Professional Documents
Culture Documents
2001 - (Onofri) LandauLevelsOnATorus
2001 - (Onofri) LandauLevelsOnATorus
2, 2001
Landau levels have represented a very rich field of research, which has gained
widespread attention after their application to the quantum Hall effect. In a
particular gauge, the holomorphic gauge, they give a physical implementation
of Bargmann’s Hilbert space of entire functions. They have also been recognized
as a natural bridge between Feynman’s path integral and geometric quantization.
We discuss here some mathematical subtleties involved in the formulation of the
problem when one tries to study quantum mechanics on a finite strip of sides
L1, L2 with a uniform magnetic field and periodic boundary conditions. There is an
apparent paradox here: infinitesimal translations should be associated to canonical
operators [ᒍx , ᒍy] ⬀ iបB, and, at the same time, live in a Landau level of finite
dimension BL1L2/(hc/e), which is impossible from Wintner’s theorem. The paper
shows the way out of this conundrum.
1. INTRODUCTION
Landau levels were introduced in 1930 (see ref. 9). They found an
important physical application only quite recently, after the discovery of the
quantum Hall effect (see refs. 2, 4, and 5 and references therein). More
recently, it has been recognized that the theory of Landau levels provides a
general bridge between Feynman path integrals and “geometric quantization”
in all cases where the classical phase space is equipped with a complex
structure which makes it a Kaehler manifold [7]. From this general viewpoint,
or to get a more realistic description of conducting thin films, it is important
to understand the case of a finite region with suitable boundary conditions.
If these correspond to a compact (smooth) manifold without boundary, the
quantization condition of Kostant and Souriau (see ref. 11) and references
therein) or, equivalently, Dirac’s quantization condition for monopole charges
requires that the total magnetic flux be quantized, i.e., it must be an integral
1
Dipartimento di Fisica, Università di Parma, and INFN, Gruppo Collegato di Parma, 43100
Parma, Italy; e-mail: [email protected]
537
0020-7748/01/0200-0537$19.50/0 䉷 2001 Plenum Publishing Corporation
538 Onofri
multiple N of the universal constant hc/e. At the same time, the degeneracy
of the ground state is finite and coincides with N, except for a topological
correction (half the Euler characteristic of the manifold). A similar, approxi-
mate, result can be obtained by a semiclassical argument [9].
Consider now the simple case of a rectangular area with sides L1, L2
and periodic boundary conditions; the problem is formulated on a toroidal
surface with a transverse magnetic field whose flux is BL1L2. Of course, this
fact implies the presence of magnetic charges, hence Dirac’s quantization.
The problem is: What is the symmetry of the Hamiltonian? We expect that
the classical symmetry of the torus (S1 ⫻ S1) be realized as a projective
representation, the two infinitesimal generators satisfying Heisenberg algebra
with a central charge ⬀ បB (at least this is what happens in the noncompact
R2 case). But this is clearly incompatible (Wintner’s theorem) with finite
degeneracy of energy levels! While we cannot expect a spontaneous symmetry
breaking in a system with a finite number of degrees of freedom, we know
from geometric quantization that not all classical symmetries survive at the
quantum level, only those which are lifted at the prequantum level and,
second, respect the polarization (in Landau level language, those symmetries
are preserved that leave the first Landau level invariant). The problem is:
What exactly is happening on the torus?
To get an answer, we shall reconstruct the explicit form of the Landau
levels in terms of sections of the Hermitian line bundle associated to the
principal bundle with connection given by the magnetic potential A. The
language of fiber bundles is the natural one to describe gauge fields and it
is becoming more familiar to physicists especially after the advent of modern
string theory. We shall explicitly construct the transition functions of the line
bundle and find a natural orthonormal basis of holomorphic sections, which
turn out to be Jacobi -functions. By inspection, it turns out that translation
invariance is broken to a discrete subgroup ZN ⫻ ZN , N being the monopole
charge. This fact has the counterpart that the Hermitian operators which
correspond to infinitesimal translations (in the noncompact case) do not leave
the Landau levels invariant, i.e., they do not commute with the Hamiltonian:
while formally commuting with the Hamiltonian as differential operators,
they fail to respect the boundary conditions given by the bundle transition
functions.
2. ᐁ ⫽ {x ⬍ x ⬍ L1 ⫹ x, 0 ⬍ y ⬍ L2}
3. ᐁ␥ ⫽ {0 ⬍ x ⬍ L1, y ⬍ y ⬍ L2 ⫹ y}
4. ᐁ␦ ⫽ {x ⬍ x ⬍ L1 ⫹ x, y ⬍ y ⬍ L2 ⫹ y}
with some choice of constants x and y. A uniform magnetic field B transverse
to the surface ⺤2 is represented by the translation-invariant two-form B ⫽
B dx ∧ dy. The connection form A, representing the magnetic potential, is
defined in each local chart in such a way that dA ⫽ B. It is well known
that a global one-form on ⺤2 satisfying this condition does not exist, since
otherwise 兰⺤2 B ⫽ 兰⺤2 dA ⫽ 0, by Stokes’ theorem, while it holds that
兰⺤2 B ⫽ BL1L2. The problem is essentially the same as the presence of a
Dirac string in the case of a three-dimensional magnetic monopole. For the
sake of simplicity, we may define the local connection forms by the same
formula A ⫽ –21 B(x dy ⫺ y dx) since the coordinates x, y are indeed differenti-
able within each local chart.2 To characterize the connection form completely,
we have to identify the transition functions which relate Ai to Aj for any
pair (i, j ) in the set {␣, , ␥, ␦} and for each connected component of the
overlap ᐁi 艚 ᐁj; we have
A(x, y) ⫽ A␣(x, y) (x ⬍ x ⬍ L1)
A␥(x, y) ⫽ A␣(x, y) (y ⬍ y ⬍ L2)
A␦(x, y) ⫽ A␣(x, y) (x ⬍ x ⬍ L1, y ⬍ y ⬍ L2)
A(x ⫹ L1, y) ⫽ A␣(x, y) ⫹ d(–21 BL1y ⫹ ␣) (0 ⬍ x ⬍ x)
A␥(x, y ⫹ L2) ⫽ A␣(x, y) ⫹ d(⫺–21 BL2x ⫹ ␣␥) (0 ⬍ y ⬍ y)
A␦(x ⫹ L1, y ⫹ L2) ⫽ A␣(x, y) ⫹ d(–21 B(L1y ⫺ L2x) ⫹ ␣␦)
(0 ⬍ x ⬍ x, 0 ⬍ y ⬍ y)
and similar transition functions for the other cases. The constants ij are
arbitrary at this level; they will play, however, a crucial role in the lifting to
the associated line bundle which describes the quantum wave functions.3
1 1
A(z, z) ⫽ Bz dz ⫺ B d앚z앚2 (1)
2i 4i
which shows that by a gauge transformation, we can adopt a holomorphic form
1
Ah ⫽ B z dz
2i
for which we have the transition functions
Ah(z) ⫽ Ah␣(z) (x ⬍ x ⬍ L1)
Ah␥(z) ⫽ Ah␣(z) (y ⬍ y ⬍ L2)
Ah␦(z) ⫽ Ah␣(z) (x ⬍ x ⬍ L1, y ⬍ y ⬍ L2)
Ah(z ⫹ L1) ⫽ Ah␣(z) ⫹ d(⫺i –21 BL1z ⫹ ␣) (0 ⬍ x ⬍ x)
Ah␥(z ⫹ iL2) ⫽ Ah␣(z) ⫹ d(⫺ –21 BL2z ⫹ ␣␥) (0 ⬍ y ⬍ y)
Ah␦(z ⫹ L1 iL2) ⫽ Ah␣(z) ⫹ d(⫺i –21 B(L1 ⫺ iL2)z ⫹ ␣␦)
(0 ⬍ x ⬍ x, 0 ⬍ y ⬍ y)
with some new choice of constants ij.
4. QUANTIZATION
The Hamiltonian for a charged particle is given by the minimal-coupling
prescription. The local expression as a differential operator must be comple-
mented by suitable boundary conditions which ensure self-adjointness. This
is easily done in terms of a line bundle associated to A as defined in the
previous section. The physical principle to adopt is the gauge principle,
according to which
冢iប⭸ 冣
e
⫺ A
c
is covariant under gauge transformations, in particular under the transition
from one chart to another (here A are the components of the gauge potential
one-form A ⫽ 兺 A dx). This can be done directly in the holomorphic
gauge, which is our choice for the sequel. As usual, the complex line bundle
has transition functions obtained by exponentiating those which character-
ize Ah:
(z) ⫽ ␣(z) (x ⬍ x ⬍ L1)
␥(z) ⫽ ␣(z) (y ⬍ y ⬍ L2)
Landau Levels on a Torus 541
冢 冣
ប2 eB
Hh ⫽ ⫺4 ⭸⫺ z ⭸ (2)
2m 2mcប
具1앚2典 ⫽ 冮
⺤2
h(1, 2)[dz]
where [dz] ⬅ (1/2i)dz ∧ dz. It is easy to check that there is a smooth match
h(i , i) ⫽ h(j , j) on each ᐁi 艚 ᐁj provided that ᑬ(ij ) is suitably chosen.
542 Onofri
To simplify the notation, let us introduce natural units adapted to the problem:
let us use (ប/m)1/2 as length unit, where ⫽ eB/2mc is the Larmor frequency.
Then we get the new transition functions which make h(, ) smooth. At
this point, we can drop the chart index from the wave function: from our
convention, there is an open set common to all local charts where the wave
function is the same in all local charts and the transition functions merely
represent the boundary conditions to be imposed on .
(z ⫹ L1) ⫽ (z) exp{L1z ⫹ –21 L21 ⫹ i␦1}
(3)
(z ⫹ iL2) ⫽ (z) exp{⫺iL2z ⫹ –21 L22 ⫹ i␦2}
It is also easily checked that these b.c. make the Hamiltonian Hermitian.
[Hint: Use complex integration by parts in the form 兰⺤2 dz ∧ dz (z) ⭸(z)
⫽ 兰⺤2 (z) d( dz) ⫽ 养 (z)(z) dz ⫺ 兰⺤2 dz ∧ dz ⭸(z).]
However, there is a consistency condition to be satisfied, which stems
from a general theorem about Hermitian line bundles due to Weil [12] (a
simple proof taken from ref. 11 is reproduced in Appendix B). In our case,
it can be found as follows: by successively applying the previous relations,
we get
(z ⫹ L1 ⫹ iL2) ⫽ (z ⫹ L1) exp{⫺iL2 (z ⫹ L1) ⫹ –21 L22 ⫹ ␦2}
⫽ (z) exp{(L1 ⫺ iL2) z ⫹ –21 앚L1 ⫹ iL2앚2 ⫹ i␦1 ⫹ i␦2 ⫺ iL1L2}
⫽ (z ⫹ iL2) exp{L1(z ⫹ iL2) ⫹ –21 L21 ⫹ ␦1}
⫽ (z) exp{(L1 ⫺ iL2) z ⫹ –21 앚L1 ⫹ iL2앚2 ⫹ i␦1 ⫹ i␦2 ⫹ iL1L2}
(4)
Hence
2L1L2 ⫽ 2N
which is the Dirac–Weil–Kostant–Souriau quantization condition. Let us
conclude this section by giving the explicit expression for 具앚H앚典, which
exhibits H as a positive operator:
具앚H앚典 ⫽ 冮⺤2
[dz] e⫺앚z앚 앚⭸앚2
2
兺 e⫺2nL2/L1
2
(z ⫹ iL2) ⫽ e(1/2)(z⫹iL2) cn e2inz/L 1
n⫽⫺⬁
兺 cn e2inz/L e⫺iL2z⫹(1/2)L2
2 2
⫽ e(1/2)z 1
n⫽⫺⬁
which gives
⬁ ⬁
兺 cn e2inz/L 兺 cn e2inz/L
2
e2iL2z⫺L2 1 e⫺2nL2/L1 ⫽ 1
n⫽⫺⬁ n⫽⫺⬁
兺 兺
2
cn e2i(n⫹N)z/L1 e⫺2nL2/L1⫺L2 ⫽ cn e2inz/L1
n⫽⫺⬁ n⫽⫺⬁
再 (z) ⫽ ᏺ e(1/2)z
2
兺
n⬅ mod(N)
exp ⫺ 冢
n2L2 2niz
NL1
⫹
L1 冣冟
⫽ 0, 1, . . . , N ⫺ 1 冎
We can obtain a new representation in terms Gaussian functions, very conve-
nient for a practical calculation of , by applying the Poisson summation
formula (see, e.g., ref. 8). We find
⬁
兺
2⫹2iz/L
(z) ⫽ ᏺ e(1/2)z 1 exp{⫺ (z ⫹ nL1/N ⫹ i L2/N )2}
n⫽⫺⬁
2
(Ta)(z) ⫽ eaz⫺(1/2)앚a앚 (z ⫺ a)
where the value of should be found through the twisted periodicity condi-
tions given in Eq. (3). It is readily checked that:
1. Ta formally commute with the Hamiltonian, i.e., with the differential
operator of Eq. (2).
2. TaTbT⫺aT⫺b ⫽ exp{ab ⫺ ab}
3. Ta does not in general leave the ground state invariant, i.e., invariance
is maintained only if Na is trivial, that is, a ⫽ (n1L1 ⫹ in2L2)/N.
4. The formal infinitesimal generators of Ta , namely ᒍ1 ⫽ iz ⫺ i(⭸ ⫹
⭸) and ᒍ2 ⫽ ⫺iz ⫺ i(⭸ ⫺ ⭸), do not leave the space of sections
[Eq. (3)] invariant.
To begin with the last statement, it is clear that we may consider the linear
combinations ⭸ and z ⫺ ⭸, neither of which is such as to transform sections
into sections. From the group point of view, let l be a translation in ⺪2, i.e.,
l ⫽ k1L1 ⫹ ik2L2, k 苸 ⺪. Let us consider Ta.
We find
546 Onofri
7. CONCLUSIONS
The problem of a constant magnetic field transverse to a torus raises
the problem of translational symmetry. By quantizing the system according
to the standard mathematical formulation of gauge theory, we have shown
that the symmetry is broken to ZN ⫻ ZN. The conclusion to which one is led
by this result is that the ambiguity in quantization, namely the two arbitrary
phases ␦1, ␦2, entering in the definition of the domain of the Hamiltonian
operator, represent some physical degree of freedom of the magnetic charge
distribution generating the uniform field on the torus: monopole charges
have, so to speak, horns. The effect is purely quantum mechanical and we
empirically established that it vanishes approximately as exp{⫺O(B/ប)}. The
mathematical roots of the result are the classic theorems of Weil (see ref.
12, Chapter VI, Proposition 3, n. 3); a thorough study of -functions can be
found in ref. 3.
APPENDIX B. DIRAC–WEIL–KOSTANT–SOURIAU
QUANTIZATION CONDITION
A general theorem (ref. 6, Theorem 21.1) relates the dimension of spaces
of closed holomorphic forms on complex vector bundles to geometrical
objects, namely Chern and Todd classes of the base space and of the bundle.
In the simple case of a line bundle (fiber equal to ⺓) over a complex two-
dimensional Riemann surface, the theorem reduces to a simple result which
has a very intuitive flavor from the point of view of geometric quantization:
the dimension of the physical Hilbert space coincides with the volume of
phase space in units ប plus a constant given by half the Euler characteristic
of the surface. This in turn implies that the volume of phase space must be
an integer. We report here what appears to be the simplest proof, covering
Dirac’s quantization condition, combining ideas from refs. 1 and 11. Let us
build a triangulation of the surface ᏹ with vertices ␣, , ␥, . . . . Let ᐁ␣
denote the union of all triangles having ␣ as vertex. By taking a sufficiently
fine mesh, nonempty intersections ᐁ␣ 艚 ᐁ consist of the union of two
triangles which share the side ␣ ⫺ . A gauge field on ᏹ is given by a
closed two-form B; in each local chart ᐁ␣, we define a potential A␣ such
that B ⫽ dA␣ in ᐁ␣. According to the Poincaré lemma, for neighboring local
charts, we have
A␣ ⫺ A ⫽ d␣
with differentiable transition functions ␣ which are antisymmetric in their
indices. On triple intersections (any triangle ᐁ␣ 艚 ᐁ 艚 ᐁ␥), we have
A␣ ⫺ A ⫽ d␣, A ⫺ A␥ ⫽ d␥, A␥ ⫺ A␣ ⫽ d␥␣
It follows that c␣␥ ⬅ ␣ ⫹ ␥ ⫹ ␥␣ is constant on each triangle.4 Let us
introduce a line bundle associated to B: it is given locally by a direct product
ᐁ␣ ⫻ ⺓ in such a way that in any overlap the complex fibers are connected by5
␣ ⫽  exp{i␣}
For consistency, on any triple overlap it must hold that
exp{i␣ ⫹ i␥ ⫹ i␥␣} ⫽ 1
which implies thar c␣␥ must be an integer multiple of 2. This is usually
4
It is useful to regard the relation between A, , and c in terms of the coboundary operator:
(␦A)␣ ⫽ d␣, c␣␥ ⫽ (␦)␣␥. See ref. 1.
5
In the physical application, the phase is e/បc.
548 Onofri
referred as the Weil theorem on holomorphic vector bundles (ref. 12, Chapter
V, Proposition 1, n. 4).
The key result for our purposes is the following:
Theorem. The integral 兰ᏹ B coincides with the discrete sum 兺⌬c⌬, where
⌬ runs over all triangles of the mesh.
Proof. The following, purely algebraic identity holds (ref. 11, p. 131):
冮⌬␣␥
B⫽
1
3
冖
⭸⌬␣␥
(A␣ ⫹ A ⫹ A␥)
1
⫽ [(␣ ⫹ ␥ ⫹ ␥␣)(␣) ⫹ (␣ ⫹ ␥ ⫹ ␥␣)()
3
1
⫹ (␣ ⫹ ␥ ⫹ ␥␣)(␥)] ⫺ {(␣(␣) ⫹ ␣())
2
⫹ (␥() ⫹ ␥(␥)) ⫹ (␥␣(␥) ⫹ ␥␣(␣))}
⫹
1
2 再冮
␣
(A␣ ⫹ A) ⫹ 冮
␥
(A ⫹ A␥) ⫹ 冮␥␣
(A␥ ⫹ A␣) 冎
Notice that the terms in curly brackets average to zero when we sum over
the whole triangulation, while the terms in square brackets are precisely the
cocycle c⌬, whose value is constant on the triangle. Hence we get
冮 ᏹ
B⫽ 兺⌬ c⌬
and as a result, the flux of B is quantized.
ACKNOWLEDGMENTS
This research was supported by Italian MURST under contract
9702213582 and by I.N.F.N. under i.s. PR11. I warmly thank P. Maraner and
C. Destri for interesting discussions and for directing my attention to refs. 3
and 5.
REFERENCES
1. O. Alvarez (1985). Commun. Math. Phys. 100, 279.
2. H. Aoki (1987). Rep. Progr. Phys. 50, 655.
3. B. A. Dubrovin (1981). Russ. Math. Surv. 36, 11.
4. R. Ferrari, In Elementary Particles, Quantum Fields and Statistical Mechanics, M. Bonini,
G. Marchesini, and E. Onofri, eds. (Università di Parma), pp. 155–201.
Landau Levels on a Torus 549