary part . I i, which is of a definite sign As a result, the random part of the overlap amplitude takes the form J, +~Jz, where both J& and Jz are real random variables. It follows that y,& is given by
y;~
g (J+iJzi )exptPi
I
(3)
where the sum runs over all forward-directed paths connecting the sites i and and P& is the phase acquired from the magnetic field along the path l. ' In (3), M is the number of oriented paths, exponential in the path length L in the VRH regime. ' In calculating the distribution of the overlaps, P(y), we follow the approach of Ref. 5 in which correlations among paths are neglected and the real parts of the path amplitudes, Jj&, are taken from a normal distribution of zero mean and variance ( i &. As for the imaginary part Jz& which represents the level broadening, we assume it to be taken from a normal distribution of finite mean, ( Jz &, and variance
'
(4)
Thus it is seen that in our picture the level broadening modifies the interference among the tunneling paths, and interferes with the effect of the magnetic field. With those considerations, it is straightforward to obthe joint probability tain P (y). One first derives P ( Jz ), where from Eq. (3)
+J J
Jz, sing,
The interference factor y,& in the expression for the e1ementary conductance g/ [Eq. (1)] describes the renormalization of the tunneling probability due to all paths conHere we adopt the picture ' acnecting the sites i and cording to which the important contribution to y,& of pairs belonging to the critical network, deep in the strong localization regime, comes from all forward-oriented paths within a cylinder-shaped region of length r,& and This imp. lies that the paths considered width (gr/)'~ for the hopping from the initia1 to the final sites are roughly of the same length, longer than the localization 1ength. The microscopic model for the individual path amplitude consists of an Anderson model, with disordered on-site energies taken from a symmetric distribution of width 8'and hopping matrix element V. Then the
1 J, = &M g (Jcosgi
),
.
f.
v'M y(Jiinyi+Jzicosyi)
The required distribution P (y ) is then
P(y)=
of the individual paths, Pi, appear as parameters. When the integration in Eq. (6) is carried out, these phases appear in the forms
in which the magnetic phases
1
g, g expi, Pi +igz,
and
49
1
...
5167
exp~20I
pl+ ~p2
'
(8)
when
and cr+ =o' = ( J1 ), the distribution (11) reduces to that obtained before (cf. Refs. 5, 11, and
a1=a2 =0
12}.
In addition, the parameters of the J& and the J2 distributions appear in P (y) in the forms
(9)
Finally, at high fields such that the magnetic flux through the hopping region is on the order of the quantum flux unit A, a &, and a2 approach zero. In that limit, the distribution (11) takes the form
1 F(s)= 1 +so'+
r=M(J )'
(10)
the effective amount of level which characterizes broadening in the hopping process. The distribution of the overlaps, P (y ), is most conveniently presented in terms of its Laplace transform
""
f
0
which is formally the same distribution as in the case where level broadening is not considered, except for the appearance of 0+. It should be noted that higher values of the magnetic field are beyond the scope of our model, as explained above.
""
F(s)=
where
dy exp(
1p(s)]/F7(s), sy)P(y)=exp[
sa2b2
(11)
sa, b, 1+sb1
1+sb2
) ]'i~
(12)
(lny ) =
dy
P (y)lny .
(18)
A(T
(13)
In this section we calculate the logarithmic average by P(y} in Eq. (18} the inverse Laplace transform of F (s), Eq. (11). This leads to the expression
using for
=p+p +1
(14)
(lny ) =
1
2m'l
The remaining two parameters in the distribution (11), a, and az, are proportional to the amount of level broadening 1 [Eq. (10)] and are given by
z g1+
1
s
s
F(s)ln(
ys),
(19)
z lg
~
1
[21M2rI19$+P'1( }1
)2)]
(15)
2
I
2
b 2b2
91+ l2+
[2P29192+Pl(91 92)]
2
in which lny=0. 577 is the Euler constant and the contour of integration is depicted in Fig. 1. Equation (19) is derived from Eq. (18) as follows. The contour L which defines the inverse Laplace transform of F(s) is a straight a+a(x} in the half plane line between and a ~~ Res & 0 of the complex s plane, for which all the singularities of F(s) lie on its left side. Those singularities [see Eqs. (11) and (12)] are branch points at s = z, where s,
We first consider the overlap distribution in the absence of a magnetic field. In that case, A =1, p&=q&=1, and @2=F2=0. Consequently, the Laplace transform of the distribution takes the form
s, 2
&0
(20)
&i, 2
Therefore, 0&a &s&. Because of the exponential factor in F(s), the singular points correspond to essential singu-
Fo(s}= exp
sr
1+s
1
sr+
u
1/2
s plane
1+s(o.++a
o 1+s(o+
(16)
)
and behaves as 1/s in the limit This implies that the inverse Laplace transform P (y ) is independent of y as y approaches zero, i.e., the distribution belongs to the Gaussian unitary ensemble. ' We note that this feature is directly related to the imaginary component of the overlap amplitude, J2. where ( J2 ) =0 and ) =0, the Laplace transform would have approached s 'n ao, which in turn leads to P(y)-y i.e., to a distribution of the Gaussian orthogonal ensemble. Second, we note that in the absence of the level broadening, i.e.,
s~ ~.
-Sp
s~
',
(z
in the s plane.
The contour C can now be transformed into the contour C~ + C (see Fig. 2), where C~ is a circle of a large radius R disconnected at the cut, and C is along the cut. On Cz one may replace g(g) by unity and R (g) by g. The integral can then be easily calculated, yielding ln[cR /) ]. The integral along C can be reduced to a simple one by noticing that for g=g, +tgz, with g, &0 and
(z~+0,
such respectively.
C
= (cg+d) that s = s,
=b&
b2
(21)
s = sz
ln
+tgz+e) (g,
R
ln
(g, tgz+e)
Bi
x
=2mt .
(28)
become
v'x (x
(23)
+ 1)
exp
+1
v'E+ &e+ I
(29)
c=d/c
)0 .
1)
(30)
dx
~
x(x+1)
c
4a c
e.
=-+
we finally obtain
(24)
+e+1+v E
dx
which can be transformed into an arbitrary contour C, encompassing counterclockwise the cut (0, 1) (see Fig. 2). Carrying out the above transformations we obtain
&x (x
+1)
exp
8)
x+1
(31)
bound by expression section to upon the
(lny ) =
where
(25)
In the last integral we have replaced the upper This ~, as the integral converges for x for the logarithmic average is used in the next determine the e6'ects of the level broadening
~ ~.
(26)
and
MC.
IV. DISCUSSION OF THE MAGNETOCONDUCTANCE
B, =
a)b)
C
B2 =
a2b2
(27)
'
g & e.
[This follows
from
(32)
lane
In order to characterize the magnetic-6eld dependence of the MC, we assume' that for any path with a phase Pt there is also a symmetric path with a phase (a valid Pt in the model of Nguyen, assumption Spivak, and Shklovskii' in which the number of paths is large). Then pz=g2=0 and consequently a, =0 and
a2
= I q)/b~ .
(34)
average takes the
Br=0, B
'(v'ad+1+ +21n
,
v'e)
g plane.
rv'1+r
Bt exp
49
...
5169
In
of level
and
M =exp
(36)
a'.
f (N
(44a)
This reproduces the results of Refs. 11 and 12, and shows that under this condition the MC is positive. At weak fields we have [see Eqs. (7), (8), and (14)]
N'ln(1/N ), N
1
&&1 .
(44b)
3=1 2h,
h2=
1
=1 h
(37)
units
0 '
H,
~
Hence, the number of paths becomes exponentially large as the hopping distance R increases, that is, as the temperature tends to zero. It follows that the sign of the MC This point will be dismay change with temperature. cussed further below. Next we consider the efFect of the level broadening upon the zero-field logarithmic average. From Eq. (35}
we find
H, =$0/R&Rg=H, (T/To)
&&H,
(38)
It, follows [see Eq. (36)] that in the absence of level broadening, the MC at weak fields is linear in the Geld, i.e., MC ~ 2 h ~. To find the weak-field behavior of the MC in the presence of level broadening, we define
~
. (8Jz) tn
(J)&
2
Bt
MC=Ah,
A,
=
Bh
r&(t+1)
(lny )o .
(39}
&0.
(45)
~= 2+
+
2
exp
1/2
Xexp
(40}
This result is examined in two limits. (i) When the J2 distribution is almost symmetric
(41)
&0.
(ii)
That is, the level broadening enhances the interference contribution to the conductance at zero Geld. This result is quite surprising as one would have expected the broadening at the intermediate states to destroy phase memory. The enhancement of the zero-field conductance can be explained as follows. Since we consider a symmetric distribution for the real part of the tunneling amplitude (J& }, then in the absence of an imaginary part (J2 =0) there is a significant cancellation of amplitudes of opposite signs. This cancellation is destroyed in the presence of J2, i.e., when the transition amplitudes are complex. The effect is larger as the J2 distribution becomes more asymmetric. %e emphasize that the derivation of Eq. (45) assumes a &0, i.e., (J, ) & (5J2z). When the broadening is stronger and the inequality is reversed, one expects the phase memory to be lost. To deduce the overall behavior of the MC up to fields such that the Aux through the hopping region is larger than $0, i.e., H H, (but H Hhigher fields are beyond the scope of our model), we consider the
'
),
difference
(42)
&
lny
) ( lny
&, ,
(46)
one may neglect the second term in (40) and obtain Thus the level broadening changes the functionA, 1. al dependence of the MC at small fields from ~h~ to h . The crossover from ~h~ to h occurs at H =H that is, the linear dependence is smeared out by the level broadening. The sign of the MC at sma11 fields, i.e., the sign of A. [see Eqs. {39)and (40)], depends on the degree of asymmetry of the J2 distribution and the number of paths. The latter can be obtained following the expressions derived in Ref. 7. One defines a parameter N,
(x&. where the first term is the logarithmic average at According to the discussion in Sec. II [see Eq. (17)] this term is
0~
(lny
Iny+lno ) =
+,
(47)
and it is independent of the asymmetry of the J2 distribution. It should be noted that this result is valid as long as
N=(p
Ng')'
'
(43)
(48)
where N is the density of hopping centers and scattering length of each of them. One then has
p is the
Assuming the magnetic phases, p~, to be Gaussian variables, we find g, -exp[ h /2]. Then the exponential
5170
O. ENTIN-VfOHLMAN,
49
factor in Eq. (48) is Mil, . Requiring small yields [see Eqs. (44)]
this factor to be
H'&,
f(N)
(49)
At the same time, we also have H & H, . Combining these two requirements together, we obtain the temperature range at which we may consider high fields:
' 1/2
FIG. 3. A
~f(g)
p
(50)
As for the second term in (46), we calculate it in the two limiting cases (41} and (42). We find that for an almost symmetric level-broadening distribution
2
1/2
1+ 1
(lny )
)0= ln (1ny
30
1
0+
'
3/2
0'+ +0
(51)
as the temperature is reduced. As a result, the J2 distriHence, bution is less symmetric at lower temperatures. the orbital MC of strongly localized electrons can be negative at low temperatures, and changes into positive at We emphasize that we have not higher temperatures. considered the effect of electron correlations, which should be important at very low temperatures. It is instructive to compare our results for the effect of level broadening with those obtained by considering a simple ring model (see Fig. 3), used in Ref. 7. In this simplified picture, there are only two paths. Therefore, the total transition probability becomes
Under condition (41) this difference is positive and hence in this case the MC is positive for all fields, though smaller than in the case where the broadening is neglected. The situation is changed when the level-broadening distribution is substantially asymmetric [Eq. (42}]. We show in the Appendix that in this case (lny ) 0=lnI' and consequently
y(4)=~(Ji++iJz+
)e"
+(J +iJz
)e
(53)
where is the Aux through the ring, measured in units of are the path amplitudes (see Fig. 3). To find $0, and the MC we need to calculate
(hlny) =
f dJ,+dJ, dJ2+dJ2
XP(J,+,J, , Ji+, J2
)ln
MC- 1+
Since for condition (42) [cf. Eq. (10)] implies that I & a+, it follows that a very asymmetric distribution of the level broadening renders the MC to be negatiue at all fields. As it is pointed out above, the amount of asymmetry of the J2 distribution [Eqs. (41) and (42)] depends upon the temperature. This dependence has two origins: the relationships between the number of paths and the hopping distance [Eqs. (44)] and the temperature dependence of the level-broadening parameters. The level width I k of a given intermediate state of energy ek is determined by the fastest transitions, i.e. , those occurring between nearest neighbors with a spontaneous phonon emission. This implies that I k depends weakly upon the temperature. Consequently, one expects J, and J2 to decrease as the temperature decreases, but not exponentially. On the other hand, the number of paths increases exponentially
I
a'+-e,
yI
(52)
y 0
(54)
Let us first confine ourselves for simplicity to the limiting case in which the path amplitudes are purely imaginary, i.e. , J*, =0. In this case, (b, lny ) is exactly the same as in Ref. 7, when one replaces the real amplitudes by the imaginary ones. Moreover, it is shown in Ref. 7 that when the distribution is symmetric, the MC is positive, while when it is asymmetric, MC&0. This finding is in accordance with our conclusions regarding the relationship between the sign of the MC and the asymmetry of the J2 distribution. The ring model with real transition amplitudes gave that the sign of the MC is related to its periodicity. For a period of $0/2, the MC is positive, while for $0 periodicity it is negative. Let us, therefore, examine the MC of a ring with complex transition amplitudes, in the case is symmetric, and the distriwhere the distribution of bution of Jz is arbitrary. One then finds from Eq. (54)
&
dJi+dJi
dJ2+dJ2
P[(Ji+ ), (Ji
),J2+, J2
]ln
1+C, sin
+C2sin~@+ 2
(55)
C, and C2 are real combinations of J~ and J2. In particular, C, vanishes when J2 =0. It follows from this equation that for nonzero J2, the periodicity of the MC is always $0, while its sign depends upon the details of the J2 distribution. The ring-model results are, therefore, in accordance with our conclusions. Moreover, this example shows that even in the most simplified situation, addwhere
ing an imaginary part to the transition amplitude essentially the resulting magnetoconductance.
changes
ACKNOWLEDGMENTS
We are grateful to A. Aharony and A. B. Harris for helpful discussions. The research was partially supported
49
...
5171
by the fund for basic research administered by the Israel Academy of Sciences and Humanities, and by the United (BSF). science foundation binational States-Israel A. G.A. is grateful to the Landau-Weizmann program for financia1 support.
(A2}
APPENDIX: THK LOGARITHMIC AVERAGE FOR A STRONGLY SYMMETRIC DISTRIBUTION Here we derive (Iny ) at zero field for a strongly asymmetric Jz distribution, i.e., under the condition (42}. To this end we write the integral in Eq. (35) in the form
In the second term, under condition (42) [see Eq. (34)] we '~ . may expand (1+t) Up to corrections of order B second integral in (Al) gives the
(A3)
v~0
lim
. 'dr
v
I+t
. 'dr
v
' Combining (A2) where Et is the exponential integral. becomes and (A3), the integral in Eq. (35), for B/s
1,
1+t exp
Bt
(Al)
'(v'a+1+&a)+InyB . 2ln
,
(A4}
'Permanent address: I. F. Ioffe Physical-Technical Institute, 194021 Saint Petersburg, Russian Federation, and Dept. of Condensed Matter, ICTP, Trieste, Italy. 'V. I. Nguyen, B. Z. Spivak, and B. I. Shklovskii, Pis'ma Zh. Eksp. Teor. Fiz. 41, 35 (1985) [JETP Lett. 41, 42 (1985)]; Zh. Eksp. Tear. Fiz. 89, 1770 (1985) [Sov. Phys. JETP 62, 1021
and B. Z. Spivak, l. Stat. Phys. 38, 267 (1988). A. Miller and E. Abrahams, Phys. Rev. 120, 745 (1960). 4V. Ambegaokar, B. I. Halperin, and J. S. Langer, Phys. Rev. B 4, 2612 (1971). 5U. Sivan, O. Entin-Wohlman, and Y. Imry, Phys. Rev. Lett. 60, 1566 (1988); O. Entin-Wohlman, Y. Imry, and U. Sivan, Phys. Rev. B 40, 8342 (1989). W. Schirmacher, Phys. Rev. B 41, 2461 (1990).
Shklovskii and B. Z. Spivak, in Hopping Transport in Solids, edited by M. Pollak and B. I. Shklovskii (NorthHolland, Amsterdam, 1991). 8J. L. Pichard, M. Sanquer, K. Slevin, and P. Debray, Phys.
B. I.
(1985)].
B. I. Shklovskii
Rev. Lett. 65, 1812 (1990). and M. Kardar, Phys. Rev. Lett. 66, 3187 (1991); Phys. Rev. B 46, 9984 (1992). H. L. Zhao, B. Z. Spivak, M. P. Gelfand, and S. Feng, Phys. Rev. B 44, 10760 (1991). ' Y. Meir, N. S. Wingreen, O. Entin-Wohlman, and B. L. Altshuler, Phys. Rev. Lett. 66, 1517 (1991). ' Y. Meir and O. Entin-Wohlman, Phys. Rev. Lett. 70, 1988 (1993). t3M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions (Dover, New York, 1964)
E. Medina