PHYSICAL REVIEW B VOLUME 23, NUMBER 9 1 MAY 1981

Quantized Hall resistance and the measurement of the fine-structure constant

R. E. Prange
Department of Physics and Astronomy and Center for Theoretical Physics,
University of Maryland, College Park, Maryland 20742
(Received 10 October 1980; revised manuscript received 17 February 1981)

An elementary, exact calculation of two-dimensional electrons in crossed electric and magnet-

ic fields with a 5-function impurity is carried out in the quantum limit. A state localized on the
impurity exists and carries no current. However, the remaining mobile electrons passing near
the impurity carry an extra dissipationless Hall current exactly compensating the loss of current
by the localized electron. The Hall resistance should thus be precisely h/e2, as found experi-
mentally by Klitzing et al. Other possible sources of deviation from this result are briefly exam-
ined.

In a recent Letter, v. Klitzing et al. ' have reported a case in which only the ground Landau level is occu-
high-accuracy measurement of ez/h (to one part in pied. We choose m as the unit of mass, mc/eB as the
10') which after improvements and together with the unit of time, and (hc/2rreB)' —
= is as the unit of
known value of the speed of light, c, potentially' length. (The cyclotron radius, krhc/2rreB, does not
would provide a measurement of the fine-structure enter the problem. ) The drift velocity, cE/B =
constant of precision greater than that currently avail- —e4„/lt'is denoted by v, where &bn is the Hall vol-
able (one part in 107). Their result is based on a
measurement of the quantized Hall resistance in a
two-dimensional electron gas, as realized in the in-
version layer of a metal-oxide-semiconductor field-
effect transistor. We here provide an elementary cal-
culation which has a bearing on their result, and is a
step in the direction of estimating theoretically the
accuracy to which ez/h is determined by the experi- M~
ment. Mp ap
Since free electrons which fill an integral number
of Landau levels give a Hall resistance precisely an M~
integral fraction of h/ez, the problem is one of treat-
Mp
ing the imperfections, which might give rise to ordi- M)
nary resistance, and/or to localized states which could Mo al
cause the Hall resistance to deviate from its ideal l"I
value. (We do not treat the electrons as interacting, 0 M,
an omission which future work must remedy. ) We
Mp
here work out the instructive, elementary, and essen-
tially exactly solvable case of two-dimensional elec- N(E) ap
trons with a single 8-function impurity.
The main result is that (a) a localized state exists,
which (b) carries no current, but (c) the remaining FIG. 1. The inset gives the geometry of the single-impurity
nonlocalized states carry an extra Hall current which ex- problem. To the right below is the energy diagram of the
actly compensates for that nor carried by the localized lowest two Landau levels, as banded by impurities, and bent
state. Thus, provided all the nonlocalized states of the by the electric potential. The lines a, —a; bound regions of
appropriate Landau level are filled, the total Hall energy states deriving from distinct levels. The lines
current carried by the level is precisely the same as in M, —M; are mobility edges separating localized and delocal-
ized states. The hatched regions are those of delocalized
the absence of impurities and localized states.
states. The electrochemical potential is given by the line
We consider then free electrons of mass m
p, —
" "
p, — p, — p, . The position of this line depends on the
(=0.2m, as appropriate for silicon) in the xy plane, total density of electrons, i.e. , on the gate voltage. The
subjected to an electric field E in the negative x direc- " "
states are occupied below the line p, —p, —p, —p, '. The
tion, and a magnetic field B in the z direction. (The "
excess of electrons (holes) in the regions p, —p, ' —p, is the
geometry is given in Fig. 1.) Spin and valley degen- source of the Hall voltage. At the left is a schematic dia-
eracies are ignored, and attention is confined. to the gram of the density of states along the cut c —c.

23 4802 1981 The American Physical Society

23 RAPID COMMUNICATIONS 4803

tage, and N is the width of the device in the x direc- none of the states carries current along the electric
tion. The Hamiltonian is H =H 0+2+!tg'(r), with field. In fact, for the exact solution to be found, the
Ho= —, [ —0'/Bx'+ (pr +x)2] —ux. The eigenstates of same is true. The current operator in the y direction
80 are is jz=p&+x =p+x+v. When this operator acts on
a state P it gives vP plus a state orthogonal to P,
Q„r(r) =exp(ipry ) H„(x) Q(x)/(2 "n!L) 'r' because the operator p +x changes the Landau level,
and by assumption, there is only one Landau level in
(The Landau gauge has been used. ) The length of
the sum defining P . Thus, all states belonging to
the sample in the y direction is L, H„ is the Hermite
the Landau level carry the same Hall current.
polynomial, x =x +p, p = pr —u, and $(x)
Since a localized state can certainly carry no cur-
=e -z" r'/n'r4.
2
The eigenenergy corresponding to rent, we conclude that whenever the approximation is
state P„r is n + vp. The values of p are 2rrk/L with k valid, there are no localized states. We shall see that
integer and — p ranges from — b to H' —b, where the for the case of the 5-function potential, there is a lo-
impurity is at the origin of coordinates which is locat- calized state and the approximation is not valid.
ed a distance b above the lower edge of the sample. We therefore must solve the complete problem,
The total number of p values is thus WL/2rr. This keeping the admixture into the wave function of all
gives the degeneracy of the Landau level. For a sam- Landau levels. An immediate difficulty arises be-
ple as described in Ref. 1, we have in our units (with cause the sum on n in (2) does not converge if per-
8 =18 T), L =7 x10, W=g x10', and formed after the sum over p, at least if it is assumed
v —1/ W — 10 4. We shall treat L and Was macro- that the latter sum may be approximated by an in-
scopic and u as small, but it will be seen that 1/L is tegral from — ~ to ~, using the largeness of L and
by no means the smallest number in the problem. 8', and that, for large n, the term pe in the denomi-
The potential strength A. is of order unity, i.e. ,
X »1/L.
To find the full eigenstates of 0, the state, denoted
nators may be neglected. However, the finiteness of
the integration range will become a factor when the
spatial extent of the wave functions P„~ starts to be-
by P, is expanded in the "unperturbed" eigenstates come equal to the sample width, and there will be an
of Ho as g (r) = Xc„rg„r(r) It is eas.ily seen that effective cutoff at n =M —W'. (The large magni-
tude of this cutoff is special to the 5-function poten-
c4 =A q„, (0)/(E —n —pu) tial. Finite-range potentials will have much smaller
with the eigenenergy E determined by effective cutoffs. )
To evaluate (2) and (3), the sum over p is replaced
(2) by a principal value integral plus a contribution corn-
„~ E —n —pv ing from the p values in the immediate neighborhood
of the singular point. By introducing k —5
and with the amplitude determined by = LE /2rru, where k is an integer and 2~8, & I, as~

well as p, =2rrk /L Eq. (2) may be rewritten as

(3)
1 =kG(E. ) +) y(p )'(r, /u . (2a)
As is familiar, except for possible bound states break- Here G (E) is the principal value integral, equal to
ing off above or below the bands of levels, the ener- — ' —
XD (E n) for ~E n && v. (Since G is of order
~

gies determined by (2) fall between the closely lnM, the large but finite cutoff does not lead to an
spaced levels of the successive p values. Thus, we intolerably large G. ) The discrete sum is convergent
may use as a label for the state o, the nearest level of and is given by a = — scot(n5 ). In the same way
the system unperturbed by the impurity. For simpli-
the amplitudes are expressed as
city only levels belonging to the zeroth Landau level
are considered. BG L( + )
We begin by making the approximation of retaining (~ )
8Ea
( ') (2n v)'
(3a)
only the term n =0 in the sums. This is suggested
by the "strong magnetic field limit" which is usually Consider first states for which p )& 1, that is,
taken in the literature. ' The idea is that it is ade- states which hardly overlap the impurity and for
quate to diagonalize the subspace consisting of the
(nearly) degenerate states of one Landau level, if the
which Q(p )' « 1/L is very small indeed. Except
for the case that G(E, ) = I/h. , o. will necessarily be
levels are sufficiently separated. The current in this enormous, the second term on the right of (3a) will
case may be obtained without the necessity of finding dominate, and the state Q will differ insignificantly
the explicit eigenfunctions and is completely indepen- from the corresponding unperturbed state.
dent of the form of the scattering potential. The For the special energy En satisfying G(En) = I/k,
(number) current operator in the x direction is however, cr — 1, and the first term on the right-
j„= (I/i)(8/Bx), and it is immediately seen that hand side of (3a) dominates. Under the cir-
 4804 RAPID COMMUNICATIONS 23

cumstances, there is exactly one level for which this carry a current.
is true. This also subsumes the case for impurities The case of two 5-function potentials may be stud-
near the sample edges, which may have bound states ied as well. In this case, if the potential sites are
lying outside the quasicontinuum.
take the thermodynamic limit, L
(One may also
~, in which case
separated by a distance l » 1, they do not interact,
and the single-impurity problem gives the answer.
the state in question becomes an exceedingly narrow (This may be extended to many well separated im-
resonance, without changing the result. ) This is a lo- purities. ) On the other hand, if the separation l is
calized state, whose wave function is P»=exp[ —(x' small compared with unity, the potential acts as a sin-
+y~ —2ixy)/4]/J2n (for u =0, and neglecting the gle site. Only when l — 1 do the impurity levels in-
interlevel mixing). It is a peculiarity of this system teract and start to form an impurity band,
that the spatial extent of the localized state is con-
trolled by the magnetic field when the potential fluc- son localization due to Mott and others, ''
The preceding results, as well as the lore of Ander-
support
tuation is short ranged and does not become large the conjecture" that the following picture holds. In
even when X and E» become small. (This may modi- the absence of an electric field, and in the presence
fy the theory of Anderson localization for weakly lo- of a considerable number of impurities, defects or
calized states. ) The current of the localized state, j», other potential fluctuations, and in a sufficiently
of course vanishes as can be verified by direct com- strong magnetic field, a given Landau level will be
putation which gives broadened into a band. The central region of this
t band will be delocalized states. Beyond a mobility
j =u 1 —(A )' g(n +I)[ (E» —n) -2 edge there will be localized states, which can only
2' conduct current by hopping processes. The delocal-
ized states of different bands will thus be separated
(E, -n--l)-'] =0 . by a region of localized states. The delocalized states
"
of each Landau "band, however, collectively carry a
total Hall current I, in the presence of an electric
(4) field or potential gradient, which is
Next, the remaining class of states is considered,
namely, those with energies E whose corresponding I = ( eu/L) L W—/2rr =e~&btt/2rr(=e'4tt/h) (8)
p is of order unity, that is, states made up of unper-
turbed eigenstates which have a significant overlap Our calculation thus tends to confirm the expectation
with the impurity. The eigenstates P do not overlap that the quantum Hall resistance is h/e~.
the impurity, of course, since they must be orthogo- This result applies to a dilute system of 5-function
nal to P~ which is localized at the site. The energies impurities. Ando et al. also obtained a quantized
E are of order u, and thus G(E, ) ReGp(p )/u —
= Hall current within the framework of their approxi-
mations, which are the assumption of "high magnetic
which is given by
"
field, the single-site approximation with the effects
G, (u. ) = Jdu —p + I'g of scattering taken into account self-consistently, and
p~ the assumption that there are gaps between the
— — — resulting "impurity bands.
" The spirit of the "high-
Then, G 1/u, $(p ) I, a I, and the second
term on the right-hand side of (3a) dominates be-
field" approximation seems to be the same as dis-
cause L is large. The Hall current carried by such a cussed earlier, but in detail it is different since the
state may be evaluated to leading order in v as elegant formalism of Kubo et al. 6 is used. (This
method breaks the current correlation expression for
iy =u+(A )'/rru . the conductivity into two parts, one of which is treat-
ed exactly, and the other approximately. ) Although

u)—
The sum over states n is smooth and may be re- the conditions for the validity of the high-field ap-
placed by an integral over p . It is found that proximation have not been spelled out, it seems like-
j'—
= $(iy (7) ly that it will be valid when the field is so great that
the potential hardly varies on the scale of l~. The 5-
function results on the other hand ought to be quali-
=v —
2
dp Im
1 tatively valid when the potential is confined to a
7T Gp(p) small region compared with l&. Given 0
the value of I&
appropriate for the experiment (70 A), it is unlikely
The integral is evaluated as m. dp p'@ p =m 2, by that either approximation is a priori very good. The
recognizing its analytic properties. Thus, j,
which is
the excess current carried by the electrons which pass
experiment, however, is evidence that the results
which have been obtained in these limiting cases
near the impurity, is just j
= v, exactly enough to must be valid under very general conditions.
compensate for the failure of the localized state to Aside from these questions which somehow must
23 RAPID COMMUNICATIONS 4805

receive a favorable answer if the experiment is to be current and lead to a potential drop along the length
explained at all, there are several other considerations of the sample. By controlling the frequency of the
which might lead to effects at the part in 10 level. radiation, something about the position of the mobili-
In particular, W is not so large that corrections of or- ty edges may be inferred.
der 1/Incan be tolerated. Thus, the edge effects The approximation of constant electric field must
must be carefully investigated. It is known from the also be examined. It is not known where the charge
theory of the Landau diamagnetism, for example, which gives rise to this field resides, at least in the
that the surface Landau levels tend to carry current case that the diagonal component of the conductivity,
in a direction counter to that of the bulk. Perhaps cr, vanishes, and the actual electric field configura-
this kind of correction can be avoided along the lines tion may depend on how the Hall current is set up.
suggested by the foregoing, namely, that any surface If the charge is localized toward the edges of the
states or other anomalous states will be compensated sample, it will attract an equal and opposite image
for by an increase in the Hall current of the delocal- charge in the facing metal a few hundred nm away.
ized states. Another small effect needing investiga- The potential of such a line of dipoles will vary most
tion is the nonparabolicity of the energy bands. pronouncedly in the first few hundred nm away from
It is interesting to ask whether the quantum Hall the edge. If this is the case, most of the Hall current
current is a supercurrent in the sense of the theory of will flow along the edges of the sample, and the inner
superconductivity, and whether a persistent current part will carry practically no current. This wi11 in-
can be set up. There is no dissipation connected with crease the effective value of the local v and correc-
the Hali current per se, since it is perpendicular to the tions of order v' could start to play a role. There is,
field. It is a supercurrent in the sense that the wave fortunately, no evidence thus far that the electric
function is locked into place by an energy gap, and it field is far from constant. The situation might arise,
is because of the vector potential that the current ex- however, in a Corbino (disk with center hole)
ists. geometry where a Hall voltage could be applied by
Thus the more interesting question is whether moving up external charges.
there will be a small current parallel to the electric In the actual experimental configuration, ' the pri-
field, that is, whether in this direction, the system is mary charge giving rise to -the Hall potential presum-
a perfect insulator. In our model, such a current can ably resides in the localized states. We thus envisage
come only by taking into account inelastic processes a situation which is schematically shown in the figure.
so far neglected, which give rise to a change in occu- The charges in these localized states will be unable to
pation of the states. When an entire level of relax toward equilibrium at low temperature since
current-carrying states are filled there is no possibility they require thermally activated inelastic processes to
of changing their occupation without large energy change their state. The Landau band will then bend
cost. The localized states also are activated so they to folio~ the potential, and a picture as in the figure
too are perfectly insulating at sufficiently low tem- will result.
perature. Thus we expect that the Hall potential can
be maintained without dissipation and that a per-
sistent current can be set up. ACKNOWLEDGMENTS
This raises the possibility of photoinducing a poten-
tial drop in the y direction. If light falls on the junc- I wish to thank Dennis Drew for introducing me to
tion, it can excite electrons into the unfilled delocal- this problem and to express gratitude to him and Vic-
ized states of a higher Landau level, and these elec- tor Korenman for fruitful discussions on the subject.
trons will provide a current in the x direction which This work was supported in part by NSF Grant No.
will in turn cause a shift in the direction of the Hall DMR 7908819.

K. v. Klitzing, G. Dorda, and M. Pepper, Phys. Rev. Lett. 4R. J. Nicholas, R. A. Stradling, and R. J. Tidley, Solid State
45, 494 (1980). Commun. 23, 341 (1977).
Several laboratories have already apparently achieved 1- 5H. Aoki and H. Kanimura, Solid State Commun. 21, 45
ppm precision by this method. E. Braun, E. Staben, and (1977).
K. v. Klitzing, PTB-Mitteilungen 90, 350 (1980); B. Taylor R. Kubo, S. J. Miyake, and N. Hashitsume, in Solid State
(private communication). Physics, edited by F. Seitz and D. Turnbull (Academic,
T. Ando, Y. Matsumoto, and Y. Uemura, J.-Phys. Soc, Jpn. New York, 1965), Vol. 17, p. 269.
39, 279 (1975), and preceding papers. ~K, v. Klitzing (private communication).