Professional Documents
Culture Documents
S. Rabello and Q. Si
Department of Physics, Rice University, Houston, Texas 77251-1892, USA
(Received 28 November 2003; revised manuscript received 25 March 2004; published 18 June 2004)
We consider momentum resolved tunneling between a Luttinger liquid and a two-dimensional electron gas
as a function of transverse magnetic field. We include the effects of an anomalous exponent and Zeeman
splitting on both the Luttinger liquid and the two-dimensional electron gas. We show that there are six
dispersing features that should be observed in magneto-tunneling, in contrast with the four features that would
be seen in a noninteracting one-dimensional electron gas. The strength of these features varies with the
anomalous exponent, being most pronounced when ␥ = 0. We argue that this measurement provides an impor-
tant experimental signature of spin-charge separation.
II. OVERVIEW
245109-2
MOMENTUM-RESOLVED TUNNELING BETWEEN A… PHYSICAL REVIEW B 69, 245109 (2004)
spin electrons and the three abrupt features each split in two
giving a total of six features in the conductance. This is
shown in Fig 5. The dispersion of these features as the mag-
netic field and applied voltage is changed also allows us to
determine the relative velocity of the spinon and holon is
shown in Fig. 6. In the absence of spin-charge separation FIG. 4. (Color online) Differential tunneling conductance in the
there would be only one dispersing singularity in the 1D absence of Zeeman splitting shown as a function of dimensionless
spectral function leading to four features in the tunneling magnetic field r = qBvF / eV for different values of the anomalous
conductance as a function of field. This then is the essence of exponent ␥ and for dimensionless holon velocity a = vF / v = 2 and
our results and in the rest of the paper we give a more precise spinon velocity, a = vF / v = 3. The graphs have been shifted and
derivation of them. rescaled for clarity. The arrows mark divergences. Notice how in-
creasing the anomalous exponent from the noninteracting value of
zero reduces the three features in the differential conductance.
III. FORMALISM
liquid and Fermi liquid may respectively be adiabatically
We consider single-electron tunneling between a one- continued.
dimensional interacting electron metal, parallel to the x axis, For the Fermi liquid we assume Ĥ2D is the Hamiltonian of
and a two-dimensional electron gas in the xy plane separated a noninteracting gas of electrons with a free particle disper-
from the 1D wire by a distance d along the z direction. A sion. This captures the electron quasiparticlelike character of
potential difference is applied between the wire and the two- the low lying excitations in a Fermi liquid up to an overall
dimensional electron gas and a magnetic field is applied in quasiparticle renormalization factor z 艋 1, which is simply a
the plane of the two-dimensional electron gas but perpen- multiplicative factor in the tunneling current. It neglects the
dicular to the wire (along the y direction). The geometry is incoherent (nonuniversal) part of the spectral function and
shown schematically in Fig. 1. The appropriate formalism the 2 quasiparticle lifetime effects (we will only treat zero
was first derived in the context of superconductivity.13 The temperature). Similarly, Ĥ1D, is assumed to be a two-branch
Hamiltonian for the system may be written as spin-full Luttinger model with inter- and intra-branch
scattering.3 This model can be characterized by four param-
Ĥ = Ĥ1D共B兲 + Ĥ2D共B兲 + ĤT . 共1兲 eters (spinon and holon velocities, and two anomalous expo-
nents) which are the universal parameters of a general Lut-
The precise forms of the one-dimensional wire Ĥ1D and the
tinger liquid.1
two-dimensional electron gas Ĥ2D are not required for deter- Equation (1) differs from the Hamiltonian considered in
mining the universal tunneling properties. Instead we will Ref. 11 since we have explicitly allowed a coupling of the
identify the universal aspects of the electron spectral func- magnetic field to the 1D and 2D electron systems. Since the
tion in these two systems by assuming that the one- field lies in the plane of the 2D electron gas however the
dimensional system forms a Luttinger liquid and the two- coupling will be via the Zeeman interaction. The orbital part
dimensional system is a Fermi liquid. Thus for Ĥ1D and Ĥ2D of the interaction with the magnetic field is included in the
we will take the canonical forms from which the Luttinger tunneling term
245109-3
S. A. GRIGERA, A. J. SCHOFIELD, S. RABELLO, AND Q. SI PHYSICAL REVIEW B 69, 245109 (2004)
ĤT = 冕 drជdxt共x,rជ兲关e−iedB共x+rx兲/2⌽̂2D,
†
ជ兲⌿̂1D,共x兲 + H.c.兴.
共r
共2兲
Here we have included the transverse magnetic field using
the following gauge A ជ = 共0 , 0 , −Bx兲 and the usual Peierls cou-
pling. The crucial assumption behind this method is that the
tunneling amplitude t共x , rជ兲 for an electron to move from po-
sition x in the 1D to position rជ in the 2D gas is translationally
invariant—i.e., that the tunnel barrier is smooth—and so may
be written as t共x − x2D , y 2D兲. We also assume that this single
electron tunneling is weak (so as not to significantly disturb
the excitation spectrum in the 1D and 2D systems) and is the
only term coupling the systems together.
In momentum space then this coupling may be written as
I共B,V兲 =
e
冕 d 兺ជ t02关f共兲 − f共 − eV兲兴A1D共kx
k
245109-4
MOMENTUM-RESOLVED TUNNELING BETWEEN A… PHYSICAL REVIEW B 69, 245109 (2004)
l␣
dx
U␣
L␣
ds
We can describe what is known about these singularities
as follows. At very small q, the function looks very similar to a1/2−2␥a1/2−␥
a spinless fermion’s function. As q is increased two peaks ⫻ ,
兩sa − x兩1/2−2␥兩sa − x兩1/2−␥兩s − x + 共r − 1兲兩1/2
become apparent, a reflection of spin-charge separation, one
at vq and another at vq, where v and v are the velocities 共9兲
of charge and spin density waves, respectively. The exponent
where we have introduced the dimensionless parameters r
of the singularity at vq is 2␥ − 21 and the corresponding one = qBvF / eV, a = vF / v, and a = vF / v and the dimensionless
at vq is ␥ − 21 . The function terminates at negative q at −vq one-dimensional variables x = qvF / eV for the 1D wire wave-
with a nonsingular exponent ␥. The parameter ␥, which vector and s = / eV for the frequency. I0 = e兩t兩2m / is the
characterizes the interactions between left and right movers natural unit for current in this problem. From this integral we
is always positive. The case of noninteracting left and right identify four different regions with different qualitative be-
branches, which was studied in Ref. 11, corresponds to the havior, R j , j = 1 , . . . , 4, corresponding to different situations
case ␥ = 0. As ␥ increases, the power law divergences of overlap between the one- and two-dimensional spectral
gradually weaken into cusp singularities, and the spectral functions. Figure 3 is a schematic representation of the rela-
weight, which for ␥ = 0 is confined within vq and vq is tive positioning between A1D and A2D in the four regions. In
gradually transferred by the electronic correlations toward terms of the dimensionless parameters these are given by R1:
higher values of q and . The spectral function should also r ⬍ 1; R2: 1 艋 r 艋 a; R3: a 艋 r ⬍ a; and R4: r ⬎ a. In each
be invariant to the transformations 共p → −p ; q → −q兲 where region, and for practical reasons only, the calculation is in
p = ± labels left and right movers, and 共 → − ; q → −q兲. turn split into different integrals of the same integrand. Table
Since the spectral functions away from these singularities I lists the upper and lower limits corresponding to each dif-
are likely to be nonuniversal, and in any case lead only to the ferent region. Although the majority of these integrals cannot
background tunneling current, we can capture the universal be integrated analytically, the asymptotic behavior in the dif-
physics by a function which has identical asymptotes to the ferent regions can be obtained by standard calculus proce-
same singularities. For ⬎ 0, the case of our interest, a func- dures. The q ⬍ 0 nonsingular part of the spectral function for
tion that has the correct singularities and asymptotic behav- ␥ ⫽ 0 [Eq. (8)] only contributes a small featureless onset of
ior can be written as conductivity at r = −a, and a background of finite conductiv-
ity noticeable only for small values of r and big values of ␥.
W共q, 兲⌰共 − qv兲 Figure 4 shows the differential conductance G = dI / dV as
A1D共q, 兲 =
兩 − vq兩1/2−2␥兩 − vq兩1/2−␥ a function of the dimensionless parameter r at T = 0. In the
following we discuss the behavior of G in each regime.
where R1.—For r ⬍ 1 and ␥ ⫽ 0 A2D overlaps with the nondiver-
再
gent part of A1D, leading to a finite but nonsingular flow of
⌰共qv − 兲 + ⌰共 − qv兲c共␥兲 if ␥ ⫽ 0 current, with negative differential conductance. When
W共q, 兲 = r reaches the value of 1, the conductance diverges as g
⌰共qv − 兲 if ␥ = 0, ⬃ −共1 − r兲−1/2+3␥, where g = G冑EFe 1 Ⲑ 2 +3␥V− 1 Ⲑ 2 +3␥ / I0 is a
共7兲 dimensionless measure of the conductance. For ␥ = 0 the
two areas do not overlap, implying that the current vanishes.
for q ⬎ 0 plus a nondivergent term for q ⬍ 0 and ␥ ⫽ 0 which R2.—For r ⬎ 1, the spectral functions for ␥ = 0 start
is proportional to overlapping as well, and the conductance diverges as
245109-5
S. A. GRIGERA, A. J. SCHOFIELD, S. RABELLO, AND Q. SI PHYSICAL REVIEW B 69, 245109 (2004)
TABLE I. Integration limits for Eq. (9) in the four different regions of spectral function overlap shown in
Fig. 3.
Region ␣ l␣ u␣ L␣ U␣
R1 1 0 r x − 共r − 1兲 1
R2 1 0 Ⲑ
r − 1 a − 1 sa sa
2 Ⲑ
r − 1 a − 1 Ⲑ
r − 1 a − 1 sa s + 共r − 1兲
3 0 Ⲑ
r − 1 a − 1 0 sa
4 Ⲑ
r − 1 a − 1 1 0 s + 共r − 1兲
R3 1 0 Ⲑ
r − 1 a − 1 sa sa
2 Ⲑ
r − 1 a − 1 1 sa s + 共r − 1兲
3 0 1 0 sa
R4 1 0 1 sa sa
2 0 1 0 sa
冋
1/2−2␥ 1/2−␥ 1/2−␥ 1/2−2␥
a a a a a a 1
g共a+ 兲 − g共a− 兲 = lim 共r − a兲␥ , g共r兲→r→a − lim 2F 1 1/2
a − 1 共a − a兲1/2−2␥ r→a a − 1 共a − a兲 1/2−3␥
r→a
␥ = 0, the conductance exhibits a discontinuity ⌬, the mag- which is a nondivergent peak (see Fig. 4).
nitude of which is R4.—The boundary is symmetric in r around a. The
asymptotic behavior is g ⬃ r−1/2 for all values of ␥.
⌬=
a
a − 1
冑 a a
a − a
. 共11兲 V. ADDING A ZEEMAN SPLITTING
In this section we consider the effects on the spectral
functions of a Zeeman coupling to the magnetic field. Very
For every nonzero value of ␥ this step is rounded off (see recent work by two of the authors of this paper12 has derived
Fig. 4) into a continuous function with a pronounced change the spectral functions of a Luttinger liquid with a Zeeman
at a, the change decreasing progressively as ␥ is increased. term in the Hamiltonian.
R3.—In this region the differential conductance becomes
positive. As r approaches the singularity line corresponding W共q, 兲⌰共 − qv − B兲
A1D共q, ,兲 =
to a, the boundary with R4, the conductance shows a pro- 兩 − vq − B兩1/2−2␥兩 − vq − Bv/v兩1/2−␥
nounced increase. Again the case of ␥ = 0 is unusual, since
共14兲
the boundary between R3 and R4 shows a singularity, which
is of logarithmic type where
where takes the value of the spin 共±1 / 2兲 times the Zeeman tivity. (This background is very small for ␥ ⬃ 0.) Further-
coupling factor. We have considered only positive q and , more, we have seen that the region of interest, where the
since as we discussed in the previous section, the contribu- features in the spectral functions are easily distinguished in
tion of q ⬍ 0 is merely to add a finite background of conduc- the differential conductivity, is restricted to small values of
245109-6
MOMENTUM-RESOLVED TUNNELING BETWEEN A… PHYSICAL REVIEW B 69, 245109 (2004)
the anomalous exponent (below ␥ ⬇ 0.2); for these values of Zeeman splitting and spin-charge separation are distinguish-
␥ the spectral weight outside the qv , qv region is so small able in this experiment—they are.
that all nondivergent contributions from A1D outside this in- The effect of an anomalous exponent is more subtle. The
terval are negligible. original proposal of Ref. 11 took the case of ␥ = 0 for cal-
The effect of the magnetic field in the two-dimensional culational simplicity. The more general treatment given here
system is taken into account by writing the spectral function shows that this is, in fact, the most singular case and other
as values for the anomalous exponent leads to less pronounced
effects. Nevertheless, if ␥ is not too far from zero, there will
冑m ⌰共 − qvF − ⬘B兲 still be six clearly distinguishable features. In the Luttinger
A2D共q, ,⬘兲 = 共16兲 model, ␥ comes from interbranch processes, while spin-
冑2 冑 − qvF − ⬘B ,
charge separation is due primarily to forward scattering in-
where ⬘ is the equivalent of for the two-dimensional sys- trabranch effects. Thus it is possible that spin-charge separa-
tem. tion and an anomalous exponent not far from the
Since we can split the current into the separate contribu- noninteracting value of one could coexist in real quantum
tions of the two possible values of the spin wires.
The role of the anomalous exponent in weakening the
I = I↓ + I↑ 共17兲 tunneling singularities has implications for other forms of
momentum-resolved tunneling experiments. Recently Car-
the problem is reduced to the calculation of the integral pentier et al.16 have analyzed the tunneling conductance be-
I↓,↑共V,B兲 =
4冑2I0共eV兲1/2+3␥
冑mvF
冕 冕dq d关f共 − eV兲
tween two Luttinger liquids in a magnetic field (though with-
out including the Zeeman effects as is done here). Again the
tunneling current can be viewed as a convolution but now of
− f共兲兴A1D共q, ,兲A2D共q, ,⬘兲. 共18兲 two Luttinger liquid spectral functions. An anomalous expo-
nent which differs from the noninteracting value will weaken
We can make use of the results of the previous section and the singularities in both functions in the convolution and will
simplify the calculation considerably by defining the spin be doubly detrimental to features in the conductance. Thus
dependent variable q⬘ = q − B / vs, and the spin dependent pa- we believe that using a two-dimensional Fermi liquid as the
rameter spectrometer, as described in this paper, optimizes the prob-
ability of seeing the dispersing features of the Luttinger liq-
⬘v − v F B
r↓,↑ = r ± . 共19兲 uid. This is because the Fermi liquid theory will always guar-
v eV antee a square-root singularity in its spectral function (after
The general expressions for the current and the differential integration over the transverse momentum) which is the best
conductance then reduce to one can do.
The experimental challenges in carrying out this experi-
1 ment should not be underestimated. We rely on a number of
IZ = 关I共r↑兲 + I共r↓兲兴, 共20兲
2 assumptions. The most obvious is that tunneling is occurring
uniformly along the 1D to 2D interface rather than via point-
1 like tunneling. The test for whether an experiment is in this
gZ = 关g共r↑兲 + g共r↓兲兴, 共21兲 regime comes from the magnetic field dependence. With
2 pointlike tunneling, one would expect only weak field depen-
where I共r兲 and g共r兲 are the functions for the current and the dence of the tunneling current since momentum would no
conductance derived in the previous section. Figure 5 shows longer be conserved along the wire. Experiments using mo-
the differential conductivity as a function of r for ␥ = 0, a lecular beam epitaxy grown interfaces have shown17 that the
= 2 , a = 3, = ⬘ = 1, and k2D 1D
f = k f . The different degree of
tunnel barriers can be sufficiently well controlled to preserve
field splitting of the different features in the conductivity can momentum conservation along the wire during tunneling,
clearly be seen. hence we believe that semiconductor fabricated quantum
wire to 2D metal interfaces will be the most promising can-
didate.
VI. CONCLUSIONS The second assumption is that the two-dimensional sys-
tem is a well-controlled Fermi liquid with a large electron
Having calculated the generalized form of the magneto- weight in the quasiparticle Z ⬃ 1. This ensures that the over-
tunneling conductance we see that the key signature of the lap between the electron and the excitations in the 2D spec-
types of excitation in a Luttinger liquid is revealed in the trometer are large. Again, estimates from semiconductor
appearance of six features which disperse with applied field. two-dimensional electron gases (2DEGs) suggest that this is
Loosely this may be viewed as the allowed transitions be- not implausible.
tween spin-split spinon and holon excitations and the spin- Our final assumption is that the rate limiting step in the
split electron in the two-dimensional electron gas. However, experiment is the tunneling process between the 1D and 2D
this differs dramatically from the case of electronlike excita- systems. This requires a clean quantum wire with no impu-
tions in the one-dimensional metal which would display four rities breaking the wire up into smaller pieces. Early results
features. So this directly addresses the issue as to whether suggest that this may be causing problems in trying to imple-
245109-7
S. A. GRIGERA, A. J. SCHOFIELD, S. RABELLO, AND Q. SI PHYSICAL REVIEW B 69, 245109 (2004)
ment this experiment in semiconductor devices.17 There the Fermi liquid properties, or indeed on the parameters of the
quantum wire is made by “pinching off” a channel in a two- Luttinger liquid which may be renormalized via screening
dimensional electron gas with an applied gate voltage. This from the 2DEG.
pushes the one-dimensional subbands through the chemical To summarize, we have considered momentum-
potential until only one remains active. Initial experimental conserving tunneling between a Luttinger liquid and a two-
results reveal magneto-tunneling occurring when multiple dimensional conventional metal. We have shown how a
subbands are conducting but the wire becomes insulating in transverse magnetic field can be used to tune the relative
the last subband. This is presumably due to impurities block- momentum of the tunneling electron. This then provides a
ing conduction (an interesting process in itself).18 Ultimately, direct measure of the spectral function of a Luttinger liquid
we believe that this should be viewed as a challenge rather via its convolution with that of a conventional Fermi liquid.
than a fundamental flaw in the experiment. However, it also The signatures of spin-charge separation are revealed as fea-
suggests that we should look at alternative realizations of this tures in the tunneling conductance and we have shown they
experiment. One possibility is to use carbon nanotubes as vary as a function of Zeeman splitting and anomalous expo-
quantum wires since these have already been used to dem- nent. The advantage of this experiment is that it can be per-
onstrate Luttinger liquidlike behavior19 via point tunneling. formed with high resolution compared to other probes of the
If a suitable interface could be found with a two-dimensional spectral function such as angle-resolved photoemission. Also
conventional metal this would be a good alternative candi- the experiment has a very straightforward theoretical inter-
date for the magneto-tunneling measurement. pretation and hence, if successful, is an unambiguous detec-
Finally we should point out that, although we have used a tor of spin-charge separation. We have also discussed the
magnetic field for tuning the relative momentum between the prospects of performing such an experiment.
wire and spectrometer, this is not the only method. Using a
semiconductor 2DEG one could back-gate the device and ACKNOWLEDGMENTS
control the carrier concentration, and hence kF, in the spec-
trometer. This gate voltage would then provide the momen- The authors have benefited from useful discussions with
tum tuning via the difference in kF between the wire and the T. S. Grigera, M. W. Long, and L. Macks and we thank A. P.
2DEG. Such a method could be used in cleaved edge over- Mackenzie for a critical reading of the manuscript. We are
growth devices where the tunnel barrier to the quantum wire grateful for the financial support of the Leverhulme Trust,
is in the plane of the 2DEG.20 Using kF to tune the momen- The Royal Society, and NATO Collaborative Research Grant
tum would, of course, mean there is no need to consider No. 971072. Two of us (S.R. and Q.S.) have been supported
Zeeman coupling. However, the results may be complicated in part by the Robert A. Welch Foundation, NSF Grant No.
by any carrier concentration dependence of the 2DEG on its DMR-0090071, and TcSAM.
1 F. D. M. Haldane, J. Phys. C 14, 2585 (1981). Schofield, Phys. Rev. Lett. 83, 1203 (1999); cond-mat/9907459.
2 S. Tomonaga, Prog. Theor. Phys. 5, 544 (1950). 12 S. Rabello and Q. Si, cond-mat/0008065.
3 J. M. Luttinger, J. Math. Phys. 4, 1154 (1963). 13 J. R. Schrieffer, D. J. Scalapino, and J. W. Wilkins, Phys. Rev.
4 A. J. Schofield, Contemp. Phys. 40, 95 (1999). Lett. 10, 336 (1963).
5 J. Voit, Rep. Prog. Phys. 58, 977 (1995). 14 G. D. Mahan, Many-Particle Physics, 2nd ed. (Plenum Press,
6 A. M. Chang, L. N. Pfeiffer, and K. W. West, Phys. Rev. Lett. 77,
New York, 1990).
2538 (1996). 15 V. Meden and K. Schönhammer, Phys. Rev. B 46, 15753 (1992).
7 A. M. Chang, M. K. Wu, C. C. Chi, L. N. Pfeiffer, and K. W. 16 D. Carpentier, C. Peca, and L. Balents, cond-mat/0103193.
West, Phys. Rev. Lett. 86, 143 (2001). 17 B. Kardynal, C. H. W. Barnes, E. H. Linfield, D. A. Ritchie, J. T.
8 P. Segovia, D. Purdie, M. Hengsberger, and Y. Baer, Nature Nicholls, K. M. Brown, G. A. C. Jones, and M. Pepper, Phys.
(London) 402, 504 (1999). Rev. B 55, R1966 (1997).
9 C. Kim, A. Y. Matsuura, Z. X. Shen, N. Motoyama, H. Eisaki, S. 18 C. L. Kane and M. P. A. Fisher, Phys. Rev. B 46, 15233 (1992).
Uchida, T. Tohyama, and S. Maekawa, Phys. Rev. Lett. 77, 19 M. Bockrath, D. H. Cobden, J. Lu, A. G. Rinzler, R. E. Smalley,
245109-8