Katzgraber - Topology in Physics

Topology in Physics
Proseminar
in Theoretical Physics
Institut f
ur theoretische Physik
ETH Honggerberg
CH-8093 Z
urich
Switzerland
July 31, 2006
Contents
1 Lie Algebras: a crash course . . . . . . . . . . . . . . . .
Cyril Stark (under the supervision of Urs Wenger)
2 Goldstone bosons and the Higgs mechanism . . . . 51

Stefan Pfenninger (under the supervision of Stefan Fredenhagen)
3 Magnetic monopoles: Dirac and Polyakov-tHooft
75
Philip Grech (under the supervision of Urs Wenger)
4 Dislocations in crystals . . . . . . . . . . . . . . . . . . . . 103

David Gablinger (under the supervision of Urs Aeberhard)
5 Solitons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
Sami Lang (under the supervision of Ilka Brunner)
6 Solitons in density wave systems . . . . . . . . . . . . . 151

Mathias Ritzmann (under the supervision of Urs Aeberhard)
7 Vortices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Thomas Eggel (under the supervision of Stefan Fredenhagen)
8 BKT transition in spin systems . . . . . . . . . . . . . . 197

Lucas Lombriser (under the supervision of Munehisa Matsumoto)
9 Berry Phase I: Mathematical Concepts . . . . . . . . 227

Simon Wood (under the supervision of Sebastian Pilgram)
10 Berry phase II: The Aharonov-Bohm effect . . . . 271

Oliver Pfaffli (under the supervision of Sebastian Pilgram)
11 Berry Phase III: Berry phase and Bloch states . . 293

Juan Carlos Andresen (under the supervision of Sebastian Pilgram)
12 Superfluidity and winding numbers . . . . . . . . . . 311

Barbara Theiler (under the supervision of Lode Pollet)
iii
iv
13 Introduction to quantum computing . . . . . . . . . 339

Lukas Gamper (under the supervision of Lode Pollet)
14 Topologically protected quantum computing . . . 361

Bojan Skerlak (under the supervision of Helmut Katzgraber)
15 Topological quantum computing implementations 393

Mario Berta (under the supervision of Fabricio Albuquerque)
16 Braid group statistics in quantum mechanics . . . 421

Klaus Jager (under the supervision of Walid Abou Salem)
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
c 2006 H. G. Katzgraber (Theoretische Physik ETH Z

urich)
Lie Algebras: a crash course
Cyril Stark
Supervisor: Urs Wenger
The goal of this paper is to give an introduction to the representation theory of Lie algebras. In the first part, the fundamentals
of Lie groups and their Lie algebras are presented in a very compact form. In the second and the third part, the representations
of su(2) and su(3) are discussed and basic terms are introduced.
In the forth part follows the generalization of the machinery introduced in the analysis of su(2) and su(3). The last part is
related to the classification of complex simple Lie algebras. Most
of the proofs are omitted in the first, the forth and the fifth part.
Introduction
Lie algebras are closely connected to symmetries. Their use in the description
of physical systems allows to apply algebraical methods to get further insight.
Concrete applications of Lie theory appear in the description of atomic, molecular
and nuclear spectra or in gauge theories. Furthermore, Lie algebras are related to
several algebraical structures like group algebras, Hopf algebras, quantum groups
and vertex operator algebras which are used in mathematical physics. The goal
of this paper is to provide an informal introduction to Lie theory. A detailed
presentation of the subject is given in [1, 2].
Fundamentals
Definition of a Lie Group

Definition (Lie Group). A Lie group G is a set G that has two structures
that are compatible to each other: the structure of a group and the structure
of a C -manifold. Compatible means that the operations multiplication and
inverse associated to the group structure of G are differentiable maps (C ).
1
Topology in Physics
Definition 1. Let G and H be two Lie groups. A homomorphism between Lie

groups is a group homomorphism which is differentiable. In the remainder of this
section we will simply write homomorphism instead of homomorphism between
Lie groups.
Definition 2. Let G be a Lie group. A representation of G is a homomorphism
between Lie groups from G to GL(V ).
The matrix group GLn (R) is a basic example for a Lie group. This group
is an open subset of the vector space of all n n matrices and therefore a C manifold. Obviously, the multiplication is differentiable as well as the inverseoperation (compare Cramers rule).
Whenever no basis is specified, the maps corresponding to the matrices in GLn (R)
must be considered as automorphisms. The set of all automorphisms V V is
denoted by GL(V ).
Definition 3. A complex Lie group is a complex manifold endowed with a group
structure in such a way that the operations product and inverse associated to
the group structure are holomorphic maps.
A complex manifold is a smooth manifold of dimension 2n such that the local
charts (U , ) have the property that the change-of-coordinate map are
holomorphic.
Definition and basic properties of Lie Algebras
The ultimate goal is the study of representations of Lie groups. For that purpose
one makes use of the differentiable structure of the Lie group. More precisely,
we will try to determine a characterization of : G H being a Lie group
homomorphism expressed by maps between tangent spaces of G and H.
The right idea is to look at the automorphisms of G given by conjugation:
g
i.e.,
: G
h
G
g (h) := g h g 1 ,
: G GL(G),
(1.1)
where GL(G) denotes the set of all automorphisms on G. We observe

group homomorphism g = (g) ,
(1.2)
since
g (h) = (g)(h)1 (g) = (g) (h).
We are going to differentiate (at the identity) two times, i.e., we use that is
differentiable. After the second differentiation we will deduce an identity based on
the assumption that is a group homomorphism that involves only the differential
of .
Differentiation at the identity of (1.2) yields
de d(g )e = d((g) )e de .
We define a new map to reformulate this result.
Definition 4. The adjoint map Ad is defined as
Ad : G
g
GL(Te G)
Ad(g) := d(g )e : Te G Te G.
Thus, the adjoint map Ad realizes a representation of G called the adjoint representation of the Lie group G.
Therefore,
group homomorphism de Ad(g) = Ad((g)) de .
(1.3)
This identity still explicitly depends on (g). Thus, we differentiate a second

time to get
de d(Ad)e (X)(Y ) = d(Ad)e (de (X), de (Y )).
(1.4)
Definition 5. One defines ad := d(Ad)e . The tangent space at GL(Te G) can be
identified with the entire space End(Te G). Thus,
ad :
Te G
X 7
End(Te G)
ad(X) := d(Ad)e (X) : Te G Te G.
Therefore, we can consider the map ad to be a bilinear map

Te G Te G Te G.
Notation:
[X, Y ] := ad(X)(Y ).
Using this definition in identity (1.4) yields the following important identity
(Assumption: is a homomorphism)
de ([X, Y ]) = [de (X), de (Y )].
(1.5)
We can even state the following two assertions which are much stronger (without
proof):
Theorem 1. Let G and H be two Lie groups, with G connected. Then: Each
homomorphism : G H is uniquely determined by its differential de :
Te G Te H at the identity.
Topology in Physics
Theorem 2. Let G and H be two Lie groups, with G connected and simply
connected. Then: A linear map L : Te G Te H is the differential of a
homomorphism : G H iff it preserves the bracket operation, i.e.,
L([X, Y ]) = [L(X), L(Y )].
Observation 1. Properties of the bracket operation [, ] on Te G:
1. Skew-symmetry, i.e.,
[X, Y ] = [Y, X].
2. Satisfaction of the Jacobi identity, i.e.,
[X, [Y, Z]] + [Z, [X, Y ]] + [Y, [Z, X]] = 0.
A general vector space with such a structure is called a Lie algebra:
Definition (Lie Algebra). A vector space V together with a skew-symmetric
bilinear form
[, ] : V V V
which satisfies the Jacobi identity is called a Lie algebra (usually denoted g).
One can also show that the contrary is true as well, i.e., that the tangent
space at a Lie group has exactly the structure of a Lie algebra. Thats the reason
why the tangent space Te G of a Lie group G is called the Lie algebra g of the Lie
group G.
An even stronger assertion is (without proof)...
Proposition 1. Every finite-dimensional (abstract) Lie algebra is the Lie algebra
of a Lie group.
A real (complex) Lie algebra is a Lie algebra whose vector space is a real
(complex) vector space.
Definition 6. A homomorphism L : g1 g2 between Lie Algebras g1 and g2
is a linear map of vector spaces which preserves the bracket operation, i.e.,
L([X, Y ]g1 ) = [L(X), L(Y )]g2 .
Definition 7. A Lie subalgebra is a linear subspace which is closed under the
bracket.
Let us try to explicitly compute the bracket operation for G = GLn R.
The Lie group GLn (R) is an open subset of the space End(Rn ). Thus, we can
identify End(Rn ) with the tangent space Te GLn (R) and the conjugation g on
GLn (R) extends naturally to End(Rn ) = Te GLn (R), i.e.,
Ad(g)(X) d(g )e (X) = g X g 1 ,
for all X Te GLn (R) = End(Rn ).
Let X and Y be two tangent vectors to GLn (R) and let : I R GLn (R)
be a curve such that (0) = e and such that 0 (0) = X. Then
[X, Y ] ad(X)(Y )
d
=
|t=0 Ad((t))(Y )
dt
d
=
|t=0 ((t)Y 1 (t))
dt
= 0 (0) Y 1 (0) + (0) Y ((0)) 0 (0)
= X Y Y X.
(1.6)
Definition (Representation). A representation of a Lie algebra g on a vector

space V is a homomorphism between Lie algebras
: g gl(V ) = End(V ),
i.e., is a linear map form the vector space corresponding to g to the vector
space corresponding to gl(V )such that
([X, Y ]g ) = [(X), (Y )]gl(V ) = (X) (Y ) (Y ) (X),
for all X, Y g.
The last equation follows from our discussion about GLn (R) with the only
difference that this time gl(V ) is considered to be the space of automorphisms
and not to be a space of matrices (i.e., no basis has been specified).
As corollaries of theorem 1 and theorem 2 we can state the following crucial
observation.
Crucial Observation. The representations of a connected and simply connected
Lie group are in one-to-one correspondence to the representations of its Lie algebra.
Definition 8. Let V be a finite-dimensional real vector space. Then the complexification VC of the vector space V is the space of formal combinations
v1 + iv2 ,
with v1 , v2 V . This is a real vector space and becomes a complex vector space,
if we define
i(v1 + iv2 ) iv1 v2 .
Topology in Physics
Proposition 2. Let g be a finite-dimensional real Lie algebra and gC its complexification to a complex vector space. Then there exists a unique extension of
the bracket operation of the Lie algebra g to the vector space gC which becomes a
Lie algebra.
Proof. Since the extension of the bracket must still be bilinear, we have that
[X1 + iX2 , Y1 + iY2 ] = ([X1 , Y1 ] [X2 , Y2 ]) + i([X1 , Y2 ] + [X2 , Y1 ]).
Therefore the extension is unique. To proof the existence of the extension of
the Lie bracket we must check that the new bracket fulfills the demands in the
definition of a Lie bracket. The extension is obviously skew symmetric and real
bilinear since this is true for the bracket on g.
To check complex bilinearity we only have to check (think of skew-symmetry)
that the new bracket is complex-linear in one argument:
[i(X1 + iX2 ), Y1 + iY2 ] =
=
=
=
[iX1 X2 , Y1 + iY2 ]
([X1 , Y2 ] [X2 , Y1 ]) + i([X1 , Y1 ] [X2 , Y2 ])
i(([X1 , Y1 ] [X2 , Y2 ]) + i([X1 , Y2 ] + [X2 , Y1 ]))
i[X1 + iX2 , Y1 + iY2 ].
What is left to check is the Jacobi identity. We first observe (using bilinearity)
that
[X, [Y, Z]] + [Z, [X, Y ]] + [Y, [Z, X]] = 0,
for all X gC and Y, Z g. The same argument applies to Y, Z and it follows
that the Jacobi identity holds for the extended bracket on gC .
Definition 9. The vector space gC together with the unique extension of the
bracket operation of a real Lie algebra g is called the complexification gC of the
real Lie algebra g.
Proposition 3. Let g be a real Lie algebra and gC its complexification. Then
an arbitrary finite-dimensional complex representation
of g can be extended
uniquely to a complex-linear representation of gC given by
(X + iY ) =
(X) + i
(Y )
for all X, Y g.
In addition we have that is irreducible iff
is irreducible.
Proof. Existence and uniqueness are trivial.
Let us prove the statement concerning the irreducibility.
: Let
be irreducible and let W be a complex linear subspace invariant
under . This implies that W is invariant under
. Therefore W is equal to {0}
or to the whole representation space is irreducible.
: Let be irreducible and let W be a complex linear subspace invariant

under
. We deduce that (W )
(W ) + i
(W ) W since W is a complex
vector space. Therefore W is equal to {0} or is equal to the entire representation
space
is irreducible.
Lemma 1. su(n)C
= sl(n, C)
Proof. The complexification su(n)C of the
are subalgebras of the Lie algebra gl(n; C)
plexification). The Lie algebra sl(n; C) is
n n-matrices. We can write each element
X=
real Lie algebra su(n) and sl(n; C)

(compare the definition of the comthe space of complex and traceless
X of sl(n; C) (uniquely) in the form
X X
X + X
+i
,
2
2i
where (X X )/2 and (X + X )/2i are both traceless and skew and are therefore
elements of su(n). This implies the claim.
The Exponential Map
Let G be a Lie group with Lie algebra g, X g and g G. The map mg : G
G is the diffeomorphism defined by left-multiplication with g, i.e., mg (h) := g h
for all h G.
Now we introduce a vector field on the Lie group G defined by
vX (g) := (mg ) (X).
(1.7)
((mg ) (X) d(mg )mg1 X.) This is locally integrable. Thus, there exists I R,
: I G differentiable with (0) = e, such that
0X (t) = vX (X (t))
(1.8)
for all t I. Since (s + t) = (s) (t) and since is defined on a finite interval
I, we can deduce that extends uniquely to all of R:
X : R G.
(1.9)
Definition 10. Let G be a Lie group with Lie algebra g. We define the exponential map by
exp
g
X
G
7 exp(X) := X (1).
Properties of the exponential map:
Topology in Physics
1. Let be a Lie group homomorphism. Then the following diagram commutes
g h
exp
exp
y
y
G H
2. The exponential map is differentiable. Its derivative at the origin of g is the
identity map. Thus, the exponential map realizes a diffeomorphism from a
neighborhood of the origin of Te G
= g to a neighborhood of the identity
element e G. Such a neighborhood of e G generates the identity
component of G.
3. Let G be a compact Lie group with Lie algebra g. Then: exponentiation of
a representation of g yields a representation of G.
4. X (t) = exp(t X)
5. exp((t + t0 ) X) = exp(t X) exp(t0 X)
6. exp(X) = (exp(X))1
7. Let G be a matrix Lie group. Then: the exponential map is equal to the
exponential map of matrices, i.e.,
exp(X) =
X
Xk
k=0
k!
for all X G.
First Classification of Lie Algebras
As it is usual in mathematics, we have started with the discussion of a certain
mathematical structure (here the Lie algebra). Once one doesnt get any further
in the general analysis, one tries to organize the different manifestations of the
general structure, i.e., one tries to classify the original structure. The next step
is the investigation of the different classes. The following classification reflects
how much a Lie algebra fails to be Abelian.
Definition 11. Let g be a Lie algebra and let h g be a subalgebra. The
subalgebra is an ideal if
[X, Y ] h
for all X h and Y g.
The bracket operation on g induces a bracket operation on the quotient space

g/h iff h is an ideal of g.
Therefore, we make the next definition.
Definition 12. A Lie algebra is called simple if dim g > 1 and if it contains no
nontrivial ideals.
Definition 13. The lower central series Dk g of subalgebras is defined by
D1 g := [g, g],
Dk g := [g, Dk1 g].
The subalgebras Dk g are ideals in g.

Definition 14. The derived series Dk g of subalgebras is defined by
D1 g := [g, g],
Dk g := [Dk1 g, Dk1 g].
The subalgebras Dk g are ideals in g and Dk g Dk g for all k.

Furthermore, D1 g = D1 g =: Dg.
Definition 15. Let g be a Lie algebra. Then:
1. g is called nilpotent if Dk g = 0 for some k.
2. g is called solvable if Dk g = 0 for some k.
3. g is called perfect if Dg = g.
4. g is called semisimple if g has no nonzero solvable ideals.
The term semisimple will become clearer in the section about semisimple Lie
algebras.
The sum of two solvable ideals of a Lie algebra g is again solvable.
Definition 16. The sum of all solvable ideals of g is a maximal solvable ideal,
called the radical Rad(g) of g.
The following proposition can be used to determine the radical of g (the
definition of the Killing form follows in section 5).
Proposition 4. Let g be a Lie algebra. Then: the radical Rad(g) of g is the
orthogonal complement to Dg with respect to the Killing form.
The quotient g/Rad(g) is semisimple according to the definition of semisimplicity. A consequence of Lies theorem (compare Fulton and Harris 9.2) is that
any irreducible representation of a solvable Lie algebra is one-dimensional. One
arrives at the following theorem which describes how to find irreducible representations of an arbitrary complex Lie algebra g.
10
Topology in Physics
Theorem 3. Let g be an arbitrary complex Lie algebra and let gsemisimple =

g/Rad(g) be its semisimple component.
Then: Every irreducible representation of g is of the form
= semisimple L,
where semisimple is an irreducible representation of gsemisimple (i.e., a representation of g that is trivial on Rad(g)), and L is a one-dimensional representation.
Simple Lie Algebras
As we will see in section 6, every simple complex Lie algebra is either isomorphic to sln (C), son (C), sp2n (C) (the symplectic Lie algebra) or to one of the so
called exceptional Lie algebras.
Semisimple Lie Algebras
There are two fundamental problems concerning the representations of general
Lie groups and general Lie algebras: first, the complete reducibility property of
representations is not valid and second, the action of an element in a Lie group
or in a Lie algebra can be diagonalizable under one representation and not under
another. This problems vanish immediately as soon as we pass to semisimple Lie
algebras.
Theorem (Complete Reducibility). Let g be a semisimple Lie algebra and let
be an arbitrary representation of g with representation space V . Assume that
W V is an invariant subspace under the action of g.
Then: there exists a subspace W 0 V complementary to W and invariant under
the action of g.
Recall that we can write any endomorphism X of a complex vector space V
in the form
X = Xs + Xn ,
where Xs is diagonalizable and Xn is nilpotent.
Theorem (Preservation of Jordan Decomposition). Let g be a semisimple
Lie algebra. Then there exist for any X g elements Xs g and Xn g such
that
(X)s = (Xs ), (X)n = (Xn )
for any representation , where (X)s is diagonalizable and where (X)n is nilpotent.
In addition, the following proposition holds:
11
Proposition 5. The following is equivalent:

1. g is a semisimple Lie algebra.
2. g is a direct sum of simple Lie algebras.
3. The Killing form is non-degenerate (i.e., the set {x V | < x, y >Killing =
0 f or all y V } is trivial).
4. g has no nonzero Abelian ideals.
5. The radical of g is zero.
Representations of su(2)
The goal of this section is to analyze the irreducible complex representations of

su(2). Motivation: First, in the analysis of the representations of general semisimple Lie algebras one uses the representations of sl(2, C)
= su(2), second, su(2) is
an easy example of how one uses commutation relations to get the representations
of a Lie algebra and third, su(2) is of physical importance.
In proposition 3 we saw that every finite-dimensional complex representation of
su(2) extends uniquely to a complex-linear representation of the complexification
su(2)C . According to Lemma 1 this complexification is isomorphic to sl(2, C).
Therefore the extension of a representation to sl(2, C) is irreducible iff the representation of su(2) is irreducible.
We conclude that the study of the irreducible representations of su(2) is equivalent to the study of the irreducible complex-linear representations of sl(2, C).
We prefer the analysis of sl(2, C) to the analysis of su(2) since we can work with
the following nice basis in the case of sl(2, C) which will simplify our computations:

1 0
0 1
0 0
H=
; X=
; Y =
,
(1.10)
0 0
1 0
0 1
with
[H, X] = 2X, [H, Y ] = 2Y, [X, Y ] = H.
(1.11)
Assume that is an arbitrary irreducible finite-dimensional representation of

sl(2, C) on a vector space V .
As a first observation we have the following remark which is a result of the
preservation of the Jordan decomposition under a representation (compare section
2):
Remark 1. (H) is diagonalizable and therefore
M
V =
V ,
where the index runs over the eigenvalues of (H) which are complex numbers.
12
Topology in Physics
With the decomposition of V into eigenspaces of (H) we have brought a first

structure (wrt the action of (H)) into the representation space V . As a result of
the commutation relations (1.11) we will see in the following that this structure of
V will be preserved under the action of (X) and (Y ) ( the structure will be
preserved under the action of (sl(2, C))). This will equip us with much insight
into representations of sl(2, C).
After this outlook let us try to answer the following question: How do (X)
and (Y ) act on the decomposition in Remark 1?
L
Remark 2. Let W =
W be a decomposition of a vectorspace W into eigenspaces
W of a linear Operator L acting on W with eigenvalues C. Whenever one
wants to determine in which eigenspace the vector K(w) (w W ; K being
another linear Operator on W ) lies, we simply apply L:
L(K(w)) = K(w), C K(w) W .
Therefore the answer to the question above is given by the following calculation using the commutation relations (1.11):
(H) (X)(v) = (X) (H)(v) + [(H), (X)](v)
= (X) (H)(v) + 2(X)(v)
= ( + 2) (X)(v),
f or v V . (1.12)
Similarly,
(H) (Y )(v) = (Y ) (H)(v) + [(H), (Y )](v)
= (Y ) (H)(v) 2(Y )(v)
= ( 2) (Y )(v),
f or v V .
(1.13)
We summarize these results in the following

Lemma 2.
(H) : V V
(X) : V V+2
(Y ) : V V2
From the 3 facts that

1. is irreducible ( the only invariant subspaces are V and {0}),
2. only those subspaces V and V are getting mapped into each other by the
action of sl(2, C) iff is an integer multiple of 2 and
3. only finitely many linear subspaces can contribute to a direct-sum-decomposition
of V (since the vector space V is finite-dimensional)
13
we deduce that
Lemma 3. There exists C and k Z, such that
V = V V+2 V+4 ... V+2k .
Define n := + 2k C.
Lemma 4. Let v Vn+2k . Then
V = span{v, (Y )(v), (Y )2 (v), (Y )3 (v), ...}.
Proof. Define W span{v, (Y )(v), (Y )2 (v), (Y )3 (v), ...}. Recall that the
representation is irreducible by assumption. This implies that it is sufficient to
show that W is invariant under the action of X, Y and H. Again we will use the
commutation relations (1.11).
W is preserved under the action of Y since (Y ) (Y )m (v) = (Y )m+1 (v) W
for all m N.
W is preserved under the action of H since all the basis elements of W lie in
different eigenspaces of (H) which are all preserved under the action of H.
It is left to show that
(X) (Y )m (v)) W,
v W.
Consider the case m = 0:

(X)(v) = 0
(compare Lemma 3).
(X) (Y )(v) = (Y ) (X)(v) + [(X), (Y )](v)
= 0 + (H)(v)
= nv.
(X) (Y )((Y )(v)) =
=
=
=
(Y ) (X)((Y )(v)) + [(X), (Y )]((Y )(v))

(Y ) (nv) + (H)((Y )(v))
n(Y )(v) + (n 2)(Y )(v)
(n + (n 2)) (Y )(v).
Repeating these calculations yields

(X) (Y )m (v)) = [n + (n 2) + (n 4) + . . . + (n 2(m 1))] (Y )(m1) (v)
= m(n m + 1) (Y )(m1) (v).
(1.14)
14
Topology in Physics
The last step can easily be proved by induction (m m+1, the cases m = 0, 1, 2
have already been checked above):
n + (n 2) + (n 4) + . . . + (n 2(m + 1 1)) = n + (n 2) + (n 4) + . . .
+(n 2(m 1)) + n 2m
= m(n m + 1) + n 2m
= mn m2 + m + n 2m
= (m + 1)(n (m + 1) + 1) X.
We have shown that (Y )m (v) is mapped to a multiple (Y )m1 (v) W is
preserved under the action of X.
W = V and {v, (Y )(v), (Y )2 (v), (Y )3 (v), ...} forms a basis of V .
Corollary 1. For each eigenspace V of (H) we have that
dim V = 1.
Since we know how the basis elements H, X and Y act on the basis vector
of each eigenspace V , we know how the representation of sl(2, C) acts on the
vector space V as linear maps. Thus, we know
L how the representation of
sl(2, C) acts on the whole vector space V =
V and we arrive at the second
corollary of lemma 4
Corollary 2. Every irreducible representation of sl(2, C) is determined by the
set { C|V is an eigenspace of (H)}.
Let m N be the smallest non-negative integer such that (Y )m (v) = 0 with
v Vn . We apply (X) and use identity (1.14) to get
0 = (X) (Y )m (v) = m(n m + 1) (Y )(m1) (v).
(1.15)
Since (Y )(m1) (v) is by assumption not equal to zero we get:

0 = (n m + 1) m = n + 1.
(1.16)
But the integer m is also equal to the dimension of V . Thus, we deduce, that
the representation space V is n + 1-dimensional (compare lemma 2, lemma 3 and
corollary 1).
L
This implies that the direct decomposition V =
V of V into one-dimensional
eigenspaces V of (H) consists of n + 1 terms. We deduce
V
= Vn Vn2 . . . Vn2n
= Vn Vn2 . . . Vn+2 Vn .
We can summarize the former results in the following theorem.
(1.17)
15
Theorem 4. Let be an irreducible representation of sl(2, C) with representation

space V . Then the spectrum of (H) is a subset of Z of the form
{n, n + 2, n + 4, . . . , 2, 0, 2, . . . , n 2, n}
or
{n, n + 2, n + 4, . . . , 1, 1, . . . , n 2, n}.
Each irreducible representation of sl(2, C) can be labeled by the largest eigenvalue n of (H). Each eigenspace of (n) (H) is one-dimensional and V (n) =
L
(n)
k{n, n+2, ..., n2, n} Vk .
Corollary 3. The number of irreducible factors in an arbitrary representation
of sl(2, C) is equal to the multiplicities of 0 and 1 as eigenvalues of (H).
Representations of su(3)
Definitions and first Observations

As in the case of su(2) we observe that we are allowed to study the representations
of sl(3, C) instead of su(3), since the representations of sl(3, C) are in one-to-one
correspondence with the complex-linear representations of the complexification
suC (3)
= sl(3, C) (compare proposition 3 and lemma 1).
The main result of our discussion about the irreducible representation of su(2)
was that each irreducible representation can be classified in term of the largest
eigenvalue of (H).
Thus, it arises the natural question, which element in (sl(3, C)) will take over
the role of (H). The answer is that there isnt any such element in (sl(3, C)).
But instead we will use a whole subspace h of (sl(3, C)).
The main goal of this section is to classify the irreducible representations of
sl(3, C) (as we have done it in the analysis of (sl(2, C))). We will prove that
these representations are determined (up to equivalence) by their highest weight
(to be defined later).
The study of sl(3, C) will equip us with the concepts necessary to classify the
finite-dimensional representations of all semisimple Lie algebras.
We define h sl(3, C) to be the two-dimensional subspace of all diagonal
matrices in sl(3, C) (this is the so called Cartan subalgebra for sl(3, C)).
a1 0 0
h := { 0 a2 0 |a1 + a2 + a3 = 0}
(1.18)
0 0 a3
h = {1 L1 + 2 L2 + 3 L3 | 1 , 2 , 3 C}/{L1 + L2 + L3 = 0}, (1.19)
16
Topology in Physics
with
a1 0 0
Li 0 a2 0 := ai .
0 0 a3
(1.20)
Definition 17. Let be an arbitrary representation of sl(3, C) with representation space V . Then we define
1. A vector v V is a weight vector if v is a simultaneous eigenvector for
(H) for each H h.
2. Let v be a weight vector with corresponding eigenvalue (H) for H h.
This quantity (H) is a complex number depending linearly on H h, i.e.,
h . The linear functional is called a weight of the representation.
3. All the weight vectors corresponding to a weight span a vector space V
which is a so called weight space.
4. The dimension of the weight space V is called the multiplicity of the weight
.
According to the preservation of the Jordan decomposition (compare section 2) all elements in h are diagonalizable in any representation of sl(3, C).
Additionally, all elements in h commute in any representation of sl(3, C) since
they commute in sl(3, C). Recall that commuting, diagonalizable matrices are
simultaneously diagonalizable.
Remark 3. This implies that any finite-dimensional representation space V decomposes into a direct sum of weight spaces (i.e., simultaneous eigenspaces) V
of all elements in h:
M
V =
V ,
where h ranges over a finite subset of h since V is finite-dimensional by
assumption.
Now that we know how to replace the basis element H from the discussion
of sl(2, C) we should determine how to replace the basis elements X and Y .
Recall that
ad(H)(X) = [H, X] = 2X, ad(H)(Y ) = [H, Y ] = 2Y,
(1.21)
i.e., the elements X and Y sl(2, C) are eigenvectors for the adjoint action
of H sl(2, C). Thus, to replace X and Y sl(2, C) we have to look for
simultaneous eigenvectors for the adjoint action of h (i.e., weight vectors in the
case of the adjoint representation). This yields a decomposition
M
sl(3, C) = h (
g ),
(1.22)
17
where ranges over a finite subset of h as explained in remark 3. The subspaces

g sl(3, C) are the analogue of the subspaces V appearing in remark 3, i.e.,
ad(H)(X) = [H, X] = (H) X,
(1.23)
for every H h and X g .

Definition 18. In case of the adjoint representation weights are called roots,
weight vectors are called root vectors and weight spaces are called root spaces.
The finite set of all roots is denoted R h .
The next goal is to determine the subspaces g sl(3, C) and to explicitly
write down the corresponding weight h .
To determine the g s observe that
a1 0 0
m11 m12 m13
[H, M ]ij = [ 0 a2 0 , m21 m22 m23 ]ij = (ai aj ) mij (1.24)
0 0 a3
m31 m32 m33
for every H h and M sl(3, C). Thus, M sl(3, C) is a root vector iff all
but one entry of M are zero. We deduce that the root spaces g are spanned by
matrices Ei,j which are defined to be the (3 3)-matrices whose (i, j)th entry is
equal to 1 and whose other entries are equal to zero. In total we span six (Ei,j
sl(3, C)) one-dimensional complex subspaces of sl(3, C) (recall that sl(3, C) is
8-dimensional, that h is 2-dimensional and remember identity (1.22)).
Example 1. We apply the above thoughts to E1,2 , which spans a certain root
space of sl(3, C):
ad(H)(E1,2 ) = (H) E1,2 = (a1 a2 ) E1,2 (H) = L1 L2 .
(Compare equation (1.20) for the last implication.)
Therefore we arrive at the
Observation 2. The root of the root space spanned by the matrix Ei,j is Li Lj .
We have seen that the representation
Lspace V decomposes into a direct sum of
weight spaces and that sl(3, C) = h ( g ). The subspace h sl(3, C) acts on
the different V s as scalar multiplication (by definition of the V s). To get more
information about how the g s act on the V s, we proceed exactly as in (1.12):
We use commutation relations. First, we have to consider the special case of the
adjoint representation (recall the Jacobi identity):
ad(H) ad(X)(Y )
=
=
=
[H, [X, Y ]]
[Y, [H, X]] [X, [Y, H]]
[[H, X], Y ] + [X, [H, Y ]]
((H) + (H)) ad(X)(Y ),
(1.25)
for every H h, X g and Y g . This crucial calculation proves the following

fundamental observation.
18
Topology in Physics
Observation 3.
ad(g ) : g g+
Let us proceed to the general case (recall that the commutation relations are
the same in every representation):
(H) (X)(v) = [(H), (X)](v) + (X) (H)(v)
= (H) (X)(v) + (H) (X)(v)
= ((H) + (H)) (X)(v),
(1.26)
for all H h, X g and v V . We deduce

Observation 4.
(g ) : V V+
As a consequence, the weights of an arbitrary irreducible representation differ
only by integral linear combinations of Li Lj , i.e., linear combinations with
integer coefficients (recall that there must not exists any other invariant subspace
than V or {0}).
Definition 19. The integral linear combinations of Li Lj generate a lattice R
called the root lattice.
Consider again the adjoint representation. From the structure of the roots
(i.e., the weights for the adjoint representation) one can deduce much knowledge
about the Lie algebra. Thus, we introduce the following diagrammatical method:
with two real
Draw the six roots Li Lj h lying in the real subspace h
dimensions as in figure 1.1.
The points symbolize the root space corresponding to each specific root. The
0 in the middle should be considered as the Cartan subalgebra h (which corresponds to the root 0 since its elements commute). The irreducibility of the
adjoint representation implies that the subspace hL
and all the g s get mapped
into each other under the action of sl(3, C) = h ( g ) (recall that there must
not be any invariant subspace different from {0} and the representation space
V = sl(3, C)). We deduce that the only possibility to draw the six roots Li Lj
is the one shown in figure 1.1 (compare also observation 3). The locations of L1 ,
L2 and L3 follow.
In the discussion of sl(2, C) we discovered an extremal eigenspace Vn for the
action of H (by extremal we mean that Vn is sent to {0} under the action of X,
i.e. Vn is sent to {0} by half of the roots).
To deduce an analog statement for the present discussion, we have to explain
what we mean by extremal in the context of sl(3, C).
Seen from another point of view, the extremal eigenspace Vn of sl(2, C) is the
eigenspace corresponding to the root n lying farthest away from the origin of
19
Figure 1.1: Root diagram for sl(3, C)

the one-dimensional vector space hsl(2,C) . In the present circumstances we are
dealing with a 2-dimensional Cartan subalgebra h . Thus, we first have to define
a specific direction (we will see that the specific choice doesnt affect the result).
More concretely, we define a linear functional
l : R C,
(1.27)
such that l never vanishes on R , i.e., the kernel of l doesnt intersect any lattice
point of R so that
R = R+ R
(1.28)
(disjoint) with l() > 0 for all R+ and l() < 0 for all R . A specific example of the present circumstances is shown in figure 1.2 (the thick line
visualizes the kernel of l).
We call a root space g extremal, if is the farthest root in the chosen specific
direction.
Observation 5. The extremal root space gL1 L3 (in case of the specific example
shown in the figure above) is sent to {0} under the action of R+ = {L2 L3 , L1
L3 , L1 L2 , }. I.e., the elements in gL1 L3 are simultaneous eigenvectors for all
elements in h and get killed by the action of half of the roots. Thus, we have
arrived at a situation which is completely analogue to the one discussed in the
case of sl(2, C).
The explicit expression of l from our specific example is of the form
l(1 L1 + 2 L2 + 3 L3 ) = 1 a + 2 b + 3 c
(1.29)
20
Topology in Physics
Figure 1.2: Visualization of a specific kernel

with a + b + c = 0 and a > b > c and 1 , 2 , 3 C.
Note that the positive root spaces are generated by the matrices Ei,j for i < j
and that the negative root spaces are generated by the matrices Ei,j for i > j.
Let us pass over to the analysis of an arbitrary finite-dimensional representation of sl(3, C) with representation space V . We summarize the observations
above in the following lemma wrt the general irreducible representation.
Lemma 5. One can always find a vector v V , such that
1. v is a simultaneous eigenvector for all elements in h and such that
2. v gets killed by the application (E2,3 ), (E1,3 ) or (E1,2 ) (i.e, by the action
of all elements corresponding to the roots in R+ ).
The vector v V in the lemma is called a highest weight vector.
In the analysis of sl(2, C) we proceeded at this point by claiming that the vectors
v, Y (v), Y 2 (v), Y 3 (v), . . . (with v being a highest weight vector) would generate
the whole representation.
Claim 1. Let be an irreducible representation of sl(3, C) with representation
space V and highest weight vector v. Then the representation is generated by the
successive application of (E2,1 ), (E3,1 ) and (E3,2 ) (which correspond to the
roots in R ) on v.
Proof. Let W be the subspace generated by the images of the highest weight
vector v under the application of (E2,1 ), (E3,1 ) and (E3,2 ). We are going to
show that W is invariant under the action of sl(3, C). Then: irreducibility of
21
will imply that V = W .

The short calculation [E2,1 , E3,2 ] = E3,1 verifies that W is spanned by the images
of successive applications of (E2,1 ) and (E3,2 ). Recall that
sl(3, C) = h C E1,2 C E1,3 C E2,3 C E2,1 C E3,1 C E3,2 .
The vector space W is obviously invariant under (h) (the operators (E2,1 ),
(E3,2 ) transform weight spaces into weight spaces; compare observation 4) and
C (E2,1 ), C (E3,1 ), C (E3,2 )(true according to the definition of W ).
It is left to check the invariance under the application of C (E1,2 ), C (E1,3 )
and C (E2,3 ):
Since [E1,2 , E2,3 ] = E1,3 it is enough to show that W is invariant under the
successive application of (E1,2 ) and (E2,3 ). We prove this by induction over
the number of factors (E1,2 ) and (E2,3 ) applied to the highest weight vector v.
We start with only one factor: First, we show that the vector (E2,1 )(v) stays
in W after the application of (E1,2 ) and (E2,3 ). We begin with the action of
(E1,2 ):
(E1,2 ) (E2,1 )(v) = [(E1,2 ), (E2,1 )](v) + (E2,1 ) (E1,2 )(v)
= [(E1,2 ), (E2,1 )](v) + 0
= ([(E1,2 ), (E2,1 )]) v W.
We have used that E1,2 (v) = 0 (by definition of v) and that [(E1,2 ), (E2,1 )] h.
On the other hand: the action of (E2,3 ):
(E2,3 ) (E2,1 )(v) = [(E2,3 ), (E2,1 )](v) + (E2,1 ) (E2,3 )(v)
= 0.
We have used that E2,3 (v) = 0 (by definition of v) and that [(E2,3 ), (E2,1 )] = 0.
An analogue statement proves that (E3,2 )(v) is invariant under the action of
(E1,2 ) and (E2,3 ).
For the remainder of this proof we define wn (v) to be the image of v under an
action which consists of a sequence of factors (E2,1 ) and (E3,2 ) with length n
or less. Additionally, we S
define Wn to be the vector space spanned by all possible
vectors wn . Thus, W = Wn .
Claim: (E1,2 ) and (E2,3 ) map Wn into Wn1 .
Proof of the claim: Each wn is equal to (E2,1 )(wn1 ) or (E3,2 )(wn1 ).
Induction-step: Assume that (E1,2 ) and (E2,3 ) map Wn1 into Wn2 . Note
that wn1 is a weight vector with some weight .
Case 1 : wn = (E2,1 )(wn1 ):
(E1,2 )(wn (v)) =
=
=
=
(E1,2 ) (E2,1 )(wn1 (v))

[(E1,2 ), (E2,1 )](wn1 (v)) + (E2,1 ) (E1,2 )(wn1 (v))
([(E1,2 ), (E2,1 )]) wn1 (v) + (E2,1 )(wn2 (v))
Wn1 ,
22
Topology in Physics
Figure 1.3: 120 -sector which contains all weights.

since [(E1,2 ), (E2,1 )] h. On the other hand we have that
(E2,3 )(wn (v)) =
=
=
=
(E2,3 ) (E2,1 )(wn1 (v))

[(E2,3 ), (E2,1 )](wn1 (v)) + (E2,1 ) (E2,3 )(wn1 (v))
0 + (E2,1 )(wn2 (v))
Wn1 ,
since [E2,3 , E2,1 ] = 0.

Case 2 : wn = (E3,2 )(wn1 ): Analogue calculation.
This finishes the proof of the claim.
Since the case n = 1 has been verified above the proof of the original claim is
concluded.
Corollary 4.
1. All the weights of an irreducible representation lie in 120 sector h with vertex (the highest weight). Compare the figure below.
2. dim W = 1. Thus all the other weight spaces (generated as in claim 1)
are at most one-dimensional.
Proposition 6. Let be an arbitrary representation of sl(3, C) with representation space V and highest weight vector v. Let W denote the vector space spanned
by the images of v under the successive applications of the operators (E2,1 ),
(E3,1 ) and (E3,2 ) (which correspond to the roots in R ). Then: W is the
representation space of an irreducible subrepresentation
.
Proof. Let denote the so called highest weight corresponding to the highest
weight vector v. The proof of Claim 1 shows that W is the representation space
23
of a subrepresentation
(i.e., W is closed under the action of sl(3, C)). We have
to show that this subrepresentation is irreducible.
Let us assume the contrary, i.e., let us assume that
is not irreducible. Therefore,
=
0
00 for some representations
0 and
00 with representation spaces W 0
and W 00 (W = W 0 W 00 ). The operations projection to W 0 or W 00 and action
of h have simultaneous eigenspaces, since they commute. The eigenspaces of the
projections are W 0 and W 00 . Therefore, we arrive at the following decomposition
of W :
W = W 0 W 00
= W W1 W2 . . . (decomposition into weight spaces)
= W0 W00 W0 1 W001 . . .
Claim 1 implies that W is one-dimensional. Thus, all the eigenspaces Wi are
at most one-dimensional. We deduce that at least one of the spaces W0 i or W00i
is equal to {0} for each weight space i. Let us assume without loss of generality
that W00 = {0}. Therefore, (successive applications of sl(3, C))(W = W0 )
must reach the whole space W (compare the definition of W ). Since the representations
0 and
00 are not allowed to mix under the successive action of sl(3, C),
this action applied to W0 cant reach any subspace of W 00 .
I.e., W is spanned without any elements in W 00 W W 00 = {0} and W 0 = W .
The representation
is irreducible.
The Appearance of the Weight Diagram
The next natural question we try to answer is: How does a concrete weight
diagram of sl(3, C) look like?
We will proceed as follows: First, we try to determine the explicit form of the
convex hull of the weights and second, we try to determine the interior of this
convex hull.
The convex Hull
Let us have a second look at figure 1.3 to recall that there arent any weights
above the line generated by the dots representing the weight spaces g , g+L2 L1 ,
g+2(L2 L1 ) , . . . and that there arent any weights to the right of the line generated by the dots representing the weight spaces g , g+L3 L2 , g+2(L3 L2 ) , . . . .
We first consider the line which is the upper boundary for the finite set of weights.
Finite dimensionality of V implies that there exists an integer m, such that
g+k(L2 L1 ) = {0} for all k m (compare figure 1.4). To determine m we make the
following important observation: The elements E1,2 , E2,1 and H1,2 [E1,2 , E2,1 ]
span a subalgebra (denoted sL1 L2 ) of sl(3, C) which is isomorphic to sl(2, C).
24
Topology in Physics
Figure 1.4: Upper boundary

Observation 6. Let i 6= j. Then the elements Ei,j , Ej,i and Hi,j := [Ei,j , Ej,i ]
span a subalgebra (denoted sLi Lj ) of sl(3, C) which is isomorphic to sl(2, C).
The subalgebra sL1 L2 shifts the weight spaces V parallel and antiparallel to
the direction L1 L2 . Obviously, the subspace
M
W :=
V+k(L2 L1 )
k
is preserved under the action of sL1 L2 . Thus, W is a representation space of a representation of sL1 L2
= sl(2, C). This representation is irreducible, since there
are no invariant subspaces. We deduce that the eigenvalues of (H12 ) are integervalued and the spectrum of (H12 ) is symmetric about the origin of Z (compare
theorem 4). This implies that the dots representing the spaces V+k(L2 L1 ) are
symmetric to the line {L h | < L, H1,2 >= 0}, where <, > denotes the pairing between space and dual space. I.e., the dots representing the weight spaces
V+k(L2 L1 ) are preserved under reflection in the line {L h | < L, H1,2 >= 0}.
See section 5 for the explicit proof.
Exactly the same consideration applies to the analysis of sL2 L3 . The result is
shown in figure 1.5.
The weight is the highest weight wrt the linear functional l given in equation (1.29). But this specific definition was arbitrary. We could have also defined
l, such that b > a > c. As a consequence, the weight (compare figure 1.5)
would have been the highest weight instead of . Under this circumstances, the
same considerations would have been applicable. Thus, we can update the line
which will enclose all the weights: compare figure 1.6.
Now, we continue to apply these considerations for other definitions of the linear
functional l (totally 6 possibilities), until the line forms a closed boundary.
Figure 1.5: Upper and right boundary
Figure 1.6: Upper, right and left boundary
25
26
Topology in Physics
Figure 1.7: Convex hull

Observation 7. Let be a representation of sl(3, C) with representation space
V and highest weight . Then: the vertices of the convex hull of the weights
are exactly the images of under the reflections in the three lines {L h | <
Hi,j , L >= 0} (compare figure 1.7). Thus, the convex hull forms a hexagon
(if
/ one of the three reflection-lines) or a triangle (if one of the three
reflection-lines).
Claim 2. Every weight is a linear combination of L2 and L3 (or equivalently
of L1 and L3 or L1 and L2 ) with integer coefficients.
Proof. Observe that from theorem 4 we deduce that all the eigenvalues of (Hi,j )
must be integer-valued, and that h = C H1,2 C H1,3 . Consider the arbitrary
weight = 1 L1 + 2 L2 + 3 L3 h together with a vector v V .
(H1,2 )(v) =
=
=
(H1,3 )(v) =
(H2,3 )(v) =
(H1,2 ) v
[1 L1 (H1,2 ) + 2 L2 (H1,2 ) + 3 L3 (H1,2 )] v
(1 2 ) v
(1 3 ) v
(2 3 ) v
(i C)
Thus, (1 2 ) Z, (1 3 ) Z and (2 3 ) Z and therefore 2 =

1 + a, 3 = 1 + b, with a, b Z. We deduce
= 1 L1 + 2 L2 + 3 L3
= 1 (L1 + L2 + L3 ) + a L2 + b L3
= a L2 + b L3 .
(a, b Z)
27
In the last step we have used identity (1.19).

Here we see explicitly that the weights lie in a real subspace in h.
The linear combinations of the Li s with integer coefficients form a lattice W
h which is the so called weight lattice.
In the case of irreducible representations we can state the following proposition
as a consequence of the claim above.
Proposition 7. All weights of an irreducible representation of sl(3, C) are elements of the weight lattice W and are congruent modulo the root lattice R (i.e.,
they differ by an integral linear combination of Li Lj ).
The Interior of the convex Hull
Let us reveal the interior of the convex hull enclosing all weights of a certain
representation.
The subspace
M
V+k(Li Lj )
W :=
of V is a representation space of a representation of sL1 L2

= sl(2, C). According
to theorem 4, there exists an upper and lower bound k1 and k2 of k Z, such
that V+k(Li Lj ) 6= {0} for all k {k1 , k1 + 1, . . . , k2 1, k2 }. If we choose
on the boundary of the convex hull, we recognize that there is a corresponding
weight on the other side of the convex hull (this is meant wrt. an arbitrary
(i, j)-direction). It follows from the main theorem in the discussion of sl(2, C)
that the weights of a representation of sl(2, C) form an unbroken string. The
space in between of these two boundary-weights has to be filled up with other
weights (according to the remark before) such that the weights are congruent
modulo R . We get the complete weight diagram (shown in figure 1.8) if we
repeat this procedure for every boundary-weight and every direction parallel to
the lattice W .
Thus, we have proven the following proposition.
Proposition 8. Let be an arbitrary irreducible representation of sl(3, C). Let
h be an extremal weight (extremal means that |l()| is maximal with l
denoting an arbitrary linear functional as in (1.29)). Then: all the weights lie in
W , are congruent modulo R and are enclosed by the convex hull whose vertices
are the images of under the reflections at the three lines {L h | < Hi,j , L >=
0}.
The Highest Weight Theorem
Assume that we are working with the specific form of the linear functional l
given in (1.29). From the observation in the last subsection about the shape of
28
Topology in Physics
Figure 1.8: Complete weight diagram

the convex hull enclosing all weights (compare observation 7), we know that all
possible highest weights must lie in the 60 -sector shown in figure 1.9. Thus,
Observation 8. Every highest weight must be of the form (note that L2 = L1
L3 )
(a + b)L1 + bL2 = aL1 bL3
for a, b being non-negative integers.
Before we continue, we have to state a basic proposition about products of
representations of Lie algebras and about the dual representation of a Lie algebra.
Proposition 9. Let 1 , 2 and be representations of a Lie algebra g with
representation space V . Then:
(1 2 )(g) = 1 (g) 1 + 1 2 (g) (representation space: V V ),
(1 sym 2 )(g) = 1 (g)sym 1+1sym 2 (g) (representation space: V sym V )
(g) = T (g) (representation space: V ),
for all g g.
This proposition is proven in the appendix.
Corollary 5. Let 1 , 2 and be representations of a Lie algebra g with representation space V . Then:
29
Figure 1.9: Locus of the highest weight vectors

The weights of the representation 1 2 are the pairwise sum of the weights
of the representations 1 and 2 .
The weights of the representation 1 2 are the pairwise sum of distinct
weights of the representations 1 and 2 .
The weights of the representation 1 sym 2 are the pairwise sum of the
weights of the representations 1 and 2 .
The weights of the representation are the negative weights of the representation .
Proof. We confine us to the proof of the second assertion. Let w1 be an eigenvector for 1 (g) with eigenvalue and let w2 be an eigenvector for 2 (g) with
eigenvalue . Then:
(1 2 )(g)(w1 w2 ) =
=
=
=
(1 (g) 1)(w1 w2 ) + (1 2 (g))(w1 w2 )

[1 (g)(w1 )] w2 + w1 [2 (g)(w2 )]
w1 w2 + w1 w2
[ + ] w1 w2 .
Assume w2 = w2 w1 w2 = w1 w1 = 0. Thus, the weights of the representation 1 2 are the pairwise sum of distinct weights of the representations
1 and 2 .
Corollary 6. Let 1 and 2 be two representations of sl(3, C). Let v1 and v2 be
the highest weight vectors with corresponding highest weights 1 and 2 . Then:
v1 v2 is a highest weight vector for 1 2 with highest weight 1 + 2 .
30
Topology in Physics
Figure 1.10: Weight diagram for the standard representation

Let us examine the consequences for some specific representations which will
be useful later on.
Let be the standard representation of sl(3, C) on C3 . Since h is the space
of diagonal matrices given in (1.19), the weight vectors are the standard basis
vectors e1 , e2 and e3 with corresponding weights L1 , L2 and L3 . This yields the
diagram given in figure 1.10.
The diagram for the dual representation is shown in figure 1.11 (compare
the corollary 5).
We immediately deduce the form of the diagrams corresponding to Sym2 (),
Sym2 ( ), Sym3 (), Sym3 ( ), ... (compare figure 1.12, figure 1.13, figure 1.14
and 1.15). We recognize that the symmetric powers Symn () and Symm ( )
yield exactly the triangular diagrams discussed in observation 7. Thus, Symn ()
and Symm ( ) are irreducible, since all the weights have multiplicity 1 (i.e., they
cant be a sum of irreducible diagrams).
Furthermore, Symn () has highest weight n L1 and Symm ( ) has highest
weight (m) L3 .
This enables us to construct representations of sl(3, C) with arbitrary highest
weight aL1 bL3 by tensor products between Symn () and Symm ( ) (compare
observation 8 and corollary 6).
We are ready now to state the following important theorem.
Theorem (Highest Weight Theorem). Let a and b be two non-negative integers. Then there exists a unique irreducible, finite-dimensional representation
a,b of sl(3, C) with highest weight aL1 bL3 .
Proof. We have to prove the existence and the uniqueness of the irreducible
representation a,b .
Figure 1.11: Weight diagram for
Figure 1.12: Weight diagram for Sym2 ()
31
32
Topology in Physics
Figure 1.13: Weight diagram for Sym2 ( )
Figure 1.14: Weight diagram for Sym3 ()
33
Figure 1.15: Weight diagram for Sym3 ( )

Existence: We have just learned that Syma ()Symb ( ) has the desired highest
weight. According to proposition 6, Syma () Symb ( ) contains an irreducible
subrepresentation with the desired highest weight.
Uniqueness: Assume that there would exist another irreducible representation
a,b with the same highest weight. Let v and v denote the highest weight vectors
a,b respectively and with representation spaces V and V respectively.

of a,b and
a,b with highest weight aL1 bL3 .
Then (v, v) is a highest weight vector of a,b
Let a,b a,b be the irreducible representation with a representation space

U generated by the highest weight vector (v, v) (compare proposition 6). The
projections 1 : U V and 2 : U V are non-zero linear maps. Furthermore,
a,b 2 . Therefore, we can apply Schurs lemma
1 = a,b 1 and 2 =
a,b (compare
to get that 1 and 2 are isomorphisms which implies that a,b
=
section 7).
Corollary 7. The irreducible representation a,b is an irreducible subrepresentation of Syma () Symb ( ).
Let us now describe the relation between a,b and Syma () Symb ( ).
Proposition 10. Assume that b a. Then:
Syma () Symb ( ) =
b
M
ai,bi .
i=0
Observation 9. The multiplicities of a,b increase by one on each hexagon of

the weight diagram and are constant on each triangle (without proof ).
To finish this section we state an algorithm to decompose an arbitrary representation of sl(3, C) into irreducible representations:
34
Topology in Physics
1. Determine the decomposition of into weight spaces.
2. Find the highest weight = aL1 bL3 of .
3. Define
such that
a,b . The weight diagram of minus the
=
weight diagram of a,b is the weight diagram of
.
4. Repeat this process for
.
The Analysis of General Simple Lie Algebras
The procedure carried out in the last section to analyze sl(3, C) can be generalized
to an algorithm for the analysis of general complex semisimple Lie algebras. Each
of the following subsections (until the subsection about the Killing form) stands
for one step in the general algorithm. We will omit the general proofs.
Verification of the semisimplicity of the Lie algebra
Semisimplicity is crucial for the following discussion. If the Lie algebra g is not
semisimple, one can restrict the analysis to the semisimple part of g to get the
irreducible representations of g (compare theorem 3).
Determination of the Cartan subalgebra h g
Definition (Cartan Subalgebra). Let g be a complex semisimple Lie algebra.
Then a Cartan subalgebra h g is a complex subalgebra which is maximally
commuting and whose action is diagonalizable in an arbitrary faithful representation.
The dimension of h is the so called rank of g.
Maximally commuting means: if we add only one element in g \ h to h we
loose the commutativity of h.
A representation is called faithful, if its homomorphism is one-to-one. From the
Jordan decomposition theorem in section 2 we deduce that the Cartan subalgebra
is diagonalizable in every faithful representation. Especially, the action of h is
diagonalizable in the adjoint representation.
All elements in h are simultaneously diagonalizable since they commute.
Decomposition of g into a direct sum of root spaces
Compare definition 17 and definition 18 for the terms weight, weight space,
multiplicity, root, root space.
In regard of the adjoint representation:
35
The adjoint action of h on g is diagonalizable. Thus, g decomposes into the

direct sum of its root spaces:
M
g = h(
g )
ad(H)(X) = (H) X,
(1.30)
for all H h and X g . The index in the direct sum ranges over the finite
set R h of all roots. Note that h = g0 .
The calculation (1.25) proves that
ad(g ) : g g+ .
(1.31)
We state without general proof the following properties of the configuration of

the roots in h :
Observation 10.
1. Every root space g is one-dimensional.
2. The rank of the root lattice R h generated by the roots is equal to the
dimension of h .
3. The set R of all roots is symmetric about the origin, i.e. R R.
In regard of a general representation:
Let be an arbitrary representation of a Lie algebra g with representation
space V . Then the representation space V decomposes into a direct sum of weight
spaces V :
M
V =
V
(H)(v) = (H) v,
(1.32)
for all H h and all v V .

The fundamental calculation (1.26) proves that
(g ) : V V+ .
(1.33)
Let us assume that is irreducible. Thus, there must not be any invariant subspace wrt. the action of g. Together with (1.26) we deduce that the weights of
an irreducible representation are congruent modulo the root lattice R .
Like in the discussion of sl(3, C) we can draw the so called weight diagram (compare for example figure 1.8).
36
Topology in Physics
Determination of subalgebras isomorphic to sl(2, C

An important step in the discussion of sl(3, C) was the discovery of subalgebras
of sl(3, C) which were isomorphic to sl(2, C). Define
s := g g [g , g ].
(1.34)
Then: s is a subalgebra of g which is isomorphic to sl(2, C) (without proof). Note

that g , g are one-dimensional and that [g , g ] is at most one-dimensional
(compare observation 10).
We can chose a basis X g , Y g and H [g , g ] satisfying the commutation relations of sl(2, C). This determines H (it must take the eigenvalues
2 and (-2) on g and g respectively, i.e., (H ) = 2).
The weight lattice W
In the discussion of sl(2, C) we learned that the eigenvalues of H are integervalued in any representation. In the present case of s this means that the
eigenvalues of the action H are integer-valued in any representation of s .
In particular, the eigenvalues of the action of all elements H h found in
subsection 5 must be integer-valued in any representation of g.
Therefore, every weight h of every representation of g must yield integers
when evaluated on any of the H s. The weights with this property form a lattice
in h called the weight lattice W of g.
Note: First, all weights of all representations of g lie in W and second, R
W R W .
Symmetry under the Weyl group
The pairing < H , > (H ) between elements in h and elements in h can be
considered as a scalar product between , h .
We define the reflections W () in the plane := { h | < H , >= 0}
parallel to :
< H , >
W () := 2
< H , >
(H )
= 2
(H )
= (H ) .
(1.35)
(Recall from subsection 5 that (H ) = 2.)
Definition (Weyl Group). Let be a weight of an arbitrary representation of
g. Then: all the reflections
W () := (H )
generate a group of reflections in h called the Weyl group W.
37
Let us break up the set of weights into equivalence classes

[] = {weights | = + n , n Z}.
Similarly,
V[] :=
V+n .
(1.36)
(1.37)
Each of these spaces V[] realizes a representation of the subalgebra s g when

we restrict the original representation space V to V[] , since V[] is closed under
the action of s (compare (1.33)).
Proposition 11. The set of weights of any representation of g is invariant under
the Weyl group W.
Proof. It is possible to explicitly write down the whole string of weights that
correspond to nonzero summands in (1.37)...
When we shift the representant of the equivalence class to the boundary of
the string and write down the whole string as
+ , + 2, . . . , + m
,
for an integer m 0. Then we arrive at the following string of integers
), ( + )(H ) = (H
) + 2, . . . , ( + m)(H ) = (H
) + 2m.
(H
According to theorem 4, this string of integers has to be symmetric about the
origin of Z, i.e.,
) = ((H
) + 2m) (H
) = m Z.
(H
(1.38)
Thus, the whole string of weights corresponding to nonzero summands in (1.37)
is
+ , + 2, . . . , (H
).
,
(1.39)
In particular,
W ( + k ) + k ( + k)(H ) H
= + (m k) .
The same consideration shows that the multiplicities are symmetric wrt. the
action of the Weyl group. This implies that the set of weights is invariant under
the Weyl group.
At this point we can introduce the so called Killing form on g. The Killing
form on g is an inner product on h with the characteristic property that it is
the unique inner product on h which is preserved under the action of the Weyl
group W (i.e., the Weyl group acts as a group of orthogonal transformations).
Equivalently, we could say that the Killing form is the unique inner product for
which the line in h spanned by is perpendicular to the plane . Thus, the
Weyl group is the group of transformations generated by the reflections in the
planes perpendicular to the root .
38
Topology in Physics
Ordering of the roots; the highest weight vector

As in the discussion of sl(3, C) we define a linear functional l, such that its kernel
doesnt intersect the root lattice R . This yields a decomposition
R = R+ R ,
where R+ := {roots | l() > 0} and R := {roots | l() < 0}.
Definition 20. Let be an arbitrary representation of g with representation
space V . Then a vector v V is called a highest weight vector if v is simultaneously a weight vector and gets killed by the action of (g ) for all R+ . The
weight corresponding to v is called the highest weight.
Proposition 12. Let g be a complex semisimple Lie algebra.
1. There exists a highest weight vector for every finite-dimensional representation of g.
2. Let be a finite-dimensional representation of g with representation space
V and let W V denote the vector space generated by the images of the
highest weight vector v under successive application of (g ) for all R .
Then: the restriction of the original representation to the representation
space W V is an irreducible subrepresentation of g.
3. The highest weight vector of an irreducible representation is unique (up to
multiplication with scalars).
A root is called primitive of simple, if it cant be written as a sum of two
positive (respectively negative) roots.
Let us go over to the appearance of weight diagrams. As in the case of sl(3, C)
we make the following observation:
Observation 11. The weights of a representation are
1. congruent to the highest weight modulo the root lattice R and
2. they lie in the convex hull with vertices being the images of under the
Weyl group.
From what we have learned in (1.39) we deduce that the highest weight of a
representation is a weight satisfying (H ) 0 for all R+ .
Definition 21. The set of points in the real span of the roots which satisfy the
inequalities (H ) 0 for every R+ is called the Weyl chamber W associated
to the specific ordering of the roots.
In terms of the Killing form, this is the set of points making an angle between 0
and 90 degrees with each positive root.
39
Classification of the irreducible representations

We arrive at the fundamental theorem which we wont prove in general but it
should be evident after what we have done so far.
Theorem (Highest Weight Theorem). Let h be a point in the intersection between the Weyl chamber W (which is associated to a specific ordering of
the roots) and the weight lattice W .
Then: there exists a unique irreducible, finite-dimensional representation of g
with highest weight .
The weights of are the elements of W congruent to modulo the root lattice
R which are enclosed by the convex hull generated by the images of under the
action of the Weyl group.
There always exist so called fundamental weights 1 , 2 , . . . n , such that
every highest weight can be expressed as a linear combination of the fundamental
weights with non-negative, integer-valued coefficients. (Geometrically these are
the weights on the edges of the Weyl chamber.) One usually writes a1 , ... , an
instead of if = a1 1 + . . . + an n .
Before we finish this section, we should say a few words about the Killing form
which we introduced in section 5.
The Killing Form
There are several different ways to define the Killing form. We mention two
possibilities.
Definition (Killing Form (version 1)). The Killing form (, ) on g is the
symmetric bilinear form (hence an inner product) on g defined by
(X, Y ) := tr(ad(X) ad(Y ) : g g),
for all X and Y in g. This induces inner products on h and h respectively which
are denoted in the same way.
Definition (Killing Form (version 2)). The Killing form (, ) on g is the symmetric bilinear form (hence an inner product) on g defined by the characteristic
property that the Killing form is invariant under the group of automorphisms on
g.
Note that the decomposition
M
g=h(
g )
R
is orthogonal according to the first definition, since ad(g ) : g g+ .

One reason why we introduce the Killing form is the following proposition:
40
Topology in Physics
Proposition 13. The line in h spanned by a root is perpendicular (wrt. the

Killing form (, )) to the hyperplane .
From this proposition one can deduce that the Weyl chamber is the set of
vectors in h which form an acute angle (wrt. (, )) with all positive roots.
Now we state an explicit isomorphism between h and h which is determined
by the Killing form.
Proposition 14. The isomorphism between h and h, which is determined by the
Killing form, maps an element to
T :=
2
H .
(H , H )
Conversely, the inverse isomorphism maps

H
2
.
(, )
Using this notation, we define the induced Killing form on h :

(, ) := (T , T ).
(1.40)
From the orthogonality of the action of the Weyl group and (1.38) one can deduce
the next proposition.
Proposition 15. Let and be two roots. Then:
2
(, )
= (H )
(, )
is an integer.
The Classification of Complex Semisimple Lie Algebras
In the first part of this section, we introduce and classify all Dynkin diagrams
whereas in the second part, we try to recover the Lie algebra from its corresponding Dynkin diagram.
Dynkin Diagrams and their Classification
As we have already indicated, the roots span a real subspace of the Cartan
subalgebra h . On this space the Killing form is positive definite (without proof).
This real vector space together with the Killing form (, ) yields a Euclidean space
E. A positive root is called simple if it is not the sum of two other positive roots.
We list the central properties of a root system:
41
Proposition 16. Let R be a root system. Then:

1. The root system R is a finite set and its elements span the Euclidean space
E.
2. R R.
3. k
/ R for all k Z \ {1}
4. The root system R is invariant under the action of the Weyl group W.
5. The quantity
n := 2
(, )
(, )
is an integer for every , R.

6. Let and be two roots such that 6= . Then: the -string through
(i.e., p, . . . , , . . . , + q) consists of at most four elements. Additionally, p q = n .
7. Let and be two roots such that 6= . Then:
(, ) > 0 is a root,
(, ) < 0 + is a root.
If (, ) = 0 , then and + are both roots or both arent roots.
8. The angle between distinct simple roots cannot be acute.
9. The simple roots are linearly independent.
10. The number of simple roots is equal to n = dimC h = dimR E.
11. Every positive root can be written uniquely as a non-negative integral combination of simple roots.
The integer n is called the rank of the Lie algebra or the root system.
The fifth property implies a strong restriction on the possible geometric configurations of the roots:
Let be the angle (wrt. the Killing form) between two arbitrary roots and .
We deduce that
kk
(, )
= 2 cos()
.
(1.41)
n = 2
(, )
kk
Thus,
n n = 4 cos2 ()
(1.42)
42
Topology in Physics
kk
kk
/6 /4 /3
3
2
1
/2
*
2/3 3/4 5/6
1
2
3
Table 1.1: The geometry of the root system

is an integer between 0 and 4. All possible configurations are shown in the
following table.
We list all possible root systems of rank 1 and 2.
Rank 1. The only possible root system is the root system corresponding to
sl(2, C):
Rank 2. The root system for sl(2, C) sl(2, C)

= so4 (C) is A1 A1 . The root
system for sl(3, C) is A2 . The root system for so5 (C)
= sp4 (C) is B2 . The root
system for the exceptional Lie algebra g2 is G2 .
Definition 22. The direct sum of two root systems is a root system. A root
system is called irreducible if it is not a direct sum of two other root systems.
43
Dynkin Diagrams. The Dynkin diagram for a certain root system is a diagram
consisting of nodes for each simple root and lines joining them. The number
of lines between the nodes is defined by the angle between the specific simple
roots. When there is more than one line joining two nodes then the corresponding
simple roots can have different length. In this case one includes an arrow into
the diagram pointing from the longer to the shorter simple root.
Note that a Dynkin diagram contains the knowledge of (i , j ) for all i 6= j.

Two Dynkin diagrams are isomorphic if there exists a one-to-one onto map
of the nodes such that the number of lines between the nodes and such that the
direction of the arrows are preserved. The root diagrams are independent of the
specific ordering of the roots.
Now we state the crucial classification theorem which classifies every irreducible root system. Its proof is completely based on Euclidean geometry (compare table 1.1).
Theorem (Classification of Dynkin Diagrams). The Dynkin diagrams given
in figure 1.16 are all possible Dynkin diagrams of irreducible root systems. With
the correspondences
(An )
(Bn )
(Cn )
(Dn )
sln+1 (C)
so2n+1 (C)
sp2n (C)
sl2n (C)
To complete the classification of complex simple Lie algebras, we would have

to prove the following theorem:
Theorem 5.
1. Let R1 and R2 be two root systems for two different Cartan
subalgebras of the same complex semisimple Lie algebra. Then: R1 and R2
are isomorphic.
44
Topology in Physics
Figure 1.16: All irreducible Dynkin diagrams.

2. A complex semisimple Lie algebra is simple iff its root system is irreducible.
3. Complex semisimple Lie algebras which have isomorphic root systems are
isomorphic.
4. Every root system arises as the root system of a complex semisimple Lie
algebra.
A semisimple Lie algebra is determined (up to isomorphisms) by the specification of its simple summands.
Reconstruction of a Lie Algebra from its Dynkin Diagram
This section consists of two parts: first, we learn how to get the root system of a
given Dynkin diagram and second, we learn how to recover the Lie algebra from
its root system. Therefore, we will be able to reconstruct the Lie algebra from
its Dynkin diagram.
45
Reconstruction of the root system from its Dynkin Diagram

The simple roots follow directly from the diagram; we denote them 1 , . . . , n .
According to 11th property of proposition 16, all positive roots can be written as
an integral linear combination with non-negative coefficients. In the following, we
will
Pn try to answer the question: Which non-negative integral linear combinations
i=0 mi i are roots? The answer to this question determines all the positive
roots andPtherefore the entire root system.
P
Let = ni=0 mi i be any positive root. Then: ni=0 mi is the so called level of
.
Claim 3.
1. Every positive root can be written in the form

= + j ,
where is a root or 0 and j is a simple root.

P
2. Let = ni=0 mi i be a positive root and let j be a simple root. Then:
+ is a root iff
n
p > nj
X
(, j )
=
mi ni j ,
2
(j , j )
i=0
where p is the lower bound for the j -string, i.e.,

p, . . . , , . . . , + q.
Furthermore, p mj .
P
Proof. Let = ni=1 ri i be a positive root with level m + 1. Since the Killing
form is positive definite on the root system, we have
X
0 < (, ) =
ri (, i ).
Thus, (, i ) > 0 for some i with ri > 0 (if all quantities (, i ) and ri were
negative, level m + 1 of wouldnt be possible). Using the 7th property of
proposition 16, we deduce that i is a root. Therefore, was of the form
+ j . This proves the first statement.
Assume we would know all positive roots with level m and let be any positive
root with level m. Question: Is + j a root?
If + j is a root it has to be contained in the j -string
pj , . . . , , . . . , + qj
through , i.e., we have to demand qj string > 0. This is equivalent to p > nj
(compare the 6th property of proposition 16).
property 11 of proposition 16 says that no root can be written as a linear combination with mixed signs. is a positive root by assumption. Therefore, p
is a positive root p mj
46
Topology in Physics
Reconstruction of the simple Lie Algebra from its root system
The process of the reconstruction of the simple Lie Algebra from its root
system can be divided into three steps:
1. Define a basis of the Lie algebra.
2. Compute the entire multiplication table of all elements in the basis.
3. Check that the structure we have gotten (i.e., the vector space spanned by
the basis together with the multiplication table) really defines a simple Lie
algebra.
What one would have to prove is the following
1. The multiplication table is determined by the Dynkin diagram.
2. Existence: There exists a simple Lie algebra for each Dynkin diagram.
3. Uniqueness: The simple Lie algebra corresponding to a certain Dynkin
diagram is unique up to isomorphisms.
Now, we describe, how one can find a basis of g.
Let 1 , . . . , n denote the simple roots which follow directly from the root
system. These simple roots realize a basis in h . After the discussion in the last
subsection we know how to get the entire root system from the set of simple
roots.
Let R+ be an arbitrary positive root. The goal is to find a basis for the
Cartan subalgebra h and every root space g (which are one-dimensional), since
M
g=h(
g ).
We need two ingredients to construct the basis. The first one is the following
lemma.
Lemma 6. Let , 6= and + be roots. Then:
[g , g ] = g+ .
Proof. We have already seen that [g , g ] g+ and that all root spaces are
one-dimensional. What is left to show is that [g , g ] 6= 0.
Consider the -string
M
g+k .
47
If 6= the -string realizes an irreducible representation of sl(2, C), since all

the root spaces in the string are one-dimensional.
Assume that [g , g ] = 0. Then: the following two parts of the string
M
M
g+n
gl
n2
l0
wouldnt mix under the action of ad(g ), i.e., they would realize subrepresentations which is a contradiction to the observation of irreducibility above.
The second ingredient is the observation we made in the last subsection that
we can write the arbitrary positive root in the form (compare claim 3)
= i1 + . . . + ir ,
(1.43)
i1 + . . . + is
(1.44)
such that
is a root for every s {1, . . . , r}.
From these two ingredients we deduce the following:
Once we have found the basis elements for all root spaces corresponding to simple roots and for gi1 + ... +ir1 , the basis element for g gi1 + ... +ir follows
directly by the application of ad(basis of gir ), since
ad(gir ) : gi1 +
... +ir1
gi1 +
... +ir
g .
Thus, to begin, we have to determine the basis for h and for all the simple root
spaces gi :
Choose Xi gi and Yi gi arbitrarily. Define Hi := [Xi , Yi ]. Adjust Yi , such
that
[Hi , Xi ] = 2 Xi and [Hi , Yi ] = 2 Yi .
Therefore, span{Hi , Xi , Yi } = si
= sl(2, C).
We repeat this procedure for every simple root i . We finally get a basis for gi
and gi for every 1 i n and a basis for h (since dim h = dim h ). Then we
can complete the basis of g as described above.
Appendix
Products of representations and dual representations

Definition 23. Let G be a Lie group and let 1 and 2 be representations of G
on vector spaces V1 and V2 . Then:
1 2 (g) := 1 (g) 2 (g)
1 2 (g) := 1 (g) 2 (g)
48
Topology in Physics
1 sym 2 (g) := 1 (g) sym 2 (g)
1 (g) := T (g 1 )
for all g G.
Proposition 17. Let 1 , 2 and be representations of a Lie algebra g with
representation space V . Then:
(1 sym 2 )(g) = 1 (g)sym 1+1sym 2 (g) (representation space: V sym V )
(g) = T (g) (representation space: V ),
for all g g.
Proof. We only prove the second assertion, because the proofs of the others follow
in an analogue way.
Let u(t) and v(t) be two general curves in a vector space V . Then:
d
u(t + h) v(t + h) u(t) v(t)
(u(t) v(t)) = lim
h0
dt
h
u(t + h) v(t + h) u(t + h) v(t) u(t + h) v(t) u(t) v(t)
= lim
+
h0
h
h
v(t + h) v(t) u(t + h) u(t)
= lim u(t + h)
+
v(t)
h0
h
h
d
d
= u(t) v(t) + u(t) v(t).
dt
dt
Let g be the Lie algebra of G, let X g and let u, v V . Then:
1 2 (X)(u v)
d
|t=0 (1 2 )(etX )(u v)
dt
d
|t=0 (1 (etX ) 2 (etX ))(u v)
dt
d
d
= ( |t=0 1 (etX )) 2 (etX )(u v) + 1 (etX ) ( |t=0 2 (etX ))(u v)
dt
dt
= (1 (X) 1)(u v) + (1 2 (X))(u v).
=
49
Schurs Lemma
Theorem (Schurs Lemma).
1. Let and 0 be two irreducible representations of a group G with representation spaces V and V 0 . Assume that
L (g) = 0 (g) L for a linear map L : V V 0 and all g G.
Then: either L 0 or L is a vector space isomorphism.
2. Let be a finite-dimensional irreducible representation of a group G with
representation space V , let L be a linear map L : V V such that L(g) =
(g) L for all g G.
Then: L = 1 on V with C.
3. Let be a finite-dimensional irreducible and Abelian representation of a
group G with representation space V .
Then: V is one-dimensional.
50
Topology in Physics
2 Goldstone bosons and the

Higgs mechanism
Stefan Pfenninger
Supervisor: Stefan Fredenhagen
Starting with the fact that the Lagrangian of electrodynamics is
locally invariant under the transformation group U (1), i.e.
exp(i(x)), A A +ie (x), the first part of this report will
consider the opposite question of how possible Lagrangians that
have such an invariance might look like in general. Astonishingly,
we get quite similar results by replacing the Abelian group U (1)
with a general non-Abelian group G. Using these experiences,
in the second part we will consider the concept of spontaneous
symmetry breaking again in two parts. Firstly we will look at
the case of a global symmetry to be broken what leads us to
Goldstones theorem and secondly we consider the case of a local
symmetry breaking, what leads us to the Higgs mechanism. This
important concept appears in many real physical problems and
we discuss briefly this appearance in the electroweak theory and
in superconductivity. Write here a short summary of your talk. It
should contain all important informations and catch the readers
attention.
51
52
Topology in Physics
Non-Abelian gauge invariance
Motivation and conventions

Considering the Lorentz invariant Lagrangian density for scalar electrodynamics1 ,
one can see that it is locally gauge invariant under the group U (1), whereby
local gauge invariance means that the transformation can depend on the spacetime point x, and global invariance would mean invariance under a constant
transformation.
In the first section, we will derive the rather astonishing result that one can also
derive the Lagrangian density for scalar electrodynamics by only assuming this
local gauge invariance under U (1). This leads us to the idea that we might create
other interesting theories by using more general groups. So we will first have a
look at the group SU (2) and then generalize the methods to a general symmetry
group.
We will use natural units (~ = 1 = c) and Einsteins sum convention in which
Greek indices run from (0, .., 3) and roman indices from (1, .., 3). In this section,
we follow [3] and [4].
Abelian gauge invariance - U(1) gauge theory
Starting with the complex-valued scalar field , we assume that our theory has
to be invariant under the transformation
(x) V (x)(x)
V (x) := ei(x) U (1).
(2.1)
The phase rotation (x) is an arbitrary real function, depending on the point of
space-time, since we want to discuss a local gauge invariance. If we want to find
out what kind of Lagrangian matches our requirement of local U (1) invariance,
we simply consider the different transformation behaviors of the possible terms
that a Lagrangian can have. Of course, if we only take U (1) invariant terms to
build a Lagrangian, we can ensure its local U (1) invariance.
First, we consider a mass term
m (x).
(2.2)
We see that such terms are invariant under both global and local U (1) transformation and do not need further considerations.
1
L(, D )
:=
1
1
F F + (D ) D V ( ) where
4
2
+ iA and A the Lorentz covariant potential.
Goldstone bosons and the Higgs mechanism
53
It becomes more interesting when we have a look at terms including derivatives

of . While the behavior under a global transformation is
(x) V (x),
the one under a local transformation is
(x) V (x) (x) + (x) V (x).
The reason of the appearance of a second term becomes more clear if we have a
look at the definition of the derivative in a given direction n,
1
n (x) := lim [(x + n) (x)].
0
(2.3)
We can see that, since the transformation (2.1) depends on the point in spacetime, both summands have a completely different transformation under (2.1).
One idea to solve this problem is to introduce a factor U (y, x) called comparator
that compensates the difference in phase transformations from one point to the
next with the transformation law
U (y, x) V (y)U (y, x)V (x)1 .
(2.4)
This ensures that (x) and U (x, y)(y) transform in the same way. Of course,
the comparator has to ensure U (x, x) = 1. We can now define an operator D
called covariant derivative:
1
n D (x) := lim [(x + n) U (x + n, x)(x)].
0
(2.5)
Using the expansion for the two points to be close together, we get
U (x + n, x) = 1 ien A (x) + O(2 ),
(2.6)
in which the A is a new real vector field called gauge field or connection. Thus,
the covariant derivative takes the form
D (x) = (x) + ieA (x)(x),
(2.7)
and by inserting(2.6) into (2.4) and comparing the -coefficients:

U (x + n, x) ei(x+n) (1 ien A (x) + O(2 ))ei(x)
= ei(x)+i (x)n +O( ) ei(x) (1 ien A (x) + O(2 ))

= (1 + i (x)n )(1 ien (A) (x)) + O(2 )
= 1 ien A + i (x)n + O(2 )
1
A (x) A (x) (x).
(2.8)
e
54
Topology in Physics
x + 2
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
x + 1
Figure 2.1: square in the (1,2)-plane of space-time, used for the construction of
a field strength
If we check the behavior of D under the local symmetry transformation (2.1) we

find
D (x) e(i(x)) (x) + (x) e(i(x)) + (ieA (x) i (x))e(i(x)) (x)
= e(i(x)) [ + ieA ](x),
hence
D (x) ei(x) D (x).
(2.9)
Note that the definition of the covariant derivative as well as the existence and
the transformation law of the gauge field A are a direct consequence of the
condition of the local symmetry. Without these constructs we would not be able
to write an invariant Lagrangian.
To complete the construction, we also need to find locally invariant terms
that depend on A and its derivatives, but not on (x) itself. Once again, we
consider the comparator U (y, x). To simplify we make some assumptions about
U (y, x); it should be a pure phase U (y, x) = ei(y,x) with (y, x) an arbitrary real
function, and U (x, y) = U (y, x). For small, this leads to an approximation 2
A (x+ n)+O(3 )
U (x + n, x) = eien
(2.10)
Now consider a small square lying in the (1,2)-plane of the space-time with
unit vectors 1 and 2 (see Fig. 2.1). Define U(x) as the product of four U (y, x) functions so that the new function U(x) is invariant under (2.1):
U(x) := U (x, x+2)U (x+2, x+1+2)U (x+1+2, x+1)U (x+1, x). (2.11)
2
ie
d
x A (
x)
Pxy
Consider the equation U (y, x) = e
which accomplish all the conditions
on a comparator and this for any line Pxy (Wilson line) running from x to y. Now
R x+n
3
A (
x)d
xn = n A (x + n
2 ) + O( ) can be proved by integrating the well known
x

equation f (x + ) = f (x + 2 ) + 2 f (x + 2 ) + O(2 ) over using the boundaries [0, ].
55
Because U(x) is invariant under (2.1) even for 0, and by using (2.10) we get
a locally invariant function of A :

U(x) = exp(ie[A2 (x + 2)
2

A1 (x + 1 + 2) + A2 (x + 1 +
2
and expanded:

2) + A1 (x +
2

1)] + O(3 )),
2
U(x) = 1 ie2 [1 A2 (x) 2 A1 (x)] + O(3 ).
(2.12)
Since U(x) is locally invariant, F := A (x) A (x) is so as well. F is

called field strength tensor. If we think of the Lagrangian of electrodynamics, we
can see that F corresponds to the electromagnetic field tensor.
A shorter approach of finding the invariance of F starts with the covariant derivative: since D (x) V (x)D (x) : [D , D ](x) V (x)[D , D ](x),
what implies that [D , D ] is invariant under (2.1) 3 . A closer look at the commutator [D , D ] shows that it is not a derivative at all:
[D , D ](x) = [ , ](x) + ie([ , A ] [ , A ])(x)
e2 [A , A ](x)
= ie( A (x) A (x))(x),
and that with the definition of F we can write
[D , D ] = ieF
(2.13)
and that it is thus invariant.

We can now create a Lagrangian out of , D and of F and its derivatives.
When we assume that operators with dimension not larger than 4 make sure that
the theory keeps renormalizable, only 5 terms can appear:
1
L = (D ) D (F )2 ic F F m + a( )2 ,
4
(2.14)
where the 14 in the second term is just conventional. The U(1) group, which
was a global symmetry has become a local symmetry group and is now called
gauged.
Note that we achieved this result only by assuming a local symmetry; furthermore, the existence of an electromagnetic vector potential appears naturally. If
we insist on having a Lagrangian L that is also invariant under time reversal or
parity, the third term disappears and our construction remind us on the Lorentz
invariant Lagrangian density for scalar electrodynamics.
3
this argument is only true when V(x) and [D , D ] do commute. In the U (1) case this is
of course the case since V(x) is a scalar function. In the next section though, we will see that
this invariance is no more true.
56
Topology in Physics
Non-Abelian Gauge Theory and the Yang-Mills Lagrangian

In this section, we want to generalize the methods of the previous section in
which we assumed the symmetry group to be the Abelian Group U (1). We now
want to consider a general group G as a symmetry group which is in particular
non-Abelian. Before we start to generalize the results from the previous section,
we first write down some conditions on the general group and very briefly some
facts about Lie algebras.
Mathematical concepts As we have seen in the last section, our gauge theory
is related to a symmetry group, which we had to gauge. In order to use similar
methods like we did in the Abelian case, the symmetry group G must be at
least a continuous group. We actually focus our attention only on continuously
generated groups, which means that the group contains elements arbitrarily close
to the identity. This condition allows us to write an infinitesimal group element
g in the form of
g() = 1 + ia X a + O(2 )
X a Lie(G).
(2.15)
The set of generators X a span a vector space4 , and so the commutator of generators X a is a linear combination of generators,
[X a , X b ] = ifabc X c
(2.16)
in which the fabc are called structure constants.

We can choose X a to be a set of hermitian generators 5 and because of the Jacobi
identity of the X a , [X a , [X b , X c ]] + [X b , [X c , X a ]] + [X c , [X a , X b ]] = 0 we get
fade fbcd + fbde fcad + fcde fabd = 0,
(2.17)
where it can be proved that fabc are totally antisymmetric.

Another nice concept we will use is the irreducible representation ad of a
Group G, called the adjoint representation whose definition is
G Aut(V )
g 7 (v 7 gvg 1 ).
(2.18)
The representation matrices of the associated representation of the Lie algebra

of G are given by the structure constants,
( (X b ))ac = ifabc
4
X b Lie(G).
(2.19)
Note that the definition of a generator A of a group element a is here the equation a = eiA
and not eA .
5
We know that there is a scalar product in the representational vector space for every
representation of the group G such that the representational matrix is unitary. According to
this fact, the generators of G can be represented by hermitian matrices X a .
57
Methods of generalization illustrated by the group SU(2) We are now

ready to generalize the methods to a general symmetry group G. Because SU (2)
is a well known example for such a group, we will apply the results to this group
to illustrate the effect more concretely.
We start now with the more general situation using a set of N fields instead
of a single field:
1 (x)
..
(2.20)
(x) =
.
N (x)
For the SU (2) case, it is enough to have N = 2,
!
1 (x)
(x) =
.
2 (x)
(2.21)
(x) transforms analogously to the Abelian case,

(x) V (x)(x) , V (x) := exp(ia (x)X a ) G
(2.22)
where X are the hermitian generators of the group G and the index a goes from
1, .., dim(G).
a
For G = SU (2), we have X a = 2 with j , (j = 1, 2, 3) the hermitian Pauli spin
matrices. Expanding V (x) for a small yields
V (x) = 1 + ia (x)X a + O(2 ).
(2.23)
Now we want to gauge the group by changing (2.22) into a local symmetry group.
This can be ensured by insisting on the condition that the Lagrangian should be
invariant under (2.22) for = (x) an arbitrary function of x. We will now try
to use the same methods as in the previous section, in which our symmetry group
was U (1). One fundamental difference is that the generators X a do not have to
commute as in the Abelian case. For G = SU (2) this is the case since the Pauli
matrices do not commute.
In general, the field theory with a non-commuting local symmetry is called a
non-Abelian gauge theory.
Exactly as in the Abelian case, we introduce a comparator U (y, x), but since
is a N component object, U (y, x) has to be a N N matrix. Again we claim
U (x, x) = id and the transformation law to be
U (y, x) V (y)U (y, x)V (x) .
(2.24)
Assuming U (y, x) to be a unitary matrix, we can expand U (y, x) in terms of the

hermitian generators of G, and for a small distance between the two points x and
y, we have
U (x + n, x) = id + ign A a X a + O(2 ).
(2.25)
58
Topology in Physics
The constant g is extracted and we will use it later. As one can see, the new
fields A are elements of the Lie algebra Lie(G),
A = A a X a
(2.26)
and the A a are called Yang-Mills fields. For G = SU (2) we get analogously
j
A = A j 2 . If we write the new expansion (2.25) in the old definition (2.5),
then we get a new expression for the covariant derivative:
D = igA a X a
(2.27)
Now we want to find the gauge transformation law of the Yang-Mills fields A .
Inserting (2.25) in (2.24):
1 + ign A a X a V (x + n)(1 + ign A a X a )V (x)
(2.28)
Using V (x + n) = (1 + n )V (x) + O( ) and V (x) = V (x) one obtains

by comparing the terms proportional to ign :
i
A a (x)X a V (x)[A a (x)X a + ]V (x).
g
The infinitesimal transformation laws for and A a are:
(1 + ia X a )
and
1
A a (x) A a (x) + a + fabc A b (x)c
g
(2.29)
(2.30)
In order to check that the new covariant derivative transforms as

D (x) V (x)D (x)
(2.31)
one might use (2.29) and show it in a straight forward way. This calculation is
not so easy, because there are lots of matrices that do not commute with each
other. Since our construction (2.5) does not need such a verification, we leave
the calculation to those who do not believe 6 .
Again, we need to find gauge-invariant terms that depend only on A . Inspired by the Abelian case, we consider the commutator of covariant derivatives
and its transformation
[D , D ](x) V (x)[D , D ](x).
(2.32)
Just as in the Abelian case, we can show that [D , D ] is not a differential operator
anymore but only a multiplicative factor in form of a N N matrix which does
not commute with the matrix V (x) anymore:
[D , D ] = igF a X a
6
For G = SU (2), this calculation is done in [3] but only for a small .
(2.33)
59
where
F a X a = A a X a A a X a ig[A a X a , A a X a ].
(2.34)
This expression can be simplified if we remember the commutation relations (2.16)

of the generators X a ,
F a X a = A a X a A a X a + gfabc A b A c X a .
(2.35)
In the SU (2) case, we remember the commutation relations of the Pauli matrices,
k
l
j
[ 2 , 2 ] = ijkl 2 , with the antisymmetric structure constants jkl . As usual, the
equation (2.35) for the SU (2) case looks the same, we only have to replace X a
j
by 2 .
From the equations (2.24) (2.32) and (2.33) we can conclude that the new field
strength tensor transforms according to the adjoint representation
F a X a V (x)F a X a V (x) ,
(2.36)
or in the infinitesimal form

F a X a F a X a + [ia X a , F b X b ]
F a F a fabc b F c
(2.37)
Note that the field strength tensor is no longer gauge-invariant, since the factors
(matrices!) in equation (2.36) do not commute anymore. However, in an appropriate normalization, we can still find gauge-invariant terms out of the field
strength tensor, for example the term
1
tr[(F a X a )2 ] = (F a )2 .
(2.38)
2
Such a term in a Lagrangian would describe a nontrivial, interacting field theory,
because it contains cubic and quartic terms in A a . Such a theory is called YangMills theory.
Using the equations (2.30) and (2.37), it is possible to show that any globally
symmetric function of , F a and their covariant derivatives is also locally
symmetric, and might possibly appear in a Lagrangian.
However, as we have already seen in equation (2.14), in a renormalizable theory,
not so many terms are possible, and under the assumption that our Lagrangian
preserves time reversal and parity, only four terms are allowed:
1
L = D D (F a )2 m + a( )2 .
4
Using the Euler-Lagrange equations
L
L
= 0,
( )
(2.39)
(2.40)
60
Topology in Physics
the corresponding classical equations of motion follow,

c
F a + gfabc Ab F
= gj a
(2.41)
j a = i[(D ) X a X a (D )].
(2.42)
where
Now we want to adjust the covariant derivative on the adjoint representation,
that means if a field (x) does not transform the way equation (2.1) shows, but
in the adjoint representation, we want D acting on (x) to respect that, in the way
that the second term of D should act on the field (x) according to the adjoint
representation and its corresponding Lie algebra representation respectively,
(D )a
a igA b ( (X b ))ac c
a + gfabc A b c ,
(2.43)
where we used equation (2.19). Together with equation (2.43) we can write
equation (2.30) in the new form
1
A a A a + (D )a
g
(2.44)
and equation (2.37) becomes

(D F )a = gj a .
(2.45)
One can see that with this generalization of the covariant derivative - now it
respects the transformation behaviors of the fields it acts on - the strange-looking
terms involving fabc are gone and even for a general symmetry group G we receive
the same results as in the previous sections only by generalizing the covariant
derivative.
To finish this section, we want to show that there is, of course, an analogue
of the homogeneous Maxwell equations in electrodynamics7 . We first consider
the antisymmetric double commutator of covariant derivatives
[D , [D , D ]] = 0.
(2.46)
The term vanishes because of its total antisymmetry8 . Since we know that
[D , D ] = igF a X a , equation (2.46) can be reduced to
(D F )a = 0
(2.47)
which is the equation corresponding to the homogeneous Maxwell equation.

7
Remember that also in the classical electrodynamic theory, the homogeneous Maxwell
equation do not follow from the Euler-Lagrange equations. Because the field strength tensor
F is defined by the Lorentz invariant potential A , the homogeneous Maxwell equation follows
automatically.
8
Since is a total antisymmetric tensor with 1234 = 1, only 3! terms are non-zero when
the first component is fixed. If for example we fix the 4th coefficient,
4123 = 4231 = 4312 = 1 ,
4213 = 4132 = 4321 = 1
, and the rest all zero.
61
Everything in this chapter will be done for classical field theory except for the
particle interpretation out of a Lagrangian density. Our results are therefore not
completely correct, but one might see that what we derive can be interpreted as
the first terms in a quantum expansion. A discussion of quantum corrections is
done in [4].
Spontaneous symmetry breaking
In the context of Lagrangians with scalar potentials, spontaneous symmetry
breakdown is defined as follows:
Given that we have a rigid group G and a potential V () that is group invariant
under this group, i.e.
V (U (g)) = V ()
(2.48)
for U (g) a unitary, continuous representation of G. If the point (x) = 0 where
V () takes its minimum is not G invariant, i.e.
U (g)0 6= 0
for
g G,
(2.49)
then the symmetry is said to be spontaneously broken.

A visual example of the spontaneous breaking of a symmetry is given in [6]:
Imagine an infinite crystalline array of spin- 12 magnetic dipoles that are interacting with their nearest neighbors. Even though the Hamiltonian is rotationally
invariant, the ground state is not because it is a state in which all the dipoles
are aligned in some arbitrary direction and is infinitely degenerate for an infinite
ferromagnet 9 . Now imagine a little (i.e. < ) man living inside this huge
crystal. He would definitely have a hard time to find out that the laws of nature
are rotational invariant. As long as his experiments interact even just a little
bit with the background field, he might detect rotational invariance only as an
approximate symmetry. He would have no reasons to suspect that it was in fact
an exact symmetry. The idea that the little man could check that he lives in
a infinitely degenerated ground state and therefore misses a symmetry is good
but an impossible task for a little man, since he had to change the directions of
infinitely many spins at the time.
Therefore
4 [D , [D , D ]] =
=
1{[D1 , [D2 , D3 ]]4 + [D2 , [D3 , D1 ]]4 + [D3 , [D1 , D2 ]]4 }

+{[D2 , [D1 , D3 ]]4 + [D1 , [D3 , D2 ]]4 + [D3 , [D2 , D1 ]]4 }
0
because of the Jacobi identity.

9
Because the ferromagnet is infinite, there are no special directions for the spins to align.
Every direction is possible and so, there are infinitely many ground states.
62
Topology in Physics
VHL
VHL
Figure 2.2: potential for 2 > 0
Figure 2.3: potential for 2 < 0
Like the little man who lives in the crystal, missing the rotational symmetry, we
assume that our laws of nature may possess symmetries which are not manifest
to us because our vacuum state is not invariant under them. With vacuum state
is meant the state of lowest energy, and we characterize the spectra of small oscillations around the vacuum as particle masses.
This idea that we might miss a symmetry is one of the main motivations for the
further discussion.
Example: spontaneous symmetry breaking of a discrete symmetry
group Consider the set of n real scalar fields (x), with the Lagrange density
1
L = ( )( ) V ().
(2.50)
2
The energy density corresponding to (2.50) is
1
1
H = (0 )2 + ()2 + V ().
2
2
(2.51)
We assume to have a potential

V () =
4 2 2
+ ,
4!
2
(2.52)
where is a positive number and 2 might be positive or negative. We can easily

see the symmetry
.
(2.53)
If 2 is positive, then the potential looks like the one in Fig. 2.4. The vacuum
is then at 0 = 0 and the symmetry is not broken and the mass of the particle
corresponding to the field is .
If 2 is negative (see Fig. 2.5), it is convenient to introduce a quantity
c2 :=
62
(2.54)
63
and the potential can be written as

V () =
2
( c2 )2
4!
(2.55)
plus an irrelevant constant. One can see from equation (2.55) that the potential
has two minima, = c. Which one we choose does not matter, the symmetry
is spontaneously broken anyway. So we choose = c and define a new, shifted
field
0 := c.
(2.56)
In terms of this shifted field, the potential is
V () =
02
c
c2 02
( + 2c0 )2 = 04 + 03 +
4!
4!
6
6
(2.57)
We can see here, that the squared of the mass of the particle that corresponds to
2
2
the field 0 is c3 and not 2 = c6 as one might think by considering equation
(2.50) and (2.52). Of course we have to be careful by interpreting particles out of
a Lagrangian density, because our vacuum state might not be at 0 = 0. In such
a case we need to expand the Lagrangian around 0 because in our definition of
particles we consider the neighborhood of the vacuum state, just the way we did
it here.
Note also that a cubic term appears, which makes it hard to detect the hidden
symmetry
0 (0 + 2c).
(2.58)
So here we can see, that for such a system with spontaneous symmetry breaking,
we might have two Lagrangians (here (2.50) and (2.57)) describing the same
system, but in one we can see easily the symmetry and in one the particles.
Example: spontaneous symmetry breaking of a continuous symmetry
group This time, we choose to be a complex scalar field defined by the
Lagrangian
L0 ((x), (x)) = V ( )
(2.59)
and its classical Hamiltonian
Z
H = d3 x[ + + V ( )]
:=
.
t
(2.60)
L is invariant under the global gauge transformation

(x) ei (x).
(2.61)
One possible vacuum state is the constant (x) = 0 , such that V (0 ) is at

its smallest value. If 0 6= 0, then 0 is not invariant under a global phase
64
Topology in Physics
VH* L
VH* L
*
Figure 2.4: potential for 2 > 0
Figure 2.5: potential for 2 < 0
transformation, hence the symmetry is spontaneously broken. The lowest state

is again infinitely degenerated because the phase of 0 is arbitrary.
We want to fix a potential
V ( ) = 2 + ( )2 + const.
(2.62)
and get the equation of motion

( + 2 ) = 2 2
(2.63)
by using the Euler-Lagrange equations.

If = 0 and 2 > 0, the Lagrangian contains no terms representing interaction
and we get the plane waves of wave vectors k with k 2 = 2 corresponding to a
single-particle state of mass .
If > 0 and 2 > 0 the potential looks like Fig. 2.2 and the lowest solution is
(x) = 0 what corresponds to no spontaneous symmetry breaking.
If > 0 and 2 < 0 the potential looks like Fig. 2.3 and we can already see that
the lowest solution (x) = 0 6= 0 and so the symmetry is spontaneously broken.
We want to study this case now.
The potential is defined only up to a constant, so that we can rewrite it in the
form:
V ( ) = ( 20 )2 + const.
and the lowest state corresponds to =
20 ,
0 6= 0
(2.64)
or
(x) = 0 ei0
(2.65)
with 0 an arbitrary real constant.

We expand (x) around the vacuum solution 0
(x) = (0 + (x))ei(x)
(0) = 0
(2.66)
and the complex fields (x) and (x) are then replaced by real fields (x) and
(x). The Lagrange density changes to
L0 = (20 + )2 2 + (0 + )2 .
(2.67)
65
V (1
0 )
0
1
0
1
massive
mode
0
1
0
1
0
1
()
1
0
0
1
()
massless mode
Figure 2.6: Two possible modes in the scalar field if the potential V implies a
vacuum state that is not invariant under the symmetry U (1).
Close to the vacuum state, is small and we drop terms higher than second order
in 10 :
L0 = [ 420 2 ] + 20 + O( 3 ).
(2.68)
If we look at equation
(2.68), we see that the terms in the brackets describe a
particle of mass 20 . The next term describes a massless scalar particle because the -field appears in the Lagrangian only by its derivatives. These modes
are illustrated in Fig. 2.6.
There exists a nice way in which we could have seen purely geometrically that
there must enter a massless particle term in the Lagrangian. If our vacuum is not
invariant under the symmetry group U (1) we have a curve in an abstract space of
all states passing through all the possible vacuum states. In our case, the curve
is a circle in the complex plane as drawn in Fig. 2.6. If we expand the potential
around the vacuum, no terms involving the variable that measures displacement
along this curve can appear, because the potential is constant along this curve.
Since this curve is parametrized by the function (x), and (x) appears in the
Lagrangian by its derivatives, we always have a massless particle.
Such a massless particle is called a Goldstone boson, what leads us to the next
section.
10
If we drop the second order term in as well, we will loose the information how the system
behaves around the vacuum solution since the Euler-Lagrange equations kill one .
66
Topology in Physics
Goldstones Theorem and Goldstone bosons

In the last example of the previous section, we have seen that the spontaneous
breakdown of a global symmetry implies the existence of a massless particle called
Goldstone boson. One should remember that the symmetry that is spontaneously
broken is still a symmetry of the system. It is manifested not through the invariance of the lowest state, but in the Goldstone mode - through the existence of
a Goldstone boson.
In this section we want to give an idea of the general version of that statement
called Goldstones theorem.
We start with a Lagrangian that is globally invariant under the symmetry
group G and assume to have a set of N real fields, and a potential V () which
is invariant under a group transformation
aXa
(x) ei
(x)
and so
aXa
V () = V (ei
(x)),
(2.69)
where the X a are a set of N hermitian matrices and the a are arbitrary real
parameters.
Now we want to construct the subgroup H of G that contains all the elements
of G that leave the vacuum state 0 of the Lagrangian invariant 11 . H does
of course depend on the potential V () and may be anything from the trivial
identity subgroup (all symmetries spontaneously broken) to the full group (no
symmetries spontaneously broken). We want to choose the indices of the group
generators so that this subgroup H is generated by the first N K generators
where N > N K > 0. Formally, this means
H = {h G|h0 = 0 }
and
X a 0 = 0
a 6 N K.
(2.70)
Now the K remaining generators do not leave 0 invariant, and according to the
discussion at the end of the previous chapter, in the abstract space of all states,
we have a K-dimensional surface of constant potential V (). Thus, by the same
geometrical arguments as before 12 , the theory must contain K massless particles,
one for each spontaneously broken symmetry. As we said before, these particles
are called Goldstone Bosons and the statement we just proved is a special case
(the classical case) of Goldstones theorem, that can be proved in much greater
generality in quantum field theory13 .
11
It is easy to see that this is in fact a subgroup (see also chapter little group).
A more mathematical proof is done in [7], p.93.
13
See [6], p.120:
given a field theory obeying the usual axioms (Lorentz invariance, locality, Hilbert space with
positive-definite inner product, etc.), if there is a local conserved current (the axiomatic version
of the statement that the Lagrangian is invariant under some continuous transformation) such
that the space integral of its time component does not annihilate the vacuum state, then the
theory necessarily contains a massless spinless meson, with the same internal symmetry and
parity properties as the time component of the current.
12
67
Higgs mechanism
In the very beginning of this report, we have seen that there are certain Lagrangians that are invariant not only under a global symmetry transformation,
but also under a local one. From the previous section we know that there are
spontaneous symmetry breakings, but we discussed that phenomenon only in case
of a global gauge invariance. In this section, we will discuss the case of a local
gauge invariance in an Abelian and a non-Abelian example.
Higgs mechanism in an Abelian model Consider the following Lagrangian
with the complex, scalar field , the covariant derivative D and the potential
V ( ):
1
L = F F + (D ) D V ( ),
4
D = ( + ieA )
V ( ) = ( 20 )2 , 0 6= 0.
(2.71)
The gauge field A transforms according to equation (2.8) and the field according to equation (2.1).
Using the Hamiltonian of the Lagrangian (2.71), it is possible to show 14 that a
lowest energy solution is
A (x) = 0
and
(x) = 0 ei0
(2.72)
where 0 is an arbitrary constant, and since 0 6= 0 we see that there is a spontaneous symmetry breaking for all 0 and without loss of generality we choose
0 = 0.
Considering the close neighborhood of the vacuum state 0 , we write (x) =
0 + (x) with (x) := 1 (x) + i2 (x) and get the Lagrangian
1
1
L(A , ) = F F + (D ) D + e2 20 A A + e2 0 1 A A
4
2
2 2
(2.73)
e0 2 ( A ) (40 1 + 40 1 ||2 + ||4 ),
Now we take advantage of the fact that the Lagrangian is invariant under a local
phase transformation and gauge away the phase of such that we can assume
to be real,
(x) = (x)
,
x R4 : (x) R.
(2.74)
That leads us to the Lagrangian
1
L(A , ) = F F + ( )2 + e2 A A 2 (2 20 )2 .
4
14
see [4], p.54
(2.75)
68
Topology in Physics
As in the previous section, we call the local gauge symmetry spontaneously broken
when 0 6= 0. A vacuum solution is = 0 and A = 0. This follows from the
equations of motion (see equation (2.78)). Because 0 6= 0, we get a spontaneous
symmetry breaking to discuss.
Again we consider the field (x) in a close neighborhood of the vacuum state 0 ,
(x) = 0 + (x),
(2.76)
and the Lagrangian changes to the form

1
L(A , ) = F F + ( )2 + e2 A A 20
4
+e2 A A 2 (420 2 + 40 3 + 4 ).
(2.77)
By comparing the two Lagrangians (2.73) and (2.77) we recognize that in (2.73),
there appears a massless particle (Goldstone Boson) according to the field 2 but
there is no such massless particle in (2.77) anymore. Since the second Lagrangian
(2.77) is reached by a local gauge transformation, we can see that the Goldstone
boson can be gauged away.
So the Goldstone Boson is just a so-called gauge phantom and by using the
right gauge transformation, it disappears. Now we want to investigate the form
of the gauge fields A in this gauge, because once we have chosen the gauge
transformation for it is fixed for the fields A as well.
Using the Euler-Lagrange equations, we get the equations of motion
F = 2e2 2 A
( + ieA )( + ieA ) = 2(20 2 ).
(2.78)
Expanding the equations of motion in the regime where and A are small
quantities leads us to the new equations of motion
( + 2e2 20 )A = 0 ,
( + 420 ) = 0
A = 0 and
(2.79)
Here we can see that the massless particle appears no more but the gauge field
A has become massive because of the non-zero coefficient 2e2 20 that corresponds
with the squared value of the mass. The conclusion can be shown in the following
way: at the beginning we can choose a fixed gauge so that the original independent fields are A1 , A2 and , . With the expansion around the vacuum state
and using the gauge which changes the field into a real field , we replace the
original fields by A1 , A2 , A3 and . Now we can see that the Goldstone boson,
which is in fact just a gauge phantom, has been eaten by the gauge field and
that one gets a massive gauge field. This magic trick was discovered by P.Higgs
(and independently by other people) and it is called the Higgs mechanism. Some
more discussion can be found at section 2 after we will have also considered the
Higgs mechanism for a general non-Abelian gauge group.
69
The little group Before we have a look at one example of the Higgs mechanism of a non-Abelian local gauge transformation, we want to discuss briefly a
useful concept called the little group.
We assume that only some spin-0 fields, Higgs fields, can have non-vanishing vacuum values, and therefore cause spontaneous symmetry breaking. The reason is
that higher-spin fields which cause spontaneous symmetry breaking would spontaneously break the Lorentz invariance in contradiction to experimental evidence.
We will use the same Lagrangian (2.71) and write for the Higgs fields. The
potential V () has its minimum at (x) = 0 with the minimum value zero:
V (0 ) = 0,
V 0 (0 ) = 0,
V 00 (0 ) > 0.
(2.80)
As we know, spontaneously symmetry breaking happens if 0 6= 0 and a vacuum

solution is
(x) = 0 and Aa (x) = 0 ( D (x) = 0).
(2.81)
As we have seen earlier in the U (1) case, 0 is not the only possible vacuum state
because ei 0 for an arbitrary would do it as well. The same happens for a
spontaneously symmetry breaking in the non-Abelian gauge theory, and we want
to discuss it the same way as in the Abelian case.
First, we specify that 0 is constant 6= 0, which means that all the components of
0 are independent of space-time points x. Since we assume the potential V to be
independent of x, this is indeed possible. As we have seen in earlier sections, there
might be certain elements h G for that h0 builds another possible vacuum
of G, that leave 0 invariant, such that
state. Consider now the elements h
0 do not differ. Because we assume to have a
the two vacuum states 0 and h
spontaneous symmetry breakdown, this must not hold for all the elements g of
the local gauge transformation Group G. We call the set of the elements that
leave the vacuum state invariant the little group, and say that the symmetry G
is spontaneously broken down to H 15 .
The Lie algebra of the little group is made of a subset of generators {x } of the
generators {Xa } of G. Since
eiw
ax
a
0 = 0
(2.82)
for arbitrary wa , all the generators of the little group must hold the equation
x 0 = 0.
15
(2.83)
For an example take G = SU (2), = (1 , 2 , 3 ) and V () := (21 + 22 + 23 a2 )2 .

We can see the vacuum solution 0 = (01 , 02 , 03 ) with 201 + 202 + 203 = a2 . If we fix
a solution 0 pointing in the direction e1 , it follows that a U 0 = 0 for a rotation U with
axis e1 . Since we can write every such U in the Form U = exp(iwX1 ) we see that SU (2) is
spontaneously broken down to U (1).
70
Topology in Physics
Since the generators of G not in the set {x } cannot annihilate 0 we can divide
the set {Xa } into two disjoint subsets:
{Xa } = {Xj , x } ,
Xj 0 =
6
0 (j = 1, ..., K)
x 0 = 0 ( = 1, .., N K)
(2.84)
For convenience, we choose real antisymmetric generators {Tj } instead of the

hermitian {Xj },
Tj := iXj (j = 1, .., K) and
t := ix ( = 1, .., N K),
(2.85)
and we assume 0 to be a real vector and therefore the representational vector

space to be a real R-dimensional vector space.
We can show that the vectors Tj 0 , (j = 1, .., K) are independent and therefore span a K-dimensional subspace of the R-dimensional representational vector
space. For the little group to be non-empty, we need to satisfy the condition
R K > 0.
(2.86)
We call the K-dimensional space spanned by Tj 0 (j = 1, .., K) the Goldstone

space and its complement the (R K)-dimensional space, the Higgs space.
It can be shown that for any vector in the representational vector space and
for any gauge group G, there is always a gauge transformation U0 G such that
U0 is orthogonal to the Goldstone space:
(Tj 0 , U0 ) = 0 ,
(j = 1, .., K),
(2.87)
P
where we use the standard scalar product (f, g) := R
n=1 fn gn . This U0 is said
16
to be in unitary gauge . Therefore we have
(x) U0 (x)(x)
16
Proof:
For fixed and , consider the mapping
f :G
f : U 7
R
(, U )
Since G is compact, the function f has an extrema, lets say at position U0 . A small variation
of U around U0 therefore gives
f = f (U0 + U ) f (U0 ) = 0.
By writing the element U = a Ta U0 and using the definition of f , we see
0 = f = (, a Ta U0 ) = a (, Ta U0 ) = a (Ta , U0 ).
Since the a are arbitrary, the last scalar product has to be zero and therefore the statement
follows.
U0 (x)(x) =
(x)
71
Goldstone space, K-dimensional
(2.88)
Higgs space, (R K)-dimensional.
Higgs mechanism in a non-Abelian model Using the results of the previous

section, the vacuum solution can be written in the unitary gauge and has then
the form
!
0
, Aa = 0.
(x) = 0 =
(2.89)
0
We expand around this vacuum solution:
!
0
(x) =
0 + (x)
Aa
small.
(2.90)
From equation (2.80) we can expand the potential

1
1
V () =
V (0 ) =
(, V 00 (0 )).
2
2
For the matter field current j of the Lagrangian
1
L = Fa Fa + (D ) (D ) V ()
4
is already known from equation (2.42) and we can also expand it to
(2.91)
(2.92)
17
ja = g 2 (Ta 0 , Tb 0 )Ab .
(2.93)
Together with the following bloc matrices,

Goldstone space
0
0
2
, and
( )rs :=
00
0 V (0 )
Higgs space

Goldstone space
(M 2 )ij 0
2
2
(M )ab := g (Ta 0 , Tb 0 ) =
0
0
Higgs space,
the linearized equations of motion can be written in the form
r + (2 )rs s = 0 ,
Ai + (M 2 )ij Aj = 0 ,
A ( A ) = 0 ,
(2.94)
(2.95)
18
(r = 1, .., R K)
( Ai = 0), (i = 1, .., K)
( = 1, .., N K).
(2.96)
(2.97)
(2.98)
The first equation describes massive particles called Higgs bosons, while the second one describes massive gauge bosons because the gauge fields of the spontaneously broken symmetry have eaten their Goldstone bosons. The third equation
describes massless gauge bosons which are the gauge particles associated with the
unbroken symmetry H.
17
18
see [4], p.78

see [4], p.78
72
Topology in Physics
K broken generators
Higgs field
global symmetry is spontaneously broken

(R K) remaining dimensions
(R K) remaining dimensions
Higgs field
Higgs bosons
Higgs bosons
local symmetry is spontaneously broken

K broken generators
Gauge field
Goldstone bosons
massive gauge bosons
local symmetry is spontaneously broken

(N K) unbroken generators of the little group H
massless gauge bosons
Figure 2.7: Scheme of Goldstones theorem and of the Higgs mechanism
Conclusions
So far we have done a lot of very theoretical work that is schematically shown
in Fig. 2.7. In the upper part, the spontaneous symmetry breaking of a global
symmetry is considered. We have seen, that for those generators of the global
symmetry group that do not leave the vacuum state invariant exist corresponding Goldstone bosons (Goldstones thm). For the other generators that leave the
vacuum state invariant, we get the same massive particles as in the case of no
spontaneous symmetry breaking.
In the lower part of Fig. 2.7, we consider the case of spontaneous symmetry
breaking of a local symmetry. As we have seen in the discussion of the Higgs
mechanism, the Goldstone bosons associated to the K generators of the gauge
group that do not leave the vacuum state invariant are only gauge phantoms
, which means that they can be gauged away. Doing so, the same number of
massless gauge bosons as the number of Goldstone bosons get massive. Considering the generators that leave the vacuum state invariant, as in the case of
global symmetry breaking, they imply their Higgs bosons corresponding to them.
When we have no Goldstone bosons to gauge away, the corresponding massless
gauge bosons stay massless.
Until now, we have ignored the experimental results in our discussion about
what kind of particles we have been able to detect. If we want to reduce all
possible theories to those that contain only the particles already measured, we
may not have any massless particles except one, the photon. To achieve this, the
entire gauge group must be spontaneously broken, except for a one parameter
subgroup. The gauge field corresponding to it remains then massless and we consider it to be corresponding with the electromagnetic potential that does indeed
contain a massless particle, the photon.
It might be useful to remember historically that the spontaneous symmetry
73
breaking and the theory of non-Abelian gauge fields were considered to be independent. Since both of the theories contain massless particles, the Goldstone
bosons and the massless gauge bosons, their physical relevance were thought to
be little. It took some time (1959-1967) to realize, that these two problems cancel each other, in the way that the Goldstone bosons disappear and the massless
gauge bosons gain mass.
Two physical applications
To see that the very theoretical work we have done until now has important
applications in many experimentally proved theories, we want to have a look at
two examples where our conclusions and the mechanisms we found are successfully
used. Of course, we will discuss here only the very general results and give an
idea how our results from the previous sections get important.
Electroweak theory The weak interaction acts between all particles and is
for example responsible for the -decay of certain particles. Its range is very
small, i.e. much less than 1 fm and as the name says, it is weak compared to the
other two fundamental interactions namely the strong and the electromagnetic
interaction. The weak interaction is though stronger than the third fundamental
interaction, the long ranged gravitation. However, it is possible to formulate a
theory called electroweak theory that connects the electromagnetic and the weak
interaction, thanks to the results of the gauge theory.
Summarized, the results of the theory of the weak interaction are the following.
Three massive particles, the two bosons W and the particle Z are responsible
for the interaction and that the theorys symmetry group is SU(2). We also know
that the massless photon is responsible for the electromagnetic interaction with
its symmetry group U(1).
The idea of the electroweak theory is now to consider the symmetry group
SU (2) U (1) with the little group U (1) for a certain vacuum state that is spontaneously broken. The existing massless gauge boson is identified as the photon
from the electromagnetic theory. The other three gauge fields get massive and
the appearing particles are identified with the three massive particles W and Z.
There also appears a massive Higgs field.
This method is exactly like the general one we described in the previous section.
The results following from this theory do mostly agree very good with the results
from experiments, except for the missing Higgs boson that has not been found
yet.
Superconductor As a second example where the Higgs mechanism actually
takes place we want to consider the phenomenon of superconductivity. Again,
the exact mathematical formulation would take too much time to formulate here
and we focus on the general procedures.
74
Topology in Physics
In the ground state of a material, under a critical temperature Tc an important process might happen: the condensation of the electrons to electron pairs
called cooper pairs. This happens because of a small electron-electron attraction.
The fermion property of one electron has disappeared because the cooper-pairs
can be treated as spinless particles. Such a procedure does force the question if
a symmetry of the system might be broken. To check that, we need some more
mathematical formalism, which we can find in the Landau-Ginzburg theory of
superconductivity.
The idea of the Landau-Ginzburg theory of superconductivity is to couple the
Landau description of a (second order) phase transition to an external electromagnetic field. The Lagrangian of this theory is quite similar to the one we used
in the section 2 where we discussed an example of the Abelian Higgs mechanism,
1
L(x) = F F + (D ) D ( 20 )2 .
(2.99)
4
where the 20 is proportional to the term (T Tc )/Tc . This Lagrangian is valid
only for temperatures below the critical temperature Tc because only boson and
no fermion fields appear. We also know already that this Lagrangian is invariant
under the local U (1) gauge transformation (see equation (2.1) and (2.8)) and
that the vacuum state is spontaneously broken. One result we can use now is
that the formerly massless gauge particle has got the mass m2A = 2e2 20 . Since
we consider the gauge field A as corresponding to the vector potential from
electrodynamics and the former massless particle to the photon, we see that the
latter gets massive with the mass mA . From the theory of electrodynamics we
know that a mass term for the fields appearing in the Lagrangian leads to the
Proca Lagrange density. This generates fields that vanish within the scale of m1A .
This explains the Meissner effect that says that an external electromagnetic field
penetrates a superconductor only to the depth m1A .
3 Magnetic monopoles: Dirac

and Polyakov-tHooft
Philip Grech
Supervisor: Urs Wenger
Although not yet found in nature magnetic monopoles are of
theoretical interest. In particular their existence would yield an
explanation for the quantisation of electric charge.
We will consider two different constructions: The first one was
given by Dirac in 1931, who examined the consequences of a magnetic monopole if a consistent description in electrodynamics and
quantum mechanics is required. We will use two approaches: The
first one is based on the quantisation of the angular momentum
in quantum mechanics, whereas the second uses gauge transformations of the electromagnetic field.
The second construction will follow t Hooft and Polyakov, who
were able to show that magnetic monopole solutions can also
arise in non-Abelian gauge theories. We will not quantise the
fields. Furthermore, we try to illuminate the connection between
the t Hooft-Polyakov monopole and the Dirac monopole.
A lower bound on the monopole mass will be calculated and
we see that under rather special assumptions the field equations
for the t Hooft-Polyakov monopole can be solved analytically.
The so obtained solution will differ from the Dirac monopole
fundamentally.
75
76
Topology in Physics
Dirac Monopole
Covariant Formulation of Electrodynamics

We will not deduce the fact that a covariant formulation of electrodynamics can
be given, yet just repeat the principal results (for further information see [10] or
[11]): The four-current and the four-potential

and (A ) =
(3.1)
(j ) =
~
~j
A
transform like four-vectors. e denotes the electric charge, ~je the corresponding
~ the vector potential.
current, the scalar and A
j = 0
(3.2)
A = j
(3.3)
is the continuity equation,

is the wave equation for the potential. We use the signature (+ - - -) for the
Minkowski metric ( ). = is the dAlembert operator, a Lorentz
scalar (Given that (j ) is a four-vector it follows from (3.3) that (A ) is a fourvector). The Lorentz gauge takes the form
A = 0.
We define the antisymmetric field tensor:
0 E 1 E 2 E 3
E1
0
B 3 B 2
(F ) = ( A A ) =
E2 B3
0
B 1
E 3 B 2 B 1
0
(3.4)
(3.5)
and its dual tensor
1
?F = F .
2
Maxwells equations now take the simple form:
F = j
? F = 0
(3.6)
(3.7)
(3.8)
Notice that the dual field tensor is obtained from the original tensor by ap~ B
~ and B
~ E.
~
plying the duality transformations E
0
Consider now the gauge transformation A = A , where is an arbitrary
function. Then we have
(F 0 ) = ( A A + ) = (F ).
Magnetic monopoles: Dirac and Polyakov-tHooft
77
Thus the electromagnetic field remains unchanged.

This formalism will be used to be able to illustrate certain parallels between
the Abelian and the non-Abelian case. In this chapter we shall content ourselves
with the usual Maxwell equations, since the covariant formulation is not really
needed for the arguments given.
Generalised Duality in Electrodynamics
We contemplate Maxwells equations but introduce as well magnetic charges m
and currents j~m :
~ E
~ = e
~ E
~ + t B
~ = ~jm
(3.9)
~ B
~ = m
~ B
~ t E
~ = ~je
(3.10)
We assumed that the magnetic continuity equation has the same form as the one
for the electric densities.19 Consider now the generalised duality transformation
which rotates the electromagnetic field and its charges and currents:

~
~0
E
E
cos sin
0
0
0
0
~0
~
0
0
0
0
B
B sin cos
0
0
cos
sin
0
0
e
e
.
0 =
m
0
sin cos
0
0
m 0
~0
0
0
0
0
cos sin j~e
je
0
0
0
0
0
cos cos
j~m
j~m
(3.11)
As an example we compute for the first Maxwell equation
~ E
~0 =
~ (cos E
~ sin B)
~ = cos e sin m = 0e .
It is easy to verify that also the other equations are invariant under the above
transformation. Assume that me is equal for every particle in nature. Then there
is an R such that
m
0m = e (sin +
cos ) = 0
e
and
0 = 0
j~m
accordingly.
It is thus a convention when we choose qe = |e| and qm = 0 for an electron.
Therefore the crucial question is not whether non-vanishing magnetic charge densities exist but whether there are two particles with different ratios qqme . We remark
that when we talk about magnetic monopoles in this article, we always think of
a particle which has a different ratio qqme than what has been measured yet.
19
Alternatively we could have written F = j and ? F = k with (k ) = (m , ~jm ).
78
Topology in Physics
Angular Momentum Approach

For the further discussion of the Dirac Monopole we refer to [12] and [13]. Dirac
discovered a quantisation condition for the electric charge by requiring a consistent description of magnetic monopoles in electrodynamics and quantum mechanics. In this chapter we show how we can connect the magnetic charge to
a certain angular momentum. The quantisation of the electric charge will then
follow directly from the quantum mechanical quantisation of this angular momentum.
Consider a particle at the point ~r with electric charge q and mass m in the field
of a magnetic monopole with magnetic charge g placed at the origin (r = |~r|):
~ =
B
g
~r
4r3
~
m~r = q~r B.
(3.12)
(3.13)
Since this is no central force we cannot expect the orbital angular momentum
~ has a rotational symmetry and we may conjecture
to be preserved. Nevertheless B
a conserved quantity. Thus compute
d
(~r m~r ) = ~r m~r
dt
qg
~r (~r ~r)
=
4r3 h
i
qg 2
=
~rr ~r(~r ~r)
4r3

d qg
=
r .
dt 4
We set r :=
as
~
r
r
and used
d ~
r
dt r
r~
r)
= ~rr ~r(~
. Now define the total angular momentum
r3
qg
J~ := ~r m~r
r,
4
(3.14)
which is indeed a conserved quantity. To understand the second term of J~ remember that the construction of the Maxwell torsion tensor yields the identification
~ B.
~ This leads to an expression for the
of the field momentum density with E
angular momentum of the electromagnetic field:
Z
~
~ B)).
~
Jem = d3 x(~x (E
(3.15)
79
~ denotes the field of the electric charge q at ~r, B

~ is given by (3.12).
E
Z
g
i
Jem =
d3 xE j (ij xi xj )
4x
Z
i
g
x
=
d3 xE j j (
)
x 4
Z
i.b.p
~ E
~ g xi .
= d3 x
4
~ E
~ = q (~x ~r), (~x) is the Dirac delta, we get the angular momentum
Using
of the electric field:
qg
J~em = r.
(3.16)
4
Thus we can argue that the total angular momentum for which we derived conservation is nothing but the sum of the orbital angular momentum of the particle
and the angular momentum of the electromagnetic field. Since
qg
qg
r J~ =
cos =
4
4J
(3.17)
~
we see that the trajectory of the particle lies on a cone around the negative Jqg
axis with semi-vertical angle arccos( 4J ) and its apex at the monopole. The first
equation in (3.17) will lead us to our goal: If we involve quantum mechanics it
is reasonable to expect that the components of J~ will satisfy the commutation
relations of the angular momentum algebra (see also chapter (1)) and have thus
eigenvalues being integer multiples of 12 ~. We consider the component along the
straight line connecting the monopoles and the charged particle and get
qg
1
= n~,
4
2
n Z,
(3.18)
which is the Dirac quantisation condition. This, of course, is not a rigorous

derivation but at least it makes things plausible.
Quantisation of Charge
To see that (3.18) is equivalent to the fact that charge is quantised consider the
case of many particles with either a magnetic charge gi or an electric charge qi .
Get:
qi gj
1
= nij ~
nij Z.
(3.19)
4
2
Fixing gj we get
2~
qi =
nij .
gj
We calculate the highest common factor n0j of the {nij } and conclude that every
~ n0j . By Euclids algorithm
electric charge is thus an integral multiple of q0 = 2
gj
80
Topology in Physics
(see e.g. [14]) n0j is a linear combination of the nij with integer coefficients, thus
q0 is a linear combination of the qi with integer coefficients. Since the total charge
of the system equals the sum of the charges of the involved particles, it is possible
to really measure q0 . The same reflections lead to an elementary magnetic charge
g0 . Nota bene: It suffices a single magnetic monopole in the whole universe to
guarantee the quantisation of electric charge.
Forces
In analogy to the force law we have for electric charges, we can assume that two
g2
magnetic charges g0 repel each other with 4r0 2 and we can compare the magnitude
of these forces:
Fm
g2
n2 q 2
= 02 = 0 ( 0 )2 5 103 n20 ,
(3.20)
Fe
q0
4 4~
which is large even for n0 = 1. It is therefore expected to be quite difficult to pairproduce magnetic monopoles and that they are much heavier than electrically
charged particles (we will calculate a lower bound on the mass in chapter 2).
Quantisation of Motion
In this section we want to analyse more precisely the quantum mechanical behaviour of our system given in 1. The Lagrangian of a particle in an electromagnetic field is
1
~ q
L = m~r 2 + q~r A
(3.21)
2
~ is the electromagnetic four-potential (
~ A
~ = B
~ and
where (A ) = (, A)
~ since this leads to the correct equations of motion:
= E),
L
rj
Ai
q
rj
rj
d L
d
Aj
Aj
= (mrj + qAj ) = m
rj + q
ri + q
dt rj
dt
ri
t
= q ri
Ai Aj
Aj
)q
q
rj
ri
rj
t
~ j + qEj .
= q(~r B)
m
rj = q ri (
A Legendre transformation gives the Hamiltonian

1
~ 2 + q
H = p~ ~r L =
(~p q A)
2m
with
p~ =
L
~
= m~r + q A.
~r
(3.22)
(3.23)
81
Note that this Hamiltonian can also be obtained by translating the four-momentum
p2
of the Hamiltonian H = 2m
of a free particle: p p + qA . Since the magnetic
~ cannot exist everywhere. We will
field is not source-free, a vector potential A
come back to this problem, yet ignore it for the moment.
We define the Poisson bracket by
3
X

{, } :=
( i i i i)
r p
p r
i=1
(3.24)
{ri , rj } = {pi , pj } = 0
{ri , pj } = {ri , mr j } = ij
{mr i , mr j } = q( i Aj j Ai ) = qijk B k .
(3.25)
(3.26)
(3.27)
and get
Using the product rule {, } = {, } + {, } we compute expressions

~ = ~r m~r :
for Poisson brackets with the angular momentum L
{Li , rj }
ijk rk
{Li , mr j }
~ B i rj )
ijk mr k + q(ij ~r B
qg
ijk mr k +
(ij ri rj )
4r
qg
ijk mr k + { ri , mr j },
4
in our case
where ri denotes ri /r. We are now ready to derive expressions for the total
qg
angular momentum J~ = ~r m~r 4
r:
{J i , mr j } = ijk mr k
{J i , rj } = ijk rk
{J i , J j } = ijk J k .
(3.28)
(3.29)
(3.30)
Since H = 12 m~r 2 we can re-derive the conservation of J~ from the first equation.
We perform the transition to quantum mechanics by applying the correspondence
principle:
1
{, }
[, ]
(3.31)
i~
~
p~ i~.
(3.32)
Equation (3.30) confirms that the total angular momentum (operator) does indeed fulfill the requirements of an angular momentum algebra as we assumed in
chapter 1.
82
Topology in Physics
We replace mr i = pi qAi by i~ i qAi = i~Di . The operator
D = + ieA
(3.33)
(e := q/~) is called covariant derivative. We will see that this concept can be
extended to the case of non-Abelian gauge theories. The gauge fields which form
the connection of the covariant derivative are then Lie algebra-valued.
The Schrodinger equation for the wavefunction of a charged particle is
~2 2
D + q = i~
2m
t
(3.34)
where denotes the wavefunction.

It can easily be shown that this equation is not invariant under a gauge
transformation A A0 = A + . However, we can guarantee that the
equation is gauge-covariant if we specify that under a gauge transformation
0 = eie .
(3.35)
In this section we have fortified the assumption that we could assume an angular
momentum algebra in section (1). Still it remains the problem that the exis~ cannot be assumed. By using the covariance of
tence of the vector potential A
Schrodingers equation, we will see in the next section an approach which avoids
these difficulties.
Vector Potential Approach
Construction of the Potential It is a classical theorem of vector analysis
~ with
~ B
~ = 0 then
(named after Helmholtz) that if we have a vector field B
~ satisfying
~ A
~ = B.
~ A
~ is the so called
there exists another vector field A
~ is
vector potential. Since in the case of magnetic monopoles the divergence of B
non-zero at the origin, it is not possible to find a vector potential that is defined
on every point in R3 . The best we can get is a vector potential being defined
everywhere except on a line - the Dirac string S - from the origin to infinity. To
be precise, we consider the magnetic field of the monopole and attach a solenoid
along the negative z-axis, which corresponds to the Dirac string:
~ sol =
B
g
r + g(z)(x)(y)
z
4r2
(3.36)
where (x) denotes the Heaviside function. From electrostatics we know that the
first term is the gradient of the fundamental solution for the Laplace operator.
Therefore we can easily compute
~ B
~ sol = 0.
(3.37)
83
~ sol can be represented by a vector potential A

~
Thus by Helmholtz theorem, B
~
~
~
such that Bsol = A. For the monopole field we find
g
~ A
~ g(z)(x)(y)
r =
z.
4r2
(3.38)
We will now construct the vector potential explicitly. Since the magnetic field
is radial, only the radial part of the rotation operator is of interest:
~ A)
~ r=
(
A
( (sin A )
).
r sin
(3.39)
~ r) =
Thus by symmetry we can expect the vector potential to have the form A(~
A(r, ) where denotes the vector (cos , sin , 0). Consider a sphere with center
at the origin. The magnetic flux through a circle C that intersects the sphere is
spherically symmetric and is therefore proportional to the solid angle subtended
by the circle C and the origin. This solid angle corresponds to the surface M of
the sphere cap that has been cut off by C:
M = (a2 + h2 ) = (r2 sin2 + r2 2r2 cos + r2 cos2 ) = (2r2 2r2 cos )
where a is the radius of C and h is the height of the cap. The magnetic flux is
gM
g ((2r2 2r2 cos ))
g
=
= (1 cos ).
2
2
4r
4r
2
Finally, we get
g
(1 cos ) =
2
Z
~ dS
~=
B
Z
~ A
~ dS
~=
~ d~l = 2A(r, )r sin

A
and
~ r) = g (1 cos ) .
A(~
(3.40)
4r
sin
Notice that for = the vector potential becomes singular. This is not sur~ on the Dirac string.
prising since we are then evaluating A
With the so obtained vector potential it is possible to describe a quantum
~ From the requirement that
mechanically behaving particle in a magnetic field B.
for different choices of the Dirac string, we should be led to equivalent equations,
we can deduce Diracs quantisation condition.
Gauge Transformation and Dirac Condition Imagining a Dirac string on

the positive z-axis we can deduce with the construction we used in the previous
chapter another expression for the vector potential:
~ S = g (1 + cos ) .
A
4r
sin
(3.41)
84
Topology in Physics
~ S has a singularity on the north pole of the two-sphere S 2 of radius 1 with

A
center at the origin. For clarity, we shall denote the field from equation (3.40) by
~N .
A
~ N and A
~ S are well defined and generate the same
On a region where both A
magnetic field, they may only differ by a gauge transformation. Such a region
would be the equator E and there we have
~N A
~ S =
A
g
g
=
= .
(3.42)
2
2
The gauge function is not continuous. We could not avoid this fact as we
see by a calculation of the enclosed magnetic charge (N: northern hemisphere
(0 6 6 /2); S: southern hemisphere (/2 6 6 )):
Z
Z
Z
~
~
~
~
~ S dS
~
g
=
B dS =
BN d S + B
2
S
N
S
Z
Stokes
~N A
~ S ) d~l = (0) (2)
(A
=
(3.43)
E
Reconsidering equations (3.34) and (3.35) we see that the only requirement for
continuous physical quantities is the continuity of the phase factor eie . Thus:
!
eie(0) = eie(2)
ge = 2n gq = 2n~.
(3.44)
So the more rigorous approach via the vector potential leads again to the Dirac
condition (3.18).
t Hooft-Polyakov Monopole
Brief Review of General Gauge Theory Formalism

We follow the introduction into the matter of gauge theory given in [15] and [12].
Consider a Lie group G with Hermitian generators {T a }. L(G) is the Lie algebra
of G, the tangent space for the Lie group at the unit element. So, any C 1 -curve
g(t) in G can be written as
g(t) = 1 + At + O(t2 )
(3.45)
where 1 denotes the unity element of G and A is an element of L(G). Let D be

a representation of G and be a Lorentz scalar field transforming under D:
D(g) where g G.
(3.46)
If g = g(x), this is a local gauge transformation. With D we can associate a

representation of the Lie algebra which since there cannot be any confusion
we will also denote by D:
D(1 + A) = 1 + D(A)
(3.47)
85
where is a small parameter. It is not too hard to show that D as a function on

L(G) preserves the Lie brackets and is linear.
We introduce Lie algebra-valued gauge fields Wa such that
W = Wa T a .
(3.48)
The usual derivative does not transform properly, since

D(g) + D(g),
(3.49)
and we replace it by the covariant derivative D for which we postulate that

D D(g)D . Such a derivative indeed exists if we specify that the gauge
fields transform according to
i
W gW g 1 + ( g)g 1 ,
e
(3.50)
because
D = + ieD(W )
i
D(g) + D(g) + ieD(gW g 1 + ( g)g 1 )D(g)
e
= D(g) + D(g) + ieD(g)D(W ) D(g)
= D(g)D .
(3.51)
Here, we used the identities D( gg 1 ) = D(g)D(g 1 ),
D(gW g 1 ) = D(g)D(W )D(g 1 ) and linearity of D() as a function on the Lie
algebra. We compute the commutator of the covariant derivative and see that it
is no differential operator:
[D , D ] = . . .
= ie{D( W ) D( W ) + ieD([W , W ])}
(3.52)
and define the antisymmetric gauge field tensor by

G = Ga T a = W W + ie [W , W ]
(3.53)
[D , D ] = ieD(G ).
(3.54)
so that we have:
This equation shows the behavior of G under a gauge transformation:
G gG g 1 .
(3.55)
We conclude this chapter with some useful formulas: The covariant derivative
obeys the Jacobi identity:
[D , [D , D ]] + [D , [D , D ]] + [D , [D , D ]] = 0.
(3.56)
86
Topology in Physics
Applying this to an arbitrary and using equation (3.54) yields

[D , [D , D ]] = ie[D , D(G )]
= ieD(D G )
(3.57)
D G = G + ie[W , G ]
(3.58)
where
is the covariant derivative for the adjoint representation gg 1 (g G),
which has this form due to equation (3.55). The Jacobi identity produces the
Bianchi identity:
D G + D G + D G
D ? G .
(3.59)
The Georgi-Glashow Model

We follow [12] and [16]. Consider a SO(3) gauge field interacting with a Higgs
field . The Lagrangian is
1
1
L = Ga Ga + D D V (),
4
2
(3.60)
where
1
V () = (21 + 22 + 23 a2 ).
4
The gauge field strength Ga is given by

Ga = Wa Wa eabc Wb Wc .
(3.61)
(3.62)
Wa is the gauge potential. We set (Ta )ij = iaij . The covariant derivative,
D := + ieD(W ), of is then given by
(D )a = a eabc Wb c .
(3.63)
R
The conditions for the action S = M 4 Ld4 x to be stationary are (we applied the
Euler-Lagrange equations):
(D G )a = eabc b (D )c
(D D )a = a (2 a2 ).
(3.64)
(3.65)
Furthermore, by construction the Bianchi identities are automatically valid:

D ? G = 0.
(3.66)
The energy density corresponds to the (0,0)-component of the energy-momentum

tensor for this theory:
2

1
00 = {(Ea i )2 + (Bai )2 + (a )2 + Di a } + V ()
2
(3.67)
87
where
Ga0i = Ea i ,
Gaij = ijk Bak
and a = (D0 )a .
(3.68)
We observe that 00 vanishes if and only if:

Ga = 0
(3.69)
(D )a = 0
(3.70)
1
V () = (2 a2 )2 = 0.
(3.71)
4
A field configuration satisfying equations (3.69), (3.70) and (3.71) is called
vacuum configuration. An example is:
a = aa3
Wa = 0,
(3.72)
where ab denotes the Kronecker delta. A configuration satisfying equations (3.70)

and (3.71) is called Higgs vacuum configuration. The introduction of a Higgs field
is of great importance: To reach a state of finite energy, we have to ensure that
our system reaches the Higgs vacuum state. The condition V () = 0 implies that
a a = a2 . Alternatively stated:
M0
where M0 = { : V () = 0}.
(3.73)
M0 is a two-dimensional sphere of radius a. Since assumes a certain value in

M0 , it is not invariant under all rotations of SO(3) but only under those around
the -axis. They form a subgroup H of SO(3). Of course, for each position
of this group is isomorphic to SO(2), or equivalently U(1), and we will call it
simply H. H is the exact symmetry group of the theory. By the introduction
of the Higgs field the SO(3) symmetry of our model is spontaneously broken
down to U(1).20
It is reasonable to identify H with the electromagnetic gauge group. The generator of this group is given by the orthogonal projection of a SO(3) generator T
onto the -axis: T /a.
We will see that in the Higgs vacuum we can naturally obtain a correspondence
with Maxwells fields. If equations (3.70) and (3.71) do not hold, fundamentally
new solutions can occur, such as magnetic monopoles.
From gauge field theory it is known that if we expand the Lagrangian about a
vacuum configuration 0 , (h(x) = 0 + (x)), the term in the Lagrangian that
is proportional (x)2 corresponds to the square of the mass divided by 2~.
20
In other words: Spontaneous symmetry breaking is the situation where the ground state
of the theory has less symmetries than the Lagrangian itself.
88
Topology in Physics
Table 3.1: Particle Attributes
Mass
Higgs particle
= a(2)1/2 ~
Photon
0
Massive gauge particles M = ae~ = aq
Spin
0
~
~
Electric charge
0
0
q = e~
To obtain the electric charge, compare the covariant derivatives:

+ ieWa Ta
Q
and + i A .
~
(3.74)
The second equation results from (3.33). This comparison can be done by performing a canonical projection for the connection of the SO(3)-covariant derivative:
W ) ( T)
(3.75)
a
a
Taking the previously mentioned projection in L(G), it is natural to identify
A with /a W . We find an electric charge operator

ieW
ie(
Q = e~
T.
a
(3.76)
This expression is valid in any representation. Let us quickly look at this

more closely: In Maxwells theory, the U (1) generator is essentially given by the
electric charge Q. A possible explanation for its quantisation would be that Q
actually results from a larger gauge group, say SO(3). Q is then also a generator
of SO(3) and element of the angular momentum algebra, so it has quantised
eigenvalues.
Up to now, no magnetic monopoles are involved. Yet we need a mechanism, that
tells us how to select the right directions within so(3). This is given by the Higgs
field (which cannot vanish in the vacuum in order to provide a distinct direction
in the Lie algebra) and automatically yields magnetic monopoles. If extra fields
are added to the model, Ta may even have any eigenvalue which is half an integer.
Thus we are finally led to the condition that Q can only assume values that are
integer multiples of q0 := 21 e~.
Monopole Solution
We are looking for simple yet non-trivial solutions for the model. A general
solution for equations (3.64) and (3.65) is not known. We can expect the solution
89
with the lowest non-zero energy to be time-independent and highly symmetric.

Several symmetry considerations lead to the ansatz (see [12]):
a (~r) = H(aer)ra /ar2
Wa 0 (~r) = 0
(3.77)
rj
(3.78)
Wa (~r) = aij 2 [1 K(aer)] .
er
To obtain finite energy we have to ensure that asymptotically, (3.71) is valid:
i
a (~r) = lim a (r
r) = a
ra .
r
(3.79)
Thus defines a map from a two-sphere at spatial infinity to M0 , which is

essentially S 2 :
homeo
homeo
2
: S2
M0
(3.80)
= S
= S 2.
Note that it is impossible to transform configuration (3.79) into the vacuum
configuration (3.72) in a continuous way: For every > 0 there is a neighborhood
2
U of the south pole of S
, such that two values of that are assumed on U
exist that would have to be mapped by two rotations with rotation axis differing
by /2. Since this difference cannot be made arbitrarily small, the deformation is
not continuous. Therefore we can expect the monopole to be stable in the sense
that it will not fall into the vacuum state.
It exists a simpler way to obtain the equations of motion than to just plug our
ansatz into equations (3.64) and (3.65). Faddeev ([17]) and Coleman ([18]) have
found a principle which roughly states the following: Suppose we have a function
(or as in our case: a functional) F defined on a set X. Let G0 be the group
describing the symmetry operations of F on X. We denote the set of fixed points
with respect to all symmetry operations by X0 . Then the stationary points of F
over X0 are also stationary points of F over X.
It is a fact that our ansatz (3.77) and (3.78) consists of those configurations
which form the set X0 , since it was obtained by symmetry considerations.
We could also argue that we are looking for those configurations which minimize the energy ( := aer):
Z
E = Ld3 r
Z
4a d 2 dK 2 1 dH
=
( (
) + (
H)2
e 0 2
d
2 d
1
+ (K 2 1)2 + K 2 H 2 + 2 (H 2 2 )2 ).
(3.81)
2
4e
Applying the Euler-Lagrange equations leads to:
2
d2 K
= KH 2 + K(K 2 1)
d 2
(3.82)
90
Topology in Physics
d2 H
= 2K 2 H + 2 H(H 2 2 ).
2
d
e
(3.83)
Since we require finite energy, equation (3.81) implies certain boundary conditions. As goes to zero we should be able to compensate the divergence resulting
from the pole of 1/ 2 . The first term of the integrand does not cause any problems. The second term gives the condition
H 6 O() ( 0).
(3.84)
K 2 1 = (K 1)(K + 1) K 1 6 O() ( 0).
(3.85)
For the third term:
These boundary conditions are also enough stringent for the two remaining terms.
As goes to infinity we need to ensure that the integrand vanishes fast enough:
K 0 and H
( ).
(3.86)
The existence of solutions of (3.82) and (3.83) with boundary conditions (3.85)(3.86) has been proven by Schwarz (see [19]) and was first conjectured by a
numerical analysis.
Equations (3.77) and (3.78) will (in the regime where ) asymptotically
take the form:
ra
rj
a = 2 aer and Wa i = aij 2
(3.87)
er
er
leading to a field strength tensor
Gaij = i Wa j j Wa i eabc Wb i Wc j
1
1
ijk ra rk
ijk rk a .
4
er
aer3
(3.88)
After a projection on the -axis we obtain

F ij =
1
Gij = 3 ijk rk
a
er
ri
, since F ij = ijk B k .
(3.89)
er3
Comparing this result with equation (3.12) gives the magnitude of the magnetic charge:
4~
4
=
.
(3.90)
g=
e
q
Bi =
91
This is consistent with the Dirac condition (3.18). By (3.76) it follows that the
smallest possible charge that may enter the theory is q0 = 12 q because the possible
electric charges are related to the eigenvalues of T. We get
q0 g
1
=
4~
2
(3.91)
and g assumes the smallest value allowed by the Dirac condition.

The set of equations (3.82) and (3.83) can be solved in the limit where .
Thus we can make a statement of how fast the asymptotic values of the boundary
conditions (3.85)-(3.86) are assumed. We expand (3.82):
d2 K
H 2 K(K 2 1)
=
K
+
K
d 2
2
2
|{z} |
{z
}
1
(3.92)
and (3.83):
d2 H
d 2
1
1
2
2
2
2K
H
+
H(H
(H + )(H ).
2
2 e2
e2 | {z }
| {z }
2
0
=:h
2 z }| {
2 (H )
e
d2 h
2
h.
d 2
e2
(3.93)
The equations of motion are now decoupled linear differential equations and as
such easy to solve. Finally, we get
K = O [exp()] = O [exp(M r/~)]
(3.94)
H = O [exp(/M ] = O [exp(r/~)]
(3.95)
where := (2)1/2 a~ and M := ae~ are the masses of the Higgs and the massive
gauge particles respectively as introduced in chapter (2). Thus by this result
(i.e. by the at least exponential decay), we can consider the t Hooft-Polyakov
monopole to have finite size, principally determined by the Compton wavelengths
~/M and ~/:
Inside we have a smooth structure; outside we obtain a field configuration
indistinguishable from that of the Dirac monopole.
Connection with Maxwells Equations
In chapter 2 we have seen that outside a radius R0 , which is determined by the
Compton wavelengths of the heavy particles in the theory, our field is exponentially close to the Higgs vacuum (3.70) & (3.71). This was based on an analysis
92
Topology in Physics
of the field configuration with the lowest non-zero energy.

We will now assume that any finite-energy solution will be very close to the Higgs
vacuum except for a finite number of areas, which we will regard as monopoles.
With a given , the general form of W satisfying (3.70) and (3.71) is
W =
1
1
+ A
2
ae
a
(3.96)
where A is an arbitrary smooth four-vector function (see [20]). We shall content

ourselves with the following calculation, which shows that (3.96) is indeed in the
Higgs vacuum:
1
e
+
(
A )
a2
a
1
1
= + 2 ( ) 2 ( ) = 0.
a
a
D = (
(3.97)
The second term in the second line vanishes due to the Leibniz rule:
0 = ( ) = ( ) + ( ) = 2 ( ).
Furthermore we get
G =
1
F
a
with F =
1
( ) + A A .
a3 e
(3.98)
This result is derived in the appendix. The field tensor points in the direction
of . So the only non-vanishing component of G is exactly the one associated
with the U (1) gauge group of rotations about , which we can identify with the
electromagnetic field strength tensor.
0 = D G = G eW G
1
1
1
1
= ( F ) e( 2 + A ) ( F )
a
ae
a
a
1
1
1
=
F + F + 3 F ( )
a
a
a
1
1
1
=
F + F 3 F a2
a
a
a
1
=
F .
(3.99)
a
Since the modulus of is finite, this equation only holds if F = 0. Analogously, we can transform the Bianchi identity D ? G = 0 into the homogeneous
Maxwell equations ? F . So outside the monopole our SO(3) theory coincides
with Maxwells U (1) theory.
93
Generalised Quantisation Condition for the t Hooft-Polyakov Monopole

We proceed with our analysis of a field in the Higgs vacuum by regarding the
magnetic flux, g , through a surface . Since outside the monopole the fields obey
Maxwells equations, it is only possible to receive a non-zero flux by integrating
over a surface that surrounds a region where (3.70) and (3.71) are not satisfied,
namely around a monopole region:
Z
~ dS
~
g =
B
Z
1
=
ijk F jk dS i
2
Z
1
1
(3.98)
=
ijk 3 ( j k )dS i .
(3.100)
2
ae
The last term in equation (3.98) vanishes due to Stokes theorem (S = ). If

2
has the shape of a sphere (as we had it before: take e.g. S
) we see immediately
i
2
that the derivatives are tangential to Sa . Thus the magnetic charge only
depends on the values that assumes on .
Furthermore consider a variation of the field
+
= 0.
(3.101)
The variation of the integrand is then given by

( j k ) = 3 ( j k ) + j ( k )

k ( j )
(3.102)
using i = i and the cyclic permutations of ~a (~b ~c). As we consider
three spatial dimensions, j k is orthogonal to and the first term vanishes. Again, we can apply Stokes theorem and the last two terms vanish when
integrated over .
Thus the magnetic flux is stable against small deformations (homotopies) of
the Higgs field. Physically, homotopies are generated by
time development of ,
continuous gauge transformations of ,
deformations of the surface within the Higgs vacuum.
Thus g is time- and gauge-independent and does not vary under continuous
deformations of . In particular, g is an additive quantum number as made
plausible in the figure 3.1:
g12 = g1 + g2 .
(3.103)
94
Topology in Physics
Figure 3.1: Continuous deformation of the surface 12 into the surfaces 1 and
2 .
The quantisation comes from the fact that we can write g = 4N where N
indicates the so called Kronecker index of the map : M0 . To be more
precise, continue with (3.100):
Z
1
ijk a abc j b k c dS i
(3.104)
g = 3
2a e
is a two-manifold and can thus (locally) be parametrised by two parameters
( {1, 2}):
xi = xi ( ).
With elementary calculus we derive
1
xm xn
dS i = imn d2
2

and j b =
b
xj
and get:
g
Z
1
1
b c xm xn
= 3
ijk a abc imn j k d2
2a e 2
x x
(using ijk imn = jm kn jn km and the chain rule)
Z
1
b c
1
abc a d2
= 3
ae 2

Z
1
b c
=
abc a d2 .
(3.105)
2e

For the square of the integrand we compute

b c
(abc a )2 = ( a ( )a )2

= 4G( , )

(3.106)
95
has unit length

where G denotes Grams determinant (we used the fact that
and is parallel to ). Thus we are integrating over 2 G. While ( , )
2
covers S 2 N+ times with the positive sign of G and
covers the sphere S
once,
N times with the negative sign. Since the a are supposed to be single-valued
functions the difference N+ N has to be an integer N, thus
g =
4N
.
e
(3.107)
Considering the map

N (~r) = a(cos N sin , sin N sin , cos ),
(3.108)
we see immediately that indeed every N can be realised. We obtain the beautiful
result that the magnetic charge is quantised by topological reasons. The smallest
electric charge entering the theory is q0 = 21 e~ and we deduce
gq0
1
= N,
4~
2
(3.109)
which is identical to the Dirac quantisation condition (3.18)!

Topological Aspects
In what comes next, we will see three further approaches to characterise a smooth
between two spheres (see [21], [22] or [16]), that emerges from the map
map
induced by the Higgs field. In the end of this section we will see the implications
of these three viewpoints.
:
Brouwer Degree Reconsider the mapping induced by the Higgs field
2
2
S S and introduce also in the target space local coordinates, say
( {1, 2}), such that = ( ). Fix a regular value = (1 , 2 ) (i.e.
does not vanish at any point in
1 ()).
the Jacobian of
at the point is defined as
The Brouwer degree of the map
) =
d(;
X
1 ()
sgn det(
).

(3.110)
It can be shown that d does not depend on the particular choice of a regular
) = d().
point in S 2 . Thus we can write: d(;

Homotopy Classes A continuous map : X Y , where X and Y are
topological spaces, is called a based map, if we can identify two base points x0 , y0
96
Topology in Physics
in X and Y respectively such that (x0 ) = y0 . Two based maps 0 and 1 are
called homotopic if and only if there is a continuous map
: X [0, 1] Y
(3.111)
=0 = 0 and |
=1 = 1 and (x
0; ) =
with a parameter [0, 1], such that |
y0 [0, 1]. It can be shown that homotopic defines an equivalence relation.
For the special case where X
=S n the set of homotopy classes of based maps
: S n Y is denoted by n (Y ). For n > 1, n forms a group, where the
group multiplication of two equivalence classes is realised as follows: Consider a
representative of a class. Due to the stereographic projection such a map can be
characterised by 1 : (I n , I n ) (Y, y0 ), where I := [0, 1] and the homotopies
1 are required to satisfy 1 (I n ) = y0 . It can be shown that
1 2 (s1 , s2 , . . . , sn ) := {
1 (2s1 , s2 , . . . , sn )
s1 [0, 1/2]
.
2 (2s1 1, s2 , . . . , sn ) s1 [1/2, 1]
(3.112)
does indeed define a group operation.

An interesting result of algebraic topology is
n (S n ) = Z n > 1.
(3.113)
Thus in our case, every homotopy class is uniquely characterised by an integer

number.
Poincar
e-Hopf Index Let ~x0 be an isolated zero of the Higgs field . Introduce a small two sphere S2 with center ~x0 . The Brouwer index of the normalized
field
: S 2 S 2 is called Poincare-Hopf index i of the zero ~x0 .

Implications Now we illuminate the connections between these three definitions: Firstly, it is a special case of a general theorem of Hopf that two smooths
maps 0 and 1 are homotopic if and only if they have the same Brouwer degree.
equals the
Secondly, if has only nondegenerate zeros, the Brouwer degree of
sum of the Poincare-Hopf indices of all zeros of . Thirdly, the Brouwer degree
is equal to the Kronecker index N .
d()
These facts enable us to determine the magnetic charge of a field configuration

by topological analysis of the Higgs field.
Relation between the Dirac and the t Hooft-Polyakov Monopole
See also [16] and [13]. It seems rather amazing that we receive the same quantisation condition for the Dirac and for the t Hooft-Polyakov monopole. There is
indeed a connection between these two.
97
Notwithstanding certain technical assumptions, we can write MH = G/H

where MH is the vacuum manifold, G the original gauge group and H the gauge
group after symmetry breaking. Thus the magnetic charge can by the previous
chapter be characterised by 2 (G/H).
Furthermore, in the case of the Dirac monopole, the gauge transformations
ie()
e
between the gauge fields on the two hemispheres of S 2 (see chapter (1))
are paths in U (1)
= S 1 and can be classified by 1 (U (1)) = 1 (S 1 ) = 1 (H).
The integer n has again the meaning of a winding number in this context.
It is a remarkable result of algebraic topology that
2 (G/H) = 1 (H),
(3.114)
2 (SO(3)/U (1)) = 2 (S 2 ) = 1 (S 1 ).
(3.115)
or in our special case:
This corresponds with equation (3.113) and yields a link between the Dirac and
the t Hooft-Polyakov point of view.
such that it points in the direction
To be more concrete, let us try to gauge
has a non-vanishing winding number, we cannot transform
of T 3 . Of course, if
it smoothly into the constant map. However, this can be done on the northern
(we denote the group element of the associated gauge transformation by g (1) ) and
southern hemisphere (g (2) ) of S 2 respectively. On the overlap region (=equator),
1
0 = T 3 and therefore lies in the U (1) subgroup of SU (2)
g (1) g (2) preserves
1
generated by T 3 : g (1) g (2) = exp(i()T 3 ).
The gauge transformations g (1) and g (2) are well defined on their respective
, N
Z. Consider now the two abelianised gauge
regions: (2) = (0) + 2 N
fields that we get after the transformation (which, for the Higgs field, looks as
(1) = D(g (1) g (2) 1 )
(2) = g (1) g (2) 1
(2) ):
follows:
1~
A(1) A(2) = .
e
(3.116)
Since the Higgs field is now constant, equation (3.98) exactly simplifies to
the case of the Dirac monopole: the total magnetic flux is determined in the
same way and does only depend on the amount by which increases around the
= N.
equator. One can show that N
Bogomolny Bound on the Monopole Mass
In the remaining part of this article we refer to [12]. Contrary to the Dirac
monopole, which we have to regard as an external source and for which it is
therefore impossible to calculate the mass, we can give an estimate for the lower
bound on the mass of the t Hooft-Polyakov monopole.
98
Topology in Physics
From differential geometry we have the following formula holding for covariant
derivatives:
X hY, Zi = hDX Y, Zi + hY, DX Zi
where the connection of D is compatible with h , i and X,Y,Z denote smooth
vector fields. Thus (without proving the compatibility of D with h , i), it follows
that
k (Bak a ) = Dk Bak a + Bak (Dk )a .
(3.117)
| {z }
=0
The first term vanishes due to the Bianchi identity. We compute the magnetic
charge:
Z
Z
Z
1
k
3
~ dS
~ = k B d r =
g =
B
k (Bak a )d3 r
a
2
S
Z
1
=
Bak (Dk )a d3 r.
(3.118)
a
Similarly we receive an expression for the electric charge
Z
Z
1
~
~
q=
E dS =
Ea k (Dk )a d3 r
a
2
S
(3.119)
using the equations of motion (3.64). Consider the center-of-mass frame of the
monopole. Einsteins relation E 2 = M 2 + p~2 (E is the energy and p~ the momentum) simplifies to E = M and we get
Z

1
M =E =
d3 r{ (Ea k )2 + (Bak )2 + (D0 )2 + (Di )2 + V ()}
2
Z
1
>
d3 r {(Ea k )2 + (Bak )2 + (Di )2 }
2
Z
Z
1
1
3
k
k
2
=
d r{Ea (D )a sin } +
d3 r{Bak (Dk )a cos }2
2
2
+a(q sin + g cos )
> a(q sin + g cos )
(3.120)
where the non-trivial step is the third equality. The estimate holds for any R
and we can interpret the last line as the standard scalar product of the vector
~ = (aq, ag) and the unit vector ~e = (cos , sin ). Since is arbitrary, we choose
~e parallel to ~ . The last line then equals the norm of ~ and we get the most
stringent inequality, the Bogomolny bound on the monopole mass:
M > a(q 2 + g 2 )1/2 .
(3.121)
For the t Hooft-Polyakov monopole (equation (3.119) would not have been
needed):
M > a|g|.
(3.122)
99
(3.107)
Choosing the smallest possible magnetic charge |g| = 4/e we can compare
the monopole mass Mg with the mass Mq = ae~ = qa of the heavy gauge bosons
in the theory (see table 3.1):
Mg >
4~
Mq .
M
=
q
q2
(3.123)
{1, 1/4} depending on whether the electron charge is q or q/2 and is the
fine-structure constant. In any case, Mg is is much larger than Mq . This makes
it impossible to observe magnetic monopoles with contemporary technologies.
Bogomolny-Prasad-Sommerfield (BPS-) Monopole
In this chapter we will see that for the special case of a vanishing potential
V () analytic solutions of our problem exist, which saturate the bound (3.122).
Thus the inequalities in the Bogomolny estimate have to become equalities. This
implies the following conditions (q=0):
D0 = 0
Ea i = 0
Bai = (Di )a
as
(3.124)
g
>
<
V () = 0.
(3.125)
(3.126)
These equations can only be simultaneously realised in the limit 0+ for

the proportionality constant of the potential V .
Although V vanishes ultimately, we will keep the boundary condition
|| a
as
r .
(3.127)
such that the charges are still defined and quantised. The equations of motion
now take the form
D G = e D
D D = 0.
(3.128)
Equation (3.125) is called the Bogomolny equation and is a first order differential
equation. While the equations of motion give a condition for stationary points
of the energy, the Bogomolny equation yields a global minimum of the energy
for a given N . With a straightforward computation, it can be shown that such a
minimum is indeed a stationary point.
Using our ansatz (3.77) & (3.78) we get:
dK
= KH
d
dH
= H (K 2 1).
d
(3.129)
100
Topology in Physics
By differentiating the one and substituting it into the other equation we would
get the original equations (3.82) and (3.83). The change of variables H = 1h
and K = k yields the equations ( denotes the derivative)
k 0 = hk
and
h0 = k 2 .
(3.130)
This system can be decoupled:

h00 = 2kk 0 = 2hk 2 = 2hh0 = (h2 )0 h0 = h2 + A,
A R.
(3.131)
The boundary conditions (3.85)-(3.86) yield that h 1 and h0 0. This

leads by substitution in the differential equation to A = 1. So we can integrate equation (3.131):
Z
1
+ B = dh
= coth1 (h), B R
2
1h
h = coth( + B).
(3.132)
B has to be zero since this is the only value for which H = h a is continuous
(only then the divergence of coth( + B) at = B can be compensated) and k
is now easily obtained:
1
k = h0 =
.
(3.133)
sinh
!
Since K 1 6 O() as 0 we choose the positive sign and finally obtain a

solution in terms of elementary functions:
H() = coth 1
sinh
The massless Higgs field is now long range because as r :
K() =
a a
ra
ra
er
(3.134)
(3.135)
(3.136)
With long range we mean that in addition to the previously derived exponential
decay we have a much slower algebraic decay. Due to the Bogomolny equation
the contribution of the gauge and the Higgs field to the mass density are equal.
This results in a twice as big density in the tail of the monopole as we had it in
the constructions by t Hooft-Polyakov (for > 0) or Dirac. This should yield
observable consequences if we analyse interactions with a gravitational field.
Furthermore, the long range force exerted by the Higgs field is always attractive
and has been found to be equal in magnitude (see [23]) to the magnetic force
101
law induced by equation (3.12): Oppositely charged monopoles experience an

enforcement of the attraction but in the case of equally charged monopoles the
two forces exactly cancel. We conclude with the remark that the BPS-monopole
shows that the t Hooft-Polyakov construction can differ essentially from the one
given by Dirac and that it is therefore more general.
102
Topology in Physics
Appendix
Field Strength Tensor in the Higgs Vacuum

We obtain this form by straightforward computation:
G = W W + ie [W , W ]
1
1
1
= 2 ( + ) + A + A
ae
a
a
1
1
1
2 ( + ) A A
a e
a
a
1
1
1
1
+ie 2 + A , 2 + A
ae
a
ae
a
2
1
1
= 2 + ( A A ) + ( A A )
a e
a

a

1
1
1
1
+ie 2 , 2 + ie 2 , A
ae
ae
ae
a

1
1
1
1
+ie A , 2 + ie A , A
a
ae
a
a
1
1
2
= 2 + ( A A ) + ( A A )
ae
a
a
1
1
+ 4 ( ) ( ) + 3 ( ) (A )
ae
a
1
+ 3 A ( )
a
2
1
1
= 2 + ( A A ) + ( A A )
ae
a
a
1
1
1
2 ( ) A + A
ae
a
a
1
1
= 2 + ( A A )
ae
a
1
=
F
a
1
(3.137)
with F = 3 ( ) + A A .
ae
Dislocations in crystals
David Gablinger
Supervisor: Urs Aeberhard
In this proseminar talk we deal with crystal dislocations, which
are a sort of topological defect. Basically, they can be divided
into screw and edge dislocations. We will then treat some of their
elastic properties, in particular a general formula for the displacement of dislocations, forces on dislocations and the interaction
between dislocation loops. We will also look at their movement
and discover properties that allow analogies to particle physics.
Dislocations Inside Solid State Physics
At first, it might be interesting to see where dislocations occur inside the vast
world of solid state physics. Dislocations are a type of crystalline defect, so let us
start with an overview of crystalline defects: Basically they can be divided into
five categories, of which two are effects from solid-state physics, so only three are
considered defects in the classical crystallographical sense. Furthermore, they
can be ordered by their dimensionality:
Different Kinds of Defects
0D: Point Defects
Point defects can be divided into defects due to an additional atom in the
lattice, and defects with a missing atom. This can be more complicated by
having a ionic crystal. These defects always occur, due to thermodynamical
reasons: they raise both the inner energy U and entropy S. In the Equilibrium, the free energy F = U T S will be minimized. With S = k ln W , the
N
d
).
concentration of such defects can be calculated as cdef = Ndef = exp( E
kT
This is done in detail for example in [24].
103
104
Topology in Physics
1D: Line Defects (Dislocations)

These are basically the defects we will deal with in this proseminar. There
are different possible defects that go along an axis or line and do not end
inside a crystal:
singularities in liquid crystals
The polarization behavior of liquid crystals permits some experiments
with nice pictures investigating these singularities.
rotation dislocations
These are dislocations where the dislocated part of the crystal gets
bigger with increasing distance to the dislocation line. These dislocations can lead to crystallites with additional faces.
translation dislocations
These are the line defects that occur most often, and furthermore we
will mainly deal with them. In principle, a dislocation is an ordered
lattice mismatch, in the sense that the mismatch continues to some
direction. Translation dislocations shift the lattice by one lattice constant, such that very far away from the dislocation, the lattice looks
exactly like the perfect lattice, but the closer we come to the dislocation line, the more we can see its
2D: Grain Boundaries
Grain boundaries are the borders where two different crystallites of the
same type meet. An example for such a border can be described when there
is a small angle between the two crystallites. In that case, the border can
be made out of a series of dislocations. Then the angle between the two
lattices is = hb , where b is the burgers vector of the edge dislocation and
h is the distance between two neighboring dislocations.
3D: Volume Defect: Thermal Movement
As known from the solid-state physics course, phonons can change physical
properties of a crystal. For example as known from the solid state physics
course, they can lower the reflexes of an x-ray diffraction experiment by the
Debye-Waller factor. In a completely different manner, thermal movement
can also be crucial for the study of movement of dislocations, as for certain
kinds of dislocation motion there is a minimum temperature.
3D: Surface Defect: Finite Crystal size
The finite size of any crystal is an imperfection in the sense that any finite
crystal is not periodical. As we will see, dislocations can not end inside
the crystal, so this means that dislocations either are closed loops, or start
(and end) at a surface.
105
Applications of Dislocations
As dislocations in general, and especially translation dislocations are very physical objects, that is they are not hypothetical in any sense, and not theoretical
constructs, but real-world objects instead, it is interesting to look at applications.
Perhaps the most obvious application of past dislocation research is chip industry. The miniaturization of electronics has made it impossible to use crystals
that have dislocations inside, as they alter the physical properties of the crystal
too much. Today, it is possible to grow crystals without dislocations inside.
Another obvious application is the so-called work hardening, which has been
used throughout at least two thousand years: Iron becomes much harder by
adding a few weight percents of carbon, i.e. it becomes steel. However a rigorous
explanation for this property could only be given in the last century, that is after
studying dynamic properties of dislocations: the carbon in the steel provides
pinning centers for the dislocation, which fixes the dislocation in its position.
From an experimental physics point of view, the whole field of dislocations is
much larger, for example, singularities in liquid crystals are treated as dislocations
too. In that sense there are many phenomena which are well understood today,
but which we will not treat.
Today, another application is, that dislocations provide a physical object to
which different new theories can be applied, that means dislocations should provide an example for a new theory with a mathematically similar behavior. The
most prominent example is maybe the analogy to vortices in a topological treatment of physics.
general assumptions
The study of dislocations can be done very rigorously, down to using quantum
mechanics, lattices and so on. However, before starting the next section, it is
maybe important to note that for the rest of this proseminar, we will restrict
ourselves only to the simplest cases. That means in particular:
In particular, we do not include properties that come from the crystal lattice and its periodicity, this means our medium will be homogeneous and
continuous.
We treat dislocations only inside elasticity theory, that means in a purely
classical theory.
We look at dislocations from a distance, that means we only work within a
linearized theory
We only consider isotropic media
usually assume an infinite crystal size, in particular avoid surfaces whenever
possible.
106
Topology in Physics
Since the reader might not be too familiar with elasticity theory, it is maybe
worthwhile to note some definitions and properties. I have tried to keep the
same notion than Prof. Sigrist in his script about continuum mechanics, so every
reader familiar with that script may omit the next section.
Introduction to Elasticity
Basic Definitions
In the following, we will always assume Einstein convention, unless explicitly
noted otherwise.
displacement and strain tensor
The strain tensor describes a deformation of a medium in the following
sense: Let us describe every position inside the undeformed medium by
using the position vector r. When we start to deform the medium, we need
an additional position dependent vector field that describes the difference
to the original position. We call this vector field the displacement ui .
ri0 = ri + ui (r)
(4.1)
Now we look at the change of distance between two nearby points (r, r+dr).
2
dr0 dr2 = dx0i dx0i dxi dxi

= (dxi + dui )(dxi + dui ) dxi dxi

ui
ui
dxl
dxj dxi dxi
= dxi +
dxi +
xl
xj
ui
ui
ui
ui
= dxi
dxl + dxi
dxj +
dxl
dxj
xl
xj
xl
xj
(4.2)
(4.3)
(4.4)
(4.5)
At this point, we can assume that the deformation and thus the first derivative of the displacement is small, such that the last term drops away. Since
we sum over both j and l, we can rename both to j. Then we get the usual
definition of the strain tensor :

1 ui uj
ij =
+
(4.6)
2 xj
xi
2
dr0 dr2 = 2ij dxi dxj
(4.7)
definition of the stress tensor

Usually a body in an equilibrium situation, in particular without outer
forces acting on it is called undeformed. When deforming the body, there
107
are stresses acting inside, that is there are forces acting between neighboring
volume elements. The Force on such a part of the volume then is:
Z
I
FV =
F dV =
ik dfk
(4.8)
V
where is called stress tensor

deformations and energy
It is now interesting to find an expression for the elastic energy of a deformed
body in terms of the new quantities and , because in the end this will
allow to find a relation between stress and strain. Now let us consider an
infinitesimal deformation:
Z
W =
wdV =
Z
Fi ui dV =
ij
ui dV
xj
(4.9)
After integrating by parts and neglecting the surface term and using that
the strain tensor is symmetric, we get the elastic energy:
Z
1
I=
ij ij dV
(4.10)
2
Similarly, the free energy density functional is
Z
F(u) =
V

I
1
ij ij 0 Fi ui dV
ui Pi dS
|{z}
2
V
|
{z
}
F su
(4.11)
F vol
Where 0 is the mass density. with the inner energy being proportional to
U = T dS I

ij =
U
ij
F = SdT + I
ij =
F
ij
(4.12)

(4.13)
the last equality can be used to derive a relation between stress and strain
relations between stress and strain
Furthermore, we assume a particular free energy, where F0 is the free energy
density without deformation, and and are the Lam
e coefficients:
F = F0 + 2ii + ij ij

2
(4.14)
108
Topology in Physics
If we now apply the above expression for the free energy into the relation
between stress and strain (4.13), we get
ij = 2ij + ij kk
(4.15)
This relation is also known as Hookes law. Formally, this relation can
be written as:
ij = cijlm lm
(4.16)
With c being the modulus of elasticity. For our special case of a homogeneous, isotropic, elastic body, it is
cijlm = ij lm + (il jm + im jl )
(4.17)
Using the above Lame coefficients, we can define Poissons module of

contraction by:
=
(4.18)
2( + )
We can also solve Hooks law for :

1
ij =
ij +
ij kk =
ij
ij kk
(4.19)
2
2(2 + 3)
2
+1
Equations of Motion
In this section, we want to derive the equations of motion for elasticity, because
they will be used throughout the rest of this paper.
Z
dtL = Ld4 x
Z
I

0
2
vol
=
(t ui ) F ) dV dt dtdSF su
2
Z
Z

0
2
=
(t ui ) (ij ij ui 0 Fi ) dV dt + (ui Pi ) dSdt
2
S=
(4.20)
(4.21)
(4.22)
(4.23)
After integration by parts in the first term of (4.22), we have

Z

0
(t ui )2 + ui j ij + ui 0 Fi dV dt
S=
2
With the Euler-Lagrange equations,

L
L

=0
u
( u)
(4.24)
(4.25)
109
we get
L
= j ij + 0 Fi
u

L
L
= t
= 0 t2 ui
( u)
(t u)
(4.26)
(4.27)
So finally, we have the following equations of motion.

0 t2 ui = j ij + 0 Fi
0 t2 ui = ( + ) ( u) + (u) + 0 Fi
(4.28)
(4.29)
The second term in (4.22) could be used to find the boundary condition:
ij nj = Pi
(4.30)
Most of the time, we will not need the full equations of motion (4.29), so there
are simplifications
Note that for the static case, the equations of motion simplify to
0 = ( + ) ( u) + (u) + 0 Fi
(4.31)
and for the case with no external forces acting on the body to
0 = ( + ) ( u) + (u) .
(4.32)
Elastic waves inside the free medium are described by

0 t2 ui = ( + ) ( u) + (u) .
(4.33)
Basic Dislocation Properties
Volterra Construction And General Properties

The Volterra construction is covered in a number of textbooks, such as [25], [26],
[27]. Let us now take a closer look at translation dislocations: Take a hollow
cylinder made of our elastic medium. Then let us cut the cylinder along a surface
parallel to the rotation axis. We can now shift the opened cylinder along the two
cut surfaces in by an arbitrary vector b, which is called Burgers vector. For
this shifting vector, there are two instructive cases:
The vector points upwards, i.e. along the cylinder axis, so the cylinder
looks now like a screw, hence the name screw dislocation. If we look
at the crystal lattice level by level, we can also make the comparison to a
parking lot with rising levels.
110
Topology in Physics
Or we can shift perpendicular to the cylinder axis, so the cylinder looks

now a bit like a slug. This dislocation is an edge dislocation. If we look
at the crystal lattice case, we can construct that dislocation by taking a
perfect lattice, and insert an additional crystal plane.
a)
b)
c)
b
b
Figure 4.1: a) the undeformed cylinder b)an edge dislocation c)a screw dislocation
The hole in the center of the cylinder must contain a singularity of the displacement field u. This construction of a line defect is called Volterra construction.
To see the properties of this Volterra construction more formally, we can trace
the displacement vector u along any path around the line defect l and note this
as
I
I
Z
dui
d2 ui
dui =
dAk = bi
(4.34)
dxk =
kmj
dxj dxm
dxk
S
{z
}
|
dislocation density
Where we applied Stokes theorem. As this path can be arbitrary, the vector
b is a conserved quantity for the singularity, in a way a topological quantum
number. The line defect thus cannot end just somewhere in the crystal, but
instead must either go from surface to surface, or along a grain boundary, or be
closed loops.
We can rewrite the dislocation density, by using that the surface integral
yields the burgers vector, and then write the burgers vector as the result of an
integration over an area delta function. Omitting the integral gives:
kmj
d2 ui
= bi lk () = ik
dxj dxm
(4.35)
with being the radial distance in the surface perpendicular to the dislocation
line. Note that this is an approximation, where we go around the dislocation
111
13a
14a
14a
12a
Figure 4.2: we can now choose an arbitrary path around this dislocation, as long
as we stay away from its core, but there will be one remaining burgers vector
with the length of one lattice constant
with a certain radius greater than the lattice parameter a, i.e. the delta function
is an approximation. With that particular dislocation density, we can find four
properties of the Burgers vector:
b is independent of the cut surface, and of the path chosen around the
dislocation line. With the dislocation density, this can be seen directly out
of the property of the delta function.
b adds like an ordinary vector, that means especially that the Burgers vector
enclosing multiple dislocations is equal to the sum of the Burgers vectors of
the dislocations. These two properties can be seen with the surface integral
over the dislocation density: If we have there a sum of delta functions, the
integration of each delta function gives a burgers vector.
and it also means that at any node, where dislocation lines cross, the sum
of all incoming
Burgers
vectors must equal the sum of all outgoing Burgers
P
P
vectors:
bin =
bout . This can be seen with an argument similar to the
one above.
dislocation lines cannot end inside the crystal, this can be seen with the
fact that the derivation of the dislocation density with respect to xm gives
0.
Now we take a closer look at screw and edge dislocations.
112
Topology in Physics
Displacement of the Screw Dislocation

When we look back to the Volterra construction, we can note that for the screw
dislocation only the z-component of u is non-vanishing, but only dependent of x
and y. The trace of the strain tensor, as well as all diagonal components are zero
for that reason, as can be seen in the figure.
Figure 4.3: cylinder coordinates for the screw dislocation
kk = u = 0
(4.36)
the elastostatic field equation without external forces (4.32)

j ij = 0 = 2j ij + i kk
(4.37)
then becomes
uz
1
0=
= x2 uz + y2 uz =
xj xj
r r

uz
1 2 uz
r
+ 2
r
r d2
(4.38)
together with the condition for the Burgers vector, there is a trivial solution
uz (r, ) =
b
2
uz (x, y) =
y
b
arctan
2
x
(4.39)
Strain and Stress Tensor of Screw Dislocations

With the definition of the strain tensor (4.6) we have

b
b sin
1 ux uz
y
+
=
=
zx =
2
2
2 |{z}
z
x
4 x + y
4 r
=0
(4.40)
zy
113
b
1 uy uz
+
=
=
2 |{z}
z
y
4
x
2
x + y2

=
b cos
4 r
(4.41)
=0
with ij = 2ij + ij kk (4.15) and because all kk are zero, the stress tensor
ij = 2ij has the same two components. If we change coordinates, only z is
non-zero:
z =
b 1
4 r
(4.42)
Stress Tensor of Edge Dislocations

Now: Burgers vector along x or y-direction. If we look back to the Volterra
i
construction, then uz = 0, and also u
= 0 , that is we have a plane deformation.
z
In such a plane deformation. Then the only stresses we need to determine are
the normal stresses xx , yy , and xy = yx . Note also that zz = (xx + yy ),
from the relations between stress and strain.
Now we define an additional function , by
xx =
2
,
y 2
yy =
2
,
x2
xy =
2
xy
We can also transform this definition to cylindrical coordinates:

1
1 2
2
1
rr =
+ 2 2 , =
, r =
r r
r
r2
r r
(4.43)
(4.44)
Now we look again with the static equation (4.32) ,

j ij = 0 = 2j ij + i kk
(4.45)
With our definition, we can note that the static equation for a plane deformation
is satisfied by any solution for the biharmonic equation
4 = 0
(4.46)
We can now make an ansatz of separation of variables,

= R(r)()
(4.47)
Then we can solve the resulting differential equations by trying first to find a
solution for , where for example a sin is a possible solution. Then we can
solve the differential equation for R(r) and see that one solution is
0 =
1
b
b
r ln r sin =
y ln(x2 + y 2 ) 2
2(1 )
2(1 )
(4.48)
114
Topology in Physics
Now we can insert this solution back into the definition for and get the stress
tensor elements:
Note that D =
b
.
2(1)
sin
r
sin
=D
r
rr = D
(4.49)
(4.50)
cos
r
sin
= 2D
r
r = r = D
(4.51)
zz
(4.52)
In Cartesian coordinates, becomes:

y(3x2 + y 2 )
(x2 + y 2 )2
y(x2 y 2 )
=D 2
(x + y 2 )2
x(x2 y 2 )
= yx = D 2
(x + y 2 )2
y
= 2D 2
(x + y 2 )
xx = D
(4.53)
yy
(4.54)
xy
zz
(4.55)
(4.56)
Displacement of Edge Dislocations

From the stress tensor elements, we can get the strain with = 2 , and then
the displacement with ij = 12 (i uj + j ui ) where uz is zero:

b
sin 2
ux =
+
(4.57)
2
4 (1 )

b
1 2
cos 2
uy =
ln r +
(4.58)
2 2 (1 )
4 (1 )
uz = 0
(4.59)
Dislocations and Vortices
Here we will look for a sort of a formal relation between dislocations and classical
vortices.
vortices
As can be seen in a textbook about fluid dynamics, the circulation around a
point in the medium is given by
I
ZZ
ZZ
=
vdl =
( v) df =
df
(4.60)
S
115
where v is the velocity field. Consider the case of a vortex with v =

then
I
I
=
vdl = rd
=
2r
S
1
2r2
( r),
(4.61)
dislocations
ui
Formally, we can set up the same equations for dislocations ( note that ij := x
)
j
Z
ZZ
ZZ
b=
dl =
( ) df =
df
(4.62)
dC
For a screw dislocation, the analogy can be seen quite directly: take the
b
, which we calculated previously, then zx = b x2y
and
displacement uz = 2
+y 2
x
zy = b x2 +y
Then
formally,
can
be
written
in
the
following
form
2

0
x
1
0 y
zi =
2(x2 + y 2 )
bz
z
From that we can see that the Burgers vector b is similar to the circulation
of a vortex. Note that this is only valid for screw dislocations, i.e. dislocations
have tensorial quantities, where vortices have vectorial ones.
=
1
(b r)
2r2
(4.63)
Static Properties of Dislocations
The General Formula for the Displacement of a Dislocation

We will now look at a more general formula for the displacement which is valid
for a closed dislocation loop.
dislocation density, strain tensor
First we look again at the and tensors which are the dislocation density
and the displacement derivative with respect to xj . These definitions allow
to formulate the strain tensor in terms of the dislocation density.
As seen from the figures for the Volterra construction, there is a discontinuity of the displacement at some point, i.e. it suddenly jumps from
one value to another value, like a step function, with the difference between
the two values equal to the Burgers vector, i.e. u (x). If we then
are interested in the local strain at the discontinuity, we get the following
expression:
1
(nl bm + nm bl )() = lm
(4.64)
2
116
Topology in Physics
where is a coordinate taken from the surface S along the normal to the
surface. We will use that later.
static equation
We start with a static equation, i.e. Strain must be proportional to applied
force. In the end, this gives nothing else but the already known static
equations of motion (4.31), this time formulated using the cijlm tensor :
Fi =
lm
1
ij
= ciklm
= ciklm (k m ul + k l um )
xk
xk
2
(4.65)
this equation can be solved for the displacement using Greens formalism
with the fundamental equation
(x) = ciklm (k l Gim + k m Gil )
(4.66)
where ci jklm is the modulus of elasticity, where we are interested in the

isotropic case. with the Greens tensor, u comes out of the convolution of
G with F :
Z
ui (r) =
Gik (r r0 )Fk d3 r
The Greens tensor for the isotropic case is then

2ik
1
+ 2r
Gik (r) =
,
8
r
+ 2 xi xk
p
with r = |~r| = x2 + y 2 + z 2 .
To justify this, consider the derivatives of r:
r =
2
r
and
1
= 4
r
(4.67)
(4.68)
(4.69)
the displacement
Now we plug into the convolution of Greens tensor with the force again the
other side of the static equation. We use the discussed dislocation strain at
the discontinuity to get an explicit expression for the force:
Z
ui (r) = Gij (r r0 ) cjklm k lm d3 r
(4.70)
Z
1
= Gij (r r0 ) cjklm k (ll bm + lm bl )()d3 r
(4.71)
2
Z
(4.72)
= cjklm bm ll k Gij (r r0 )dS 0
117
Here, we insert the exact Greens tensor and write out the isotropic cjklm
and get six terms. By using Stokes theorem and (4.69), one obtains the
following expression:
I
I
b
b
dr0
1
dr0
u(r) =
+
+
(b (r r0 ))
(4.73)
4 4
8(1 )
L
L
with
= |r r0 |
(4.74)
u(r)
dislocation line
Figure 4.4: the displacement at a point r distant from the dislocation loop
simple example
We can now check whether this formula makes sense with the simplest
example, that is the screw dislocation: assume the dislocation parallel to
the z-axis, with the burgers vector in the same direction. Then the formula
should yield the already known displacement in z-direction:
I
I
bi
1
dl
bz
dlj
+ zij
+
z (ijk bj (rk lk ))
uz (r) =
4
4 L
8(1 )
|
{z
}
|L H
{z
}
zzi =0, only bz 6=0
:=
dl
L (izk bz (rk lk ))
(4.75)
is 6= 0 only for i 6= z , so z yields 0
uz =
bz
bz
=
4
2
(4.76)
118
Topology in Physics
Energy of Screw Dislocations
It requires energy to create dislocations, because a dislocation is a permanent displacement of the crystal lattice. If we assume the crystal size and
the dislocation length to be infinite, then the elastic energy of the dislocation will be divergent, analogous to the divergence of global vortex energy.
However, it is possible and meaningful to define an energy per length.
From elasticity we have the following relations:
Z
1
E=
ij ij dV
2
Z
E
1
1
=
( ij ij
ii kk )dS
l
2
2
2(2 + 3)
:= Eall Ediag
(4.77)
(4.78)
(4.79)
= Use this to calculate the elastic energy of screw/edge dislocation

For the screw dislocation, all diagonal entries of the stress must be zero, as
discussed above, so the diagonal energy part vanishes.
Ediag = 0
(4.80)
With the note that in cylindrical coordinates, z 6= 0

Eall
Z r

1
2
2
=
3
+ 3
2rdr
2 r0

Z r 2 2
b
1
2 b 2
=
+
2rdr
2 r0 (4)2 r2 (4)2 r2

b2
r
E
=
ln
+0=
4
r0
l
(4.81)
(4.82)
(4.83)
where r0 is the core radius

Energy of Edge Dislocations
With the calculated stress and strain from the previous sections, we can calculate
the energy of an edge dislocation:
Z
1
E
=
(rr rr + + 2r r ) rdrd
(4.84)
l
2

r
2
(1 )
= D ln
(4.85)
r0
So the expression shows the same logarithmic dependence on the radius!
119
Forces Acting on Dislocations

We want to look at the force on a dislocation in an external stress field. This
treatment was first given by [28], and is also covered in many books, for example
in [29]. Then the interaction energy of the dislocation loop is as given by (4.9)
. However in our case, we look at an infinitesimal translation of the dislocation
loop, so then with xijj = Fj , and integration by parts (omitting surface terms),
we get
I
Z
I =
Fi xi dl =
ik uik dS
(4.86)
L
Using Gauss law, one can transform the surface integral with differential
element Si to a volume integral.
Z
Z
i (ik uik ) dV
ik uik dSi =
S
(4.87)
Z
i ik uik dV +
V
ik ik dV
(4.88)
Since we assumed an external, constant stress field, i.e. i ik = 0, the first term
vanishes.
For an infinitesimal displacement of an element of the dislocation line, the
change of the according surface area is
Si = ikm xk tm dl
(4.89)
With the note that the inelastic volume change becomes

kk = xi ilm bl tm ()dV
kk dV ilm bl tm xi dl
(4.90)
(4.91)
Then we can write the following

Fi = ilm tl mk bk
(4.92)
Note that the formula we just derived is analogous to the Biot-Savart formula
from electrodynamics, where the Force on a conductor is :
dFi = Iijk dlj Bk
(4.93)
Note however that our case is more complex, as the stress field which corresponds to the magnetic field B is a tensor, and the burgers vector corresponds
to the (scalar) current I.
120
Topology in Physics
Mirror Charges and Dislocations

In this section we will try to see what happens when we give up on the idea of an
infinite medium, meaning we now have an elastic medium at the lower half and
a vacuum at the upper half with one boundary plane.
x=0
x0
y
z
Figure 4.5: the dislocation and its mirrored version

For that we take the case of a screw dislocation parallel to the surface. The
boundary condition of a strain free surface should still be satisfied, so Pi =
ij nj = 0, where n is a vector normal to the surface. For simplicity, we assume the
following coordinates: the boundary between free space and our elastic medium
is the x = 0 plane, with the dislocation being located at x = x0 and y = 0,
the line being along the z-axis. The condition of a stress free surface is then
xz x=0 = 0. The stress induced by the dislocation is:

y
(x + x0 )
xz =
yz =
(4.94)
(x + x0 )2 + y 2
(x + x0 )2 + y 2
By introducing a mirror dislocation with opposite sign, we can gain a strain free
surface:

b
y
y

Px =
(4.95)
=0
2
2
2
2
2 (x + x0 ) + y
(x x0 ) + y
x
However, Py 6= 0, and thus there is an attractive net force on the dislocation,
i.e. the dislocation is forced to the surface because of the mirror dislocation:
Fi = ilm tl mk bk = izm tz mz bz

b1 b2 x0
Fx = tz yz bz =
4pi
x20
(4.96)
(4.97)
121
Interaction Between Dislocation Loops

simple example
We can first look at a simple example, that is the interaction energy between two
parallel screw dislocations. We start by assuming two screw dislocations along
the z-axis, and calculate the force the two exert on each other:
0
0
Fi = ilm tl mk
bk = izm tz mz
bz
(4.98)
Then we assume both of them are at y = const. So x-component of the force

is:
0
Fx = xzy tz yz
bz
(4.99)
Now we are only interested in the x-coordinate, so we transform to radial

coordinates,
b2 1
Fx = b1
(4.100)
4 r
R
The interaction energy is then obtained by using F dx = I, and it gives with
the core radius r0
b1 b2
r
I=
ln( )
(4.101)
4
r0
Now we want to look at a more general formula for the interaction energy of
two dislocation loops.
additional definitions
Before starting with the derivation, we need to note some additional formalisms
we denote Kroeners incompatibility tensor as
ik = ilm kpq l p mq
(4.102)
similarly, Kroeners stress function is defined by

ik = ilm kpq l p mq
(4.103)
we can define the modified stress function 0

20ik = ik
+2
nn ik
(4.104)
then the biharmonic equation holds

4 0 =
(4.105)
122
Topology in Physics
the Greens function for the biharmonic equation is

Z
1
0
=
(r0 )|r r0 |d3 r0
8
(4.106)
for our usual dislocations, the Greens function becomes, when inserting
the exact
Z
1
0
=
bl m |r r0 |(ilm dlk + klm dli )
(4.107)
16
Lastly, the stress function for such a dislocation then is

1
0
0
0
ik = 2 ik +
(i k nn ik nn )
1
(4.108)
Interaction between two loops Let us start with two dislocation loops s1 and
s2 .
Their interaction energy then can be obtained by looking at the stress field
generated by one loop acting on the other:
Z
I=
b1 s2 dS1
(4.109)
s1
Now we insert the solution for the stress function in terms of 0

Z
0nn
1
0
0
I = b1 ik +
ik nn dS1
1 xi xk
(4.110)
Inserting the Greens function for 0 into this expression again gives four
terms, which by using Stokes theorem give Blins formula:
2
I=
4
dl1 dl2
Z Z
(b1 dl1 )(b2 dl2 )

+
4 L1 L2
Z Z
(b1 dl1 )(b2 dl2 ) (4.111)
4(1 ) L1 L2
(b1 b2 )
L1
L2
with
= |l1 l2 |
(4.112)
123
Figure 4.6: a dislocation gliding along a plane needs to overcome a periodic stress
Dynamic Properties of Dislocations
Introductory note: As a real translation dislocation occurs only in real crystals,

that is within an atomic lattice, the motion of dislocations is step-wise in the
sense that the lattice periodicity provides also a stress field which the dislocation
needs to overcome. the full treatment of dislocation motion requires integral
equations, because:
Consider Peierls Nabarro model: gliding dislocation needs to overcome periodic stress generated by the lattice, the easiest possible function is of course
a sinus.
Z
uz
1 duz (, y)
d
=
=

y y=const
x
d
y=const
b
4uz (x)
sin
2a
b

zy (x)
(4.113)
(4.114)
Note that this is an approximation outside of our purely elastic approach!

Motion of Dislocations
If we stay in our narrow set of assumptions, we can note that the fast movement
of dislocations resembles the relativistic movement of a massive particle. For that
we start with the equation of motion (4.29) for elasticity, which describe both
and note that dislocations move within the same framework
relativistic motion
0 t2 ui = ( + ) ( u) + (u) + 0 Fi
If we now consider the motion of a screw dislocation, then
2

2
1 2
+
+
uz = 0
x2 y 2 c2 t2
with c =
(4.115)
124
Topology in Physics
transforming the equation with x0 = (x vt) yields again the equation for
the static case
x20 u0 + y20 u0 = 0
(4.116)
the solution for the screw dislocation is still applicable

b
uz =
arctan
2
0
y
x0
(4.117)
transformation to original coordinates gives that the displacement of the

moving dislocation becomes narrower (as to be expected)
b
arctan
uz (x, y, t) =
2
y
(x vt)

(4.118)
Note for the next section: Solution can of course be built using elementary
solutions of the form
uz = exp (i(x vt) iy)
(4.119)
where := i 1 = ( vc2 1)1/2 and i = (1 2 )1/2

mass of a dislocation If we look at dislocations analogous to particles, then
the energy required to build a dislocation could be seen as its mass, similarly. As
the energy of a moving dislocation increases, we see that the analogy to particle
physics is still valid, as the mass of the dislocation increases by a factor of .
"
= E0
2
2 #

uz
uz
b2
r
+
=
ln
E0 = Eelast,0 = dV
x
y
2
r0
"
2
2
2 #
Z
uz
uz
uz
Etot = Eelast,mov + Ekin = dxdy
+
+
x
y
t
Z
(4.120)
(4.121)
(4.122)
With E0 = m0 c2 , the dislocation can be considered as having a mass!

Supersonic Dislocations
Supersonic dislocations are still being researched, both from an experimental and
a theoretical point of view,for example [30], [31]. In this section, we will follow
one of the first papers on the topic from J.D. Eshelby who treated the case of
125
a screw dislocation moving faster than the speed of elastic waves in a dispersive
medium in [32].
When the dislocation is faster than sound, the wave equation becomes hyper2
bolic (denote := i 1 = ( vc2 1)1/2 )
2 uz
2 uz
2
=0
y 2
(.x vt)2
(4.123)
Note again that for screw dislocations, zy = 2zy = z uy + y uz = y uz

Now the hyperbolic wave equation together with zy gives a relation that makes
Peierls-Nabarro equation unnecessary:
zy =
uz
uz
= 0
y
x
(4.124)
Now we look at a dispersive medium, i. e. c(k) = (k)/k, and look at the

Fourier transform of the displacement:
Z
uz (k) exp (ik(x vt)) dk
(4.125)
uz =
The stress can also be Fourier transformed and be expressed in terms of the
displacement.
dz
du
(k) = vik ubz (k)
dt
c
bz (k)
zy (k) = (k)ki(k)u
(4.126)
(4.127)
Then we have an energy flux density out of the slip plane, namely
Z
E = 2
duz
zy
dx = 4
dt
Z
zy (k)
z
du
(k)dk
dt
(4.128)
Where the last equation comes with Percevals theorem.

Using that the lattice provides a maximum k = km , and the above Fourier
transforms, we can simplify the energy flux density integral:
Z km
E = 8v
(k)k 2 (k)uz (k)uz (k)dk
(4.129)
k0
When we now assume additionally

that the subsonic components vanish
and that we are only slightly above vm , so is the only quantity over which
integration is performed, all other values are replaced by km
126
Topology in Physics
furthermore, can be expressed in terms of (k)/k ; using that

1
2 2 (k k 0 ) 2
= (v/c(k) + 1) (v/c(k) 1)
km
1
2
1
2
(4.130)
These properties then give

2
E = (1621/2 /3)v(km )km
ubz (km )ubz (km )
(km k 0 )3/2
(4.131)
3/2
km
When we now assume that a solution to the wave equation for uz is still
proportional to
b
uz =
arctan(x/)
(4.132)
2
then we have the following identities
1. due to the proportionality of stress and strain, and the simple form of the
strain for a screw dislocation, together with the Fourier transform property
of the derivative, we have that
c
yz (k) = k uz (k)
2. note that the Fourier transform of is symmetric
c
c
yz (k) =
yz (k)
3. c
yz (k) =
b
2
exp(|k|)
This gives then approximately with km =
and (km ) =
4
2
v 3/2
) exp(2 )
(4.133)
vm
b
Which means, this is the stress required to keep the dislocation in motion,
otherwise it gets damped. Note that this effect bears some analogy to Cherenkov
radiation!
a (
Conclusion
We have now seen in this proseminar both some static properties of dislocations
including general formulas for displacement and interaction energy, and some
dynamic properties including an analogy to Cherenkov radiation. What remains
to say, is what these statements are good for:
First of all, we notice, that we have only done the simplest elastic treatment.
There are many ways to build up more complicated, improved models, and it is
still possible to do active research.
127
Second, we have seen that dislocations are experimentally accessible, so if

there is any new model that can be applied to dislocations, then in principle it
can also be verified in experiments.
And third, we have seen that it is actually possible to make analogies between
dislocations and other fields of physics.
128
Topology in Physics
Solitons
S
ami Lang
Supervisor: Ilka Brunner
This text is about solitons. Solitons are particle-like objects occurring in classical field theories. There is a special focus on
kinks, which are solitons corresponding to one space dimension.
The first chapter will give a general introduction on solitons,
while the second chapter is mostly concerned with the topological classification and the existence of solitons. In chapter 3 explicit kink solutions will be discussed, namely the 4 kinks and
the sine-Gordon kinks.
Introduction
What is a soliton?
Before we start to investigate solitons, we must clarify what a soliton actually
is. There is no commonly used definition of a soliton, some authors require
more features for an object to be called a soliton and some less. Nevertheless,
in this text we shall have a clear notion of what we call a soliton. There are
classical field theories, which have interesting, particle-like solutions to their fully
nonlinear field equations. By particle-like we mean that these solutions are
smooth-structured and stable, and that they have a finite mass as well as an
energy density, which is localized on some finite region of space. But the real distinctive property of these new particles is, that the have a topological structure
that differs from the vacuum. The topological character of a field configuration is
often captured by a simple integer N , called the topological charge. This topological charge remains constant under time evolution, since time evolution is assumed
to be an example for a continuous deformation and an integer simply cant change
under a continuous deformation. This really is the reason for the stability of the
new particles. Although they are often of large energy, they remain stable, since
they cannot change their topological charge. The integer N can be interpreted
129
130
Topology in Physics
as the net number of new particles, with the energy increasing as |N | increases.
The vacuum has N = 0. The minimal energy field configuration with N = 1 is
particle-like in the sense we mentioned above (if its energy density is localized).
It is called a topological soliton or just soliton. The analogue configuration with
N = 1 is an antisoliton (usually a reflection symmetry reverses the sign of N ).
Soliton-antisoliton pairs can annihilate or be pair-produced. Field configurations
with N > 1 are interpreted as multi-soliton states. Depending on what is energetically more favourable such a state can relax to a N -soliton bound-state or
decay into N well separated charge 1 solitons.
The depending of the interaction energy of two well separated solitons on
their separation is completely determined by the linearized, asymptotic field of
the solitons and is therefore usually rather simple. The force between two solitons
is identified with the derivative of the interaction energy with respect to the
separation.
Static soliton solutions
As we mentioned, solitons are solutions of the fully nonlinear field equations of a
theory. These nonlinear partial differential equations (PDEs) are of second order
and in general very hard to solve. But Bogomolny showed, that in many theories
it is possible to reduce the second order PDEs to first order PDEs. These first
order PDEs are generally called Bogomolny equations. Bogomolny equations
never involve time derivatives and hence their solutions are static soliton or multisoliton configurations.
Bogomolny found out that in these theories the energy of a field configuration
is bounded from below by a numerical multiple of the modulus of the topological
charge N , with equality if the field satisfies the Bogomolny equation. Therefore solutions of the Bogomolny equation minimize the energy within a given
topological class of fields (defined by N ) and are indeed (static) solitons in our
sense.
As we will see, kinks in one space dimension are solutions of a Bogomolny
equation.
Soliton dynamics
We have already stated, that there is interaction between solitons. It is therefore
important to understand this interaction and the dynamics of solitons. If a theory
is relativistic, a static soliton solution can be boosted to move with an velocity
less than the speed of light.
If solitons are well separated they can be treated as point-like objects carrying charges (or even more complicated inner structure) and there are usually
parameters that can be naturally interpreted as the locations of the solitons. The
forces between the solitons can directly be calculated by considering integrals of
Solitons
131
the energy-momentum tensor. Given these forces, one can compute the relative
motion of the solitons and interpret the result in terms of charges.
But if the solitons come closer together the picture of them being point-like
objects breaks down. The treatment of the scattering can become very complicated and only numerical methods can help.
Classification and existence of solitons
In this section we will make some rather general statements on solitons. Admittedly we will restrict ourselves mostly to the case of one space dimension, since
our main task in the following sections is to discuss kinks, which are objects corresponding to one space dimension. We shall first discuss the topological structure
and classification of solitons. As a consequence of these considerations, there will
be some conditions on the topological structure of a particular Lagrangian field
theory for solitons to exist. After that we will present another statement on the
existence or non-existence of solitons, the so called Derricks scaling argument.
For the purpose of classification of solitons, we have to review some notions
of topology. Thats what we do next.
Some topology and notation
For our discussion, we will need the following definitions.
Definition 1. Let X and Y be two manifolds without boundary. Consider
continuous maps 1 , 2 : X Y between them. 1 is said to be homotopic
: X [0, 1] Y with
to 2 , 1 ' 2 , if there exist a continuous map
=0 = 1 and |
=1 = 2 .
is
parameterizing the interval [0,1], such that |
called a homotopy between 1 and 2 .
Definition 2. Let A X be a submanifold. Let 1 , 2 : X Y be
continuous maps with 1 |A = 2 |A . A homotopy relative to A between 1 and
between 1 and 2 with the further property: a A, t
2 is a homotopy
t) = 1 (a) = 2 (a). One writes: 1 ' 2 rel. A.
[0, 1] : (a,
Often A consists only of a single point, say x0 . We call continuous maps
from X to Y , which send x0 to some fixed point y0 in Y , (x0 ) = y0 , based
maps. Homotopy is an equivalence relation and therefore maps can be classified
into homotopy classes. A case of interest is when X is a sphere. The n-sphere
S n is the set of points in Rn+1 at unit distance from the origin. The north pole
(NP) is the point (0, ..., 1) S n . For n 1, we denote the set of homotopy
classes rel. NP of based maps f : S n Y , which send NP to some fixed y0 Y ,
by n (Y, y0 ). One can define a composition for elements of 1 (Y, y0 ) and verify
that this defines a group structure on 1 (Y, y0 ). The composition of equivalence
classes is just the composition of paths for some representatives of the equivalence
classes. The group 1 (Y, y0 ) is called the fundamental group of Y with base point
y0 . One can show that 1 (S 1 , y0 ) = Z.
132
Topology in Physics
The set of homotopy classes of (unbased) maps f : S n Y we denote by

n (Y ). For n = 0 we regard S 0 as consisting of just one single point (not of
the two points 1). Thus the classes 0 (Y ) are then maps from a single point
to Y , up to homotopy equivalence. Since maps with image points in the same
connected component of Y are homotopic, 0 (Y ) is the set of distinct connected
components of Y .
Topology and solitons
Here we want to elucidate the topological aspects of scalar fields defined on flat
space Rd (mostly d = 1). In the last section we introduced the notion of homotpy.
Homotopy theory is one of the two basic techniques for classifying solitons in
theories with scalar fields. The other is topological degree theory, which we will
not discuss. Homotopy theory is more general and more distinctive, on the other
hand topological degree theory is a more refined tool. Also the topological charge
of a field configuration (which in the end is the relevant physical quantity) is more
closely linked to the topological degree of the field than to its homotopy class.
This is because the topological charge often is nothing but the topological degree
of the field. But in the examples, that we will discuss in chapter 3, there is always
a physically natural way to define the topological charge of a field, if one knows
in which homotopy class it lies.
One essential point in the topological classification of fields is, that well have
to combine topological and energetic considerations in order to make some interesting statements. For example it is not sufficient, just to assume the fields
are continuous. If there are no further restrictions, linear fields are topologically
trivial: Any field configuration (x) can be replaced by (1 )(x) and if runs
from 0 to 1 this is a homotopy taking (x) to the trivial field, 0.
The nonlinear case, where is a map from Rd to a manifold Y , is a bit different. Since Rd is contradictable to a point, every field configuration is homotopic
to a constant map: ((1 )x) is a homotopy taking (x) to the constant map
(0). Therefore field configurations are classified by 0 (Y ).
But if we require that the energy density is zero at infinity (note that this
condition is necessary to have finite energy but not sufficient) the topological
classification becomes more interesting (at least for linear fields).
Consider a multiplet of n scalar fields = (1 , ..., n ) with an energy functional of the form
Z
1
E=
l l + U (1 , ..., n ) dd x.
(5.1)
2
If the fields are time independent, then E is the total energy. Assume the potential
U (1 , ..., n ) has minimal value Umin = 0 (this can in general be arranged by
adding a constant to U , which doesnt alter the field equations). The set V Rn ,
where U takes its minimal value, is called the vacuum manifold of the theory.
Solitons
133
At spatial infinity must take its values in V, in order to have zero energy
density there. The values may be different in different directions. Thus a field
configuration defines a map
: S d1 V
e 7 lim (
e).
In a linear theory we lose no topological information, if we only keep track of

these asymptotic data: Two field configurations with the same asymptotic data
are homotopic. Moreover, field configurations , with the asymptotic data ,
, are still homotopic if is homotopic to . Therefore the topological
character of a field configuration (x) is captured the homotopy class of ,
which is an element of d1 (V).
As we mentioned before, we are interested in the case d = 1. Here S d1 =
{1} R. Hence is a map from two points to V. The set of topologically
distinct vacua is the set of distinct connected components of V, which we denoted
by 0 (V). Since 0 (V) is the set of topologically distinct vacua, an element v1
of 0 (V) is a homotopy class of vacuum solutions. But sometimes we will write
1 0 (V), where 1 is a vacuum configuration and identify the homotopy class
with its representative. The topological class of a field configuration (x) is
defined by an element (v1 , v2 ) of 0 (V) 0 (V). If, for example, V consists of p
points in Rn then there are p2 topologically different types of fields. If v1 = v2
then the field lies in the class of the vacuum solution, v1 . If v1 6= v2 then
the field is said to be kink-like and interpolates between different vacua v1 (at
) and v2 (at +). If there is only one scalar field in the theory (i.e n = 1)
then the topological solitons (if they exist) are called kinks. We will see explicit
kink-solutions in chapter 3.
Now lets turn to nonlinear scalar fields : R Y , where Y is a closed
manifold. A topological soliton on R, taking its values on a closed manifold Y ,
is called a nonlinear kink. There may be a potential U (), taking its minimal
value on some submanifold V Y . We still require that () V, but in
the nonlinear case its not useful to look at the homotopy class of the map ,
since it does not fully determine the topological character of the field configuration (x). Nevertheless, if we insist that (+) = () = y0 , or if the
potential V consists only of the point y0 , then is a constant map with value
y0 . This boundary condition allows a topological compactification of space R to
S 1 . We add a single point at spatial infinity and identify it with NP (this can be
achieved by stereographic projection). : R Y then extends to a continuous
map : S 1 Y , taking NP to y0 . The topological class of is therefore an
element of 1 (Y, y0 ). But if (+) 6= (), then the topological classification
is a bit more complicated. As we shall see, the sine-Gordon kink has also an
interpretation as a nonlinear kink: Usually the field is regarded as linear and the
potential U as periodic. But it is also possible to force the field to take its values
on a circle and then the potential has unique minimum on this circle.
134
Topology in Physics
If d 2, the topological classification of nonlinear scalar fields is a bit different.

One requires finite energy of the fields in order to make a reasonable classification.
But we will not discuss this here.
In summary we have the following: Possible (linear) kinks are classified by
an element of 0 (V) 0 (V), where V is the vacuum manifold of the theory, and
possible nonlinear kinks are classified by an element of 1 (Y, y0 ), where Y is the
target manifold of the theory. It is therefore clear, that for topological solitons
to exist, these sets have to consist of more than one element.
Derricks theorem
In this section we consider only static field configurations with finite energy.
Weve seen that such configurations are characterized by their homotopy class.
The vacuum solution, which is spatially constant and minimizes the energy within
all possible fields, lies in the trivial class. But since we are interested in solitons,
we are interested in fields, that minimize the energy within a given homotopy
class, other than the trivial one. Such minima are usually stable solitons. What
we would like to know is: Are there any minima of the energy within other
homotopy classes than the trivial one? This question is hard to answer but more
generally we can ask: Are there any stationary points of the energy other than
the vacuum?
A simple non-existence theorem is the so called Derricks theorem. It applies
for theories defined in Rd . Derrick made the observation that in many theories,
the energy functional for static fields has the following property: If you take
any non-vacuum field configuration, subject it to a spatial rescaling (see below)
and compute the variation of the energy with respect to this rescaling, then it
will never be zero. But since a stationary point of the energy must be stationary
against all variations, including spatial rescalings, such theories can have no static
finite energy solutions in any homotopy class other than the trivial one.
To be more precise: In Rd a spatial rescaling is a map x 7 x, with > 0.
Let (x) be a finite energy field configuration, with any kind of fields, or
multiplet of fields, and let () (x), 0 < < , be the 1-parameter family of field
configurations obtained from (x), if one applies the map x 7 x. We shall
clarify how () (x) is related to (x) below. Let
e() = E(() )
(5.2)
denote the energy of the field configuration () (x), as a function of . Then we

have Derricks theorem:
Suppose that for an arbitrary, finite energy field configuration (x), which is
not the vacuum, the function e() has no stationary point. Then the theory has
no static solutions of the field equation with finite energy, other than the vacuum.
The theorem is only useful, if we define () in an appropriate way so that it
Solitons
135
is easy to compute e(). For a scalar field configuration (x) one simply defines
() (x) = (x)
(5.3)
For a vector field V one defines

V() (x) =
1
V(x).
(5.4)
Note that the rescaling is done in an appropriate way, depending on the geometrical meaning of the fields. In the same sense all other kinds of fields are rescaled.
Furthermore, under these rescaling rules the boundary condition V at spatial
infinity is preserved; also if is a nonlinear scalar field, the rescaling is consistent:
if Y , then Y .
Assume we are dealing with a theory involving just a scalar field with an
energy of the form
Z
E() =
W () + U () dd x
(5.5)
E2 + E0
where we have decomposed the energy into its component parts. The explicit
powers of , occurring when the integrand is rescaled, are indicated by the subscripts. Then
Z
e() = E( ) =
W ( ) + U ( ) dd x
Z
=
2 W ((x))()(x) ()(x) + U ((x)) dd x
= 2d E2 + d E0 ,
(5.6)
where the last step follows by a change of variables from x to x. Therefore e()
is a simple algebraic function of , with coefficients E2 and E0 , that depend on
the initial choice of a field configuration (x).
E2 and E0 are in general both positive. In this case the characteristics of e()
are widely determined by the spatial dimension d. If d = 3 or d = 2,
(
1
E2 + 13 E0
d=3
(5.7)
e() =
1
E2 + 2 E0
d=2
so e() decreases monotonically as increases. There is no stationary point and
therefore there exist no non-vacuum solutions of the field equation with finite
energy. In other terms, there exist no finite energy solitons. If d = 1,
e() = E2 +
1
E0 ,
(5.8)
136
Topology in Physics
p
which is stationary at = E0 /E2 . So in this case finite energy soliton solutions
are not ruled out. Thus, in a purely scalar theory, that has an energy of the form
(5.5), finite energy solitons are only possible in one dimension, but not in higher
dimensions. In chapter 3 we will see, that in one dimension finite energy solitons
are not only possible, but do indeed exist.
Note that for the vacuum E2 = E0 = 0 and therefore the vacuum evades
Derricks theorem in all dimensions.
Another way to make use of the condition that the energy of a solution is
stationary under rescaling, is the following. Suppose that d = 1, and let (x)
be a solution of the field equation of a theory with an energy of the form (5.5).
Then
1
e() = E2 + E0 ,
(5.9)
so
de
1
(5.10)
= E2 2 E0 .
d
Since (x) is a solution of the field equation, the derivative must be zero at = 1.
Therefore E2 must equal E0 , and that means that if the gradient term and the
potential term are integrated over R, they each contribute half of the total energy.
This relation is called a virial theorem.
Kinks
Bogomolny bounds
The simplest examples of topological solitons occur in one space dimension and
involve a single real valued scalar field (x, t). Consider the Lagrangian density
L = U (),
(5.11)
where U is a real valued non-negative function of . The field equation, that

follows from this density is
+
dU
=0
d
(5.12)
Let Umin be the global minimum of U , which we will from now on assume to
be equal to zero, Umin = 0. As before V will denote the set of constant vacuum
fields, i.e. V = {0 R, such that U (0 ) = Umin = 0}. The potential energy is
given by
Z
1 02
+ U () dx,
(5.13)
V =
2
where 0 is the gradient of the field, 0 =
. The kinetic energy is
x
Z
1 2
T =
dx,
2
(5.14)
Solitons
137
where is the time derivative of the field, =

. The total energy is T + V .
t
We know from the last chapter, that finite energy field configurations of such
a theory are classified by an element ( , + ) of 0 (V) 0 (V), where =
limx7 (x). We also know, that in order to have solitons, 0 (V) needs to be
non-trivial. That is the case, if we assume V to consist of several isolated points.
Solutions, which interpolate between different vacua, that is + 6= , are called
kinks. Kinks lie in different homotopy classes than the vacuum solutions, and
therefore they cannot be continuously deformed to a constant zero energy solution
by deformations which keep the energy finite, i.e. deformations which preserve
the boundary condition () V. This really is, where the stability of a
kink solution comes from, because time evolution is an example for a continuous
deformation, that keeps the energy finite.
Recall from the last section, that we are now in a situation (d = 1, combination
in V of a potential term and a term quadratic in the field gradient), where
Derricks theorem allows static soliton solutions. Also, for these solutions the
virial theorem
Z
Z
1 02
dx =
U () dx
(5.15)
2
holds.
We will now show that it is possible to derive a lower bound on the energy
E of any field configuration with the bound only depending on ( , + ). This
means, that the bound is actually given in terms of solely topological data. We
start with the simple inequality
p
1
( 0 U ())2 0.
2
(5.16)
Expanding this on both sides and integrating over space, we obtain

Z
Z p
1 02
2U ()0 dx.
+ U () dx
2
Therefore, for static fields

Z p
Z
0
E |
2U () dx |=|
2U () d | .
(5.17)
(5.18)
Since T is positive, this bound also holds for time dependent fields. We assumed
that U () 0 and this allows us to introduce a superpotential W (), that satisfies
U () = 21 ( dW
)2 . If we substitute this, then the right-hand side of (5.18) can be
d
integrated and the bound takes the form
E | W (+ ) W ( ) | .
(5.19)
Bounds of this general form, where the energy is bounded from below in terms
of solely topological data, are known as Bogomolny bounds.
138
Topology in Physics
For our purposes such a bound is extremely useful. We are interested in

fields, that minimize the energy within a given topological class of fields. Such a
topological class is completely determined by the topological data that occur in
the bound. So the only thing we have to do, is to look for fields, which attain
equality in the bound and of course check, that they also satisfy the field equation.
To attain equality in the Bogomolny bound the field must be time independent,
= 0, and satisfy one of the first order Bogomolny equations
p
0 = 2U (),
(5.20)
where solutions of the equation with the + sign (if they exist) are called kinks
and those with the - sign antikinks. For these solutions, the gradient term and
the potential term of the energy density, 12 02 and U (), are pointwise the same,
which is a stronger statement than the virial theorem (5.15).
It is easy to check, that solutions of the Bogomolny equations (5.20) satisfy
the field equation: Differentiating (5.20) yields
1 dU 0 dU
00 =
=
.
d
2U d
One could have also argued in an other way: Since solutions of (5.20) are global
minima of the energy within the topological class of fields defined by ( , + ),
they are critical points of the energy and therefore automatically static solutions
of the second order field equation (5.12).
4 kinks
In this section we start to be concrete and make a detail discussion of the simplest
model with kinks, where there are two topologically different vacua, that is 0 (V)
consists of two elements. A potential which satisfies U () 0 and attains global
minima Umin = 0 at two distinct points is for example
U () = (m2 2 )2
(5.21)
where m and are positive real constants, m, > 0.

Obviously, the global minima is degenerated and is attained at the two points
= m. Thus there are two vacua, which we denote by V+ and V . This model
is known as the 4 -model. Its full Lagrangian density is given by
1
L = (m2 2 )2
2
(5.22)
and the corresponding field equation reads as

4(m2 2 ) = 0.
(5.23)
Solitons
139
Finite energy configurations are classified by an element ( , + ) of 0 (V)0 (V),

but we can capture the topological content of a given field in a even more compact
notation, if we define the topological charge
N=
+
,
2m
(5.24)
where = limx7 (x). The possible values of N are {0, 1, 1}. If N = 0,

then the field lies in the same homotopy class as one of the vacuum solutions,
(x) = m, to which it may be continuously deformed. The kink is the minimal
energy solution with N = 1 and interpolates between the vacua V and V+ as
x increases from to . The antikink is the corresponding solution with
N = 1, which is obtained by making the replacement 7 in the kink
solution.
We know that these kink solutions are given by the solutions of the Bogomolny
equations. The Bogomolny energy bound for the 4 model is
Z +
1
4
E|
2(m2 2 ) d| = | 2[m2 3 ]+ | = m3 2|N |. (5.25)
3
3
Since |N | = 1 for both the kink and antikink, the bound in these sectors is
E 43 m3 2. To attain equality, one of the first order Bogomolny equations
(5.20) has to be satisfied, which in the 4 model takes the form
0 = 2(m2 2 ).
(5.26)
The + sign gives a kink, the - sign an antikink. This equation (with the + sign)
can be integrated to yield the kink solution
(x) = m tanh( 2m(x a)),

(5.27)
where a is an arbitrary constant of integration. This solution has the energy
density
1
E = 02 + (m2 2 )2 = 2m4 sech4 ( 2m(x a)).
(5.28)
2
R
And the total energy of the kink therefore is E = E dx = 34 m3 2, as it
should be. Since the solution is static, the total energy E is also the rest mass
M of the kink. Note that at the point x = a, (x) is equal to zero, the value
mid-way between the asymptotic values m. The point x = a is also the point,
where the energy density is maximal, and equal to 2m4 . It is therefore natural
to interpret the point a as the position of the kink. It is a free parameter of the
solution, corresponding to the translation invariance of the Lagrangian density.
In figure (5.1) we plot the kink solution (5.27) and its energy density (5.28) for
the choice of parameters = 21 , m = 1 and a = 0.
140
Topology in Physics
1
0.5
0
-0.5
-1
-4
-2
0
x
Figure 5.1: The kink solution (solid curve) and its energy density (dashed curve)
Since the Lagrangian density (5.22) is relativistic, we can get a moving kink
solution by simply applying a Lorentz boost
(t, x) = m tanh( 2m(x vt a)),

(5.29)
where 1 < v < 1 is the velocity of the kink and

= 1/ 1 v 2 is the Lorentz
factor. The energy of the moving kink is E = 34 m3 2.
The possible values of the topological charge N are 0 and 1, so there are
no multi-kink solutions with N > 1. However, it is possible to have a field
configuration with a finite mixture of kinks and antikinks. As an example and
to conclude this section we compute the interaction energy of a well separated
kink-antikink pair, and show that there is an attractive force between the two.
To simplify the analysis we set = 21 and m = 1, since by a simple redefinition of
the field and length units, the constants and m can always be scaled to equal
any given positive values.
We will derive the interaction energy by identifying the force produced on one
soliton by the other with the rate of change of momentum. For a general theory
of the form (5.11), the momentum on the semi-finite interval (, b] is
Z
P =
0 dx.
(5.30)
The force on this interval is then

Z b
0 + 0 dx = [ 1 ( 2 + 02 ) + U ()]b .
F =P =
(5.31)
The final expression is obtained by using the field equation (5.12) and integrating
over the total time derivative terms.
Now consider a kink-antikink pair, with the antikink at position a and
the kink at position a, where a 1, so that they are well separated. A field
Solitons
141
configuration, that describes such a situation, is for example

(x) = 1 (x) + 2 (x) + 1,
(5.32)
where 1 (x) is the antikink and 2 (x) is the kink, given explicitly by
1 (x) = tanh(x + a), 2 (x) = tanh(x a).
(5.33)
Let the endpoint of the interval, b, lie between the kink and antikink, and far
from each, that is, a b a. Then throughout the interval the terms (2 + 1)
and 02 are both close to zero, so we can linearize in these terms. To first order
this yields the result
dU
1
0 0
F = [ 02
(1 )]b = [01 02 +(1+2 )001 ]b , (5.34)
1 +U (1 )1 2 +(1+2 )
2
d
where, to obtain the second expression, we have used the fact that the antikink
solves the Bogomolny equation (5.20) to cancel the first two terms in the first
expression and replaced the last term, using the static version of the field equation
(5.12). Our field configuration has the property, that its spatial derivatives fall
of exponentially fast at infinity, and so there is clearly no contribution from the
lower limit in the expression (5.34). Since the point b is far from both the kink
and antikink, we may evaluate the contribution from the upper limit, using the
asymptotic forms
1 (x) 1 + 2e2(x+a) ,
2 (x) 1 + 2e2(xa) .
(5.35)
So the expression for the force asymptotically reads as

F = 32e2R =
dEint
,
dR
(5.36)
where we have defined the kink-antikink separation R = 2a, and equated the
force with the derivative of the interaction energy Eint . The force F is independent of b, as it should be since b has non physical significance and was only
introduced to perform the analysis. Thus we can indeed identify F with the force
on the antikink, produced by the kink. Finally, we have the following asymptotic
interaction energy
Eint = 16e2R .
(5.37)
which is negative and decreases as the separation decreases. This indicates that
the force between the kink and antikink is attractive.
Numerical simulations of the full time-dependent field equation, starting with
a well separated kink-antikink pair at rest, confirm this picture of the kinkantikink interaction: The kink and the antikink move toward each other and
annihilate into radiation.
142
Topology in Physics
Figure 5.2: A physical system described by the sine-Gordon model

Sine-Gordon kinks
In the 4 model we have not seen multi-kink solutions, since they were not consistent with the finite energy boundary condition. A model, where multi-kink
solutions exist, is the sine-Gordon model, which is defined by the Lagrangian
density
L = (1 cos )
(5.38)
with corresponding field equation
+ sin() = 0.
(5.39)
We can visualize a physical model, which would be described by such an equation.

Consider a long straight, horizontal clothesline with identical pendula of mass m
and length l attached at equal distance along its length. The whole system lies in
an elastic with coefficient of force k and the masses are acted on by gravity. If n
is the angle of the nth pendulum, then the nth mass has the following Lagrangian
L=
m 22
k
k
l n mgl(1 cos(n )) (n+1 n )2 (n1 n )2 ,
2
2a
2a
(5.40)
the equation of motion therefore reads

k
ml2 n = mgl sin(n ) + (n+1 + n1 2n ).
a
(5.41)
Let the x direction be along the clothesline and xn = na, then if a is small we
can approximate
k
(n1 n )/a (n n1 )/a
(n+1 + n1 2n ) = ka
ka00 (xn ).
a
a
(5.42)
Solitons
143
So the equation of motion becomes

ml2 = mgl sin() + ka00 .
(5.43)
Now let a 7 0 and k 7 , such that ka 7 ml2 is finite. Then

g
00 = sin(),
l
(5.44)
which is the sine-Gordon equation, if units are suitably chosen.

It is obvious that the vacuum configurations of the sine-Gordon model are
given by the constant solutions = 2n, where n Z is any integer, so
0 (V) = 2Z.
(5.45)
Once again finite energy solutions are classified by an element (+ , ) 0 (V)

0 (V). But since the Lagrangian density (5.38) is invariant under 2 shifts of the
fields, 7 2, these can equivalently be classified by the topological charge
N=
+
,
2
(5.46)
which is an integer and counts the net number of solitons.

The Bogomolny bound for the sine-Gordon model is
Z +
Z 2N
2| sin( )| d =
2| sin( )| d = 4|N |[ cos( )]2
E
= 8|N |, (5.47)
2
2
2 0
0
where we have used the periodicity of the integrand and the fact that the range
of integration is an N -fold cover of the interval [0, 2], to evaluate the integral.
Equality in the bound is attained by solutions of one of the first order Bogomolny
equations
0 = 2 sin( ).
(5.48)
2
If we restrict ourselves to kink solutions, by choosing the + sign, this can be
integrated directly to yield
(x) = 4 tan1 (exa ),
(5.49)
where a is an arbitrary constant of integration. We see that = 0 and + = 2,

so this solution has topological charge N = 1 and describes a single kink. The
energy density of the kink is
E = 4 sech2 (x a),
(5.50)
which is maximal at the point x = a. This is also the point, where has value
, the value half-way between the vacuum values 0 and 2. Thus a can be
144
Topology in Physics
interpreted as the position of the sine-Gordon kink - just as in the case of the
4 -kink.
R From the expression of the energy density it is easy to confirm that
E = E dx = 8.
The general solution of the Bogomolny equations is a kink of unit charge, and
therefore there are no multi-kink solutions of the Bogomolny equations. From
this we can conclude that there is a repulsive force between two kinks: Any field
configuration with N = 2 has to obey the strict Bogomolny bound E > 16,
but in the limit in which two kinks are infinitely separated, the energy must be
equal to the sum of the energies of the two individual kinks, that is E = 16. So
the potential energy of two kinks decreases as they separate, and this indicates
that there is a repulsive force between them. In the last section we computed
the asymptotic kink-antikink interaction energy in the 4 model. We cam do an
analogue calculation to find the asymptotic interaction energy of two sine-Gordon
kinks. We start from (5.31) and consider the two kink field
(x) = 1 (x) + 2 (x),
(5.51)
where 1 (x) is a kink at position a and 2 (x) is a kink at position a and a 0.

We assume also that a b a. Then throughout the interval (, b] the
terms 2 and 02 are both close to zero and we can linearize in this terms. This
yields the result
1
dU
0 0
F = [ 02
(1 )]b = [01 02 + 2 001 ]b ,
1 + U (1 ) 1 2 + 2
2
d
(5.52)
where, to obtain the second expression, we have used (5.20) and (5.12) - just as
in (5.34). Again, there is no contribution from the lower limit in the expression
(5.52), since the derivatives fall off rapidly. Because b is far from both kinks, we
may use the asymptotic expressions of 1 and 2 to evaluate the contribution
from the upper limit. These asymptotic forms are
1 (x) 4(
e(x+a) ),
2
2 (x) 4exa .
(5.53)
Therefore the asymptotic expression for the force is

F = 32eR ,
(5.54)
where R = 2a is the kink-kink separation. Thus the asymptotic interaction

energy is
Eint = 32eR .
(5.55)
The fact that there is a repulsive force between two kinks implies that there
are no static multi-kink solutions of sine-Gordon equation. But all the same, there
exist time dependent solutions, that describe multi-kink states and in particular
Solitons
145
the scattering of two or more kinks. As a simple example we will now explicitly construct a two-kink solution of the sine-Gordon equation using so called
Baecklund transformations.
If we introduce lightcone coordinates, x = 12 (x t), with corresponding
derivatives, = x , the sine-Gordon field equation (5.39) becomes
+ = sin().
(5.56)
The following pair of equations is known as a Baecklund transformation

+ = + 2 sin(
+
),
2
= +
sin(
),
(5.57)
where is a non-zero constant and is called the Baecklund parameter. The

Baecklund transformation (5.57) may be thought of as determining the field ,
given the field , ore vice versa. Since partial derivatives commute, one can
subject (5.57) to the compatibility condition + = + , which implies
that
+ cos(
+
1
)( + ) = + + cos(
)(+ + ). (5.58)
2
And this reduces to the sine-Gordon equation, + = sin(), if we eliminate the

first derivative terms using equations (5.57). Of course one can also subject the
Baecklund transformation (5.57) to the compatibility condition + = + ,
which gives the sine-Gordon equation for , that is, + = sin(). Thus, if
and satisfy the equations (5.57) and are enough regular (i.e the compatibility
conditions hold), each of them automatically satisfies the sine-Gordon equation.
But we are not interested in finding general functions and , that satisfy equations (5.57). We will think of the Baecklund transformation as a map between
solutions of the sine-Gordon equation. That is, we start with a known solution
of the sine-Gordon equation and generate a new solution via the Baecklund
transformation. Since the Baecklund transformation contains a free parameter ,
this extra parameter is introduced into the new solution, in addition to a constant
of integration.
As an example, if we start with the trivial vacuum solution = 0, then
equations (5.57) take the simplified form
+ = 2 sin( ),
2
= sin( ).
(5.59)
These equations are integrated to give the solution

(x+ , x ) = 4 tan1 (ex+ x /+ ),
(5.60)
146
Topology in Physics
where is a constant of integration. If we make the identifications

v=
1 2
,
1 + 2
1 + 2
1
=
,
2
1 v2
a=
2
,
1 + 2
(5.61)
where < 0, this solution can be written as

(t, x) = 4 tan1 (e(xvta) ),
(5.62)
which we recognize as the Lorentz boosted version of the one-kink solution (5.49).
Now we use the Baecklund transformation to construct a two-kink solution
in a purely algebraic way, evading the task of having to explicitly integrate equations (5.57), which could be very hard for a complicated seed solution . If we
start with the seed solution = 0 , we can generate two new solutions 1 and
2 , using different Baecklund parameters 1 and 2 . One can show, that the following theorem of permutability holds: With an appropriate choice of integration
constants, the solution 12 , obtained by applying the Baecklund transformation
with parameter 2 to the seed solution 1 , is equal to the solution 21 , obtained
by applying the Baecklund transformation with parameter 1 to the seed solution
2 . The consistency condition 12 = 21 yields the relation
12 = 21 = 4 tan1 [(
1 + 2
1 2
) tan(
)] 0 ,
2 1
4
(5.63)
which gives a new solution in terms of the triplet of known solutions 0 , 1 ,

2 .
We have seen, that if we start with the vacuum solution 0 = 0, we can obtain
the one-kink solutions j = 4 tan1 (ej ) (j = 1, 2), where j = j x+ x /j +
j . So we can get another solution, by substituting these into equation (5.63)
(x+ , x ) = 4 tan1 [(
2
)
1 + 2 sinh( 1
2
)
].
+
2 1 cosh( 1 2 2 )
(5.64)
For simplicity, set 1 = 1/2 , and 1 = 2 = 0. Then (5.64) becomes

(t, x) = 4 tan1 [
v sinh(x)
],
cosh(vt)
(5.65)
where we made the same identifications (5.61) as before. This solution has topological charge N = 2, since it interpolates between the vacua 2 and 2. It
therefore describes a time dependent two-kink field.
We can interpret this solution, if we write it in the form
tan( ) = e(xa) e(x+a) ,

4
(5.66)
Solitons
147
where a > 0 is the time dependent function

a(t) =
1
2
log( cosh(vt)).
(5.67)
If |vt| 1, then cosh(vt) 21 e|vt| and therefore a |vt| log(v)/ |vt| +

is also large. In this limit, near the point a, the second term on the right-hand
side of (5.66) is exponentially small and can be neglected. The remaining term
describes a single kink, located at x = a |vt| + and moving with speed v.
Similarly, near x = a, one can neglect the first term and the remaining term
describes a kink at position x = a (|vt| + ) and moving with speed v.
Thus this solution indeed describes a two-kink state. If |t| is large, the two kinks
are well separated and both move toward the origin at speed v for t < 0 and
away from the origin at the same speed for t > 0. This is consistent with our
statement that there is a repulsive force between two kinks: When they come
closer together, they feel the repulsive force and smoothly bounce back off each
other. The motion is symmetric about t = 0, since the solution is an even function
of t. The time of closest approach is t = 0. The interpretation of a as half the
separation of the two kinks is only valid, when a is large, so we should not use it
near t = 0 to estimate the distance of closest approach. In figure (5.3) we plot
the energy density at various times for the two-kink solution (5.65) with v = 0.2.
The total energy is 16.
We can validate the expression (5.55) for the asymptotic interaction energy,
if we compare the exact two kink solution to the approximate motion one would
predict using the equation of motion, that follows from (5.55). For two kinks at
positions a, the approximate equation of motion reads as
a
= 4e2a ,
(5.68)
where we have equated the force F = 32e2a with the product of the kink acceleration a
and its mass, which is 8. Using the conditions a()
= v and a(0)
= 0,
so that the kinks each have initial speed v and the time of closest approach is at
t = 0, then (5.68) has the solution
2
a(t) = log( cosh(vt)).
v
(5.69)
This is the non-relativistic limit of the exact expression (5.67), which is obtained
by replacing the Lorentz factor by 1. So for low speeds v 1, the asymptotic
force law yields a very good approximation for the true dynamics. One source
of the error for high speeds is that when the two kinks come closer together, the
terms neglected in the asymptotic expression become relevant.
Figure (5.4) shows the exact kink trajectories given by the expression (5.67)
and the approximate trajectories obtained form (5.69), for speeds v = 0.2 and
v = 0.6.
148
Topology in Physics
t=-20
3
2
1
0
-10
-5
0
x
10
-5
0
x
10
-5
0
x
10
t=0
3
2
1
0
-10
4
t=20
3
2
1
0
-10
Figure 5.3: The energy density of the two kink solution at various times
Solitons
149
6
v=0.6
5
4
3
2
1
0
-10
-5
0
t
10
Figure 5.4: Exact solution for a(t) (solid curve) versus asymptotic force solution
for a(t) (dashed line) at various speeds
In Chapter 2 we mentioned, that the sine-Gordon kink has also an interpretation as a nonlinear kink. To see this, we need to formulate the sine-Gordon
model as a nonlinear scalar field model. For that purpose we introduce the twocomponent unit vector
= (1 , 2 ) = (sin(), cos()).
(5.70)
In terms of this field the sine-Gordon Lagrangian density (5.38) becomes

1
L = (1 2 ) + (1 ),
2
(5.71)
where we introduced the Lagrange multiplier to constrain to lie on the circle

S 1 . To ensure finite energy, the field must attain the (unique) vacuum value
= (0, 1) at spatial infinity, which corresponds in the previous formulation to
the field being an integer multiple of 2 at infinity. Therefore the vacuum
manifold consists only of one singly point, which - as we mentioned - allows us
to compactify space R to S 1 . The field is therefore a map : S 1 S 1 and is
classified by an element N of 1 (S 1 ) = Z. The integer N is the winding number
of the map and can be computed as
Z
1
ab 0a b dx,
(5.72)
N=
2
where ab is the alternating tensor in two dimensions, with 12 = 21 = 1, and all
other components zero. One can easily check, that this is equal to the topological
charge (5.46), which we defined previously.
150
Topology in Physics
6 Solitons in density wave

systems
Mathias Ritzmann
Supervisor: Urs Aeberhard
We treat the formation of the charge density wave ground state
in a 1D mean-field framework. We analyse solitons in transpolyacetylene with half band-filling, deriving charge, spin, and
charge distribution of the soliton. An extension to third bandfilling is discussed briefly. For the half-filled band case, the soliton
has unusual charge-spin relations whereas for the third-filled case,
the soliton has fractional charge.
Introduction
Some materials with a highly anisotropic band structure have a ground state
called charge-density wave in which the crystal lattice is deformed periodically.
We treat the formation of the charge density wave ground state due to electronphonon interaction in a one-dimensional model using the nearly-free electron
approximation and linear response theory for the electron gas.
Solitons in polyacetylene with a half-filled pi band are investigated, charge, spin,
and charge distribution are calculated using Greens Functions. Extensions to
the case of a third-filled pi band are discussed briefly.
Experimental Signature
In Figure 6.1, one can see that the material is a metal at temperatures above
T2 , the resistance gets lower as the temperature decreases, whereas below T2 the
resistance increases with decreasing temperature, as it is characteristic for a semiconductor. The same metal-semiconductor transition can be seen for potassium
molybdenum blue bronze in Figure 6.2, above the phase transition temperature,
151
152
Topology in Physics
Figure 6.1: Resistivity measurements for N bSe3 taken from [35]. Solid circles
denote resistivity, open circles its derivative with respect to temperature.
the conductivity increases with decreasing temperature, below it decreases with
decreasing temperature.
Figure 6.3 shows a divergence in the derivative of the thermal conductivity of
rubidium blue bronze at the temperature where the charge density wave forms,
which indicates a second order phase transition.
The Formation of the Charge Density Wave
We discuss the formation of a charge density wave in a one-dimensional free

electron gas coupled to the underlying lattice, following [38] very closely.
The Hamiltonian of the electrons is
X
Hel =
k ak ak ,
(6.1)
k
where the electron energy k equals ~2mk , m denoting the effective electron mass.
ak is the annihilation operator for an electron with momentum k, ak the corresponding creation operator.
Our considerations are always for one spin direction, we will sum over spins at
the end of the calculation.
2 2
Solitons in density wave systems
153
Figure 6.2: Natural logarithm of the conductivity as a function of inverse temperature for potassium molybdenum blue bronze. Image from [36]
.
Figure 6.3: Thermal conductivity as a function of temperature measured for two

samples of rubidium blue bronze. In the inset the temperature derivative of the
thermal conductivity near the phase transition is shown. Image from [37].
154
Topology in Physics
The Hamiltonian of the lattice vibrations is

X Pq Pq M q2
Hph =
+
Qq Qq ,
2M
2
q
(6.2)
where Qq is the normal coordinate for the oscillator normal mode with wavevector
q, Pq is the associated momentum, q the normal mode frequency, M the mass
of the ion.
We write the harmonic oscillator with creation and annihilation operators for
phonons with wavevector q,
Hph =
X
q
1
~q (bq bq + ),
2
(6.3)
where
12

~
bq + bq ,
Qq =
2M q

1

~M q 2
Pq =
bq + bq ,
2

(6.4)
(6.5)
which gives for the lattice displacement at a point

X
u(x) =
d
q
~
2M q
12
bq +
bq
eiqx
x = n d,
(6.6)
with d denoting the distance between two ions in equilibrium position.

We assume that the potential depends on the position of the ions only (which is
technically an approximation because the ions are moving charges), therefore
X
hk |V (r n d u)| k 0 i ak ak0
Helph =
(6.7)
k,k0 ,n
ei(k k)(nd+u) Vkk0 ak ak0 ,
(6.8)
k,k0 ,n
where n counts through the lattice positions, u is the displacement from the
equilibrium position, Vkk0 is the k k 0 -component of the Fourier transform of
V (r), the potential of a single ion.
We assume small displacement from the equilibrium position, enabling us to
linearise the exponential
X
0
ei(k k)u u 1 + i(k 0 k)u = 1 + i(k 0 k) d
uq eiqnd ,
(6.9)
q
155
by inserting the definition of u. The electron-phonon interaction Hamiltonian is

then
!
X
X
0
Helph =
uq eiqnd Vkk0 ak ak0
(6.10)
ei(k k)nd 1 + i(k 0 k) d
=
k,k0 ,n
i(k0 k)nd
Vkk0 ak ak0
XX
X 0
0
(k k)Vkk0 ak ak0
+i d
eind(k k+q) uq
k,k0 ,n
k,k0
|n
{z
= d1 qk+k0
(6.11)
X
1 X 0
0
(k k)ukk0 Vkk0 ak ak0 ,
=
ei(k k)nd Vkk0 ak ak0 + i
d k,k0
k,k0 ,n
(6.12)
where the first term describes the interaction between the electrons and the ions
in the equilibrium position. It leads to the formation of Bloch states from the
free electron states. The effect on the dispersion relation is strongest close to the
Brillouin zone edge. Since we are interested in materials where the Fermi wavevector is far away from the zone edge, we will not treat this term explicitly, instead
we will just assume that our index k counts through Bloch states in the following.
The second term, describing the interaction between the electrons and the phonons,
can be written out using phonon creation and annihilation operators,

12

~
bkk0 + bk0 k Vkk0 ak ak0

Helph = i
(k k)
2M kk0
k,k0
"
21 #

X
~
bkk0 + bk0 k a(kk0 )+k0 ak0
=
i(k k 0 )Vkk0
2M kk0
(kk0 ),k0
{z
}
|
X
gkk0

gq bq + bq aq+k0 ak0 ,
(6.13)
(6.14)
(6.15)
q,k0
where we have introduced the electron-phonon coupling constant gq .

The full model Hamiltonian is called Frohlich Hamiltonian:

X
X
X
gq ak+q ak bq + bq .
H=
k ak ak +
~q bq bq +
k
(6.16)
k,q
We now assume that the electron-phonon coupling gq = g is independent of q.

The equation of motion for the normal coordinates is
h
i
q = i Q q , H = 1 [[Qq , H] , H] ,
Q
(6.17)
~
~2
156
Topology in Physics
which we compute using [Qq , Pq0 ] = i~q,q0 .

"
#
X Pr Pr
[Qq , H] = Qq ,
2M
r
X [Qq , Pr Pr ]
=
2M
r
X [Qq , Pr ] Pr + Pr [Qq , Pr ]
=
2M
r
1
2M
X X
= 2i~Pq
[Pq , H] =
X M 2
r
[Pq , Qr Qr ] +
(6.18)
(6.19)
(6.20)
(6.21)
!
1
2M r 2
ak+r ak g
[Pq , Qr ]
~
!
1
X
M q2
2M q 2
akq ak g
=
(2i~Qq ) i~
2
~
k
!!

12 X
2M
q
= i~ M q2 Qq g
akq ak
~
k
!!

12 X
1
2M
i~
q
q =
akq ak
Q
i~ M q2 Qq g
~2 M
~
k

21
2q
= q2 Qq g
q ,
M~
where q denotes the q-Fourier coefficient of the electronic density,
!
!
X
X
X
ak+q ak =
ak akq =
akq ak .
q = q =
k
(6.22)
(6.23)
(6.24)
(6.25)
(6.26)
(6.27)
We can write the effective force on Qq in the form of an ionic potential

V (Qq ) = g
writing
2M q
~
12
Qq ,
q = M 2 Qq q V (Qq ).
MQ
q
Qq
(6.28)
(6.29)
We use the ionic potential only to calculate the charge density q . The relation
between the ionic potential and the induced charge density is given in linear
response theory by
ind
= (q, T )Vq ,
(6.30)
q
157
Figure 6.4: Lindhard response function at T = 0, normalised to its (negative)

value at q = 0 as a function of wavevector q. Drawn for different dimensionalities.
Image taken from [38]
where (q, T ) is the wavevector- and temperature-dependent Lindhard response
function. It is given in d dimensions by
Z
dk f (Ek , T ) f (Ek+q,T )
(q, T ) =
,
(6.31)
(2)d
Ek Ek+q
where Ek denotes the energy of the electron with wavevector k, f (E, T ) is the
Fermi function. In Figure 6.4 the Lindhard response function at T = 0 is drawn
for different dimensions. Considering that for T = 0 and d = 1 the Lindhard
response function goes to as q 7 2fF , we are most interested in its behaviour
at this point. Linearising the dispersion relation around 2kF as shown in Figure
6.5, we have
Z 0
2kB T tanh(x)
(2kF , T ) = e2 n(F )
dx,
(6.32)
x
0
where 0 is a cutoff energy (usually one takes 0 = F ) which is made necessary
by our introduction of a linearised dispersion that is only valid near kF . n(F ) is
the density of states at the Fermi surface. This leads us to
(2kF , T ) = e2 n(F ) ln
1.140
,
kB T
(6.33)
which is drawn for various temperatures in figure 6.6. Using the above calcula-
158
Topology in Physics
Figure 6.5: Linearised dispersion relation around 2kF taken from [38]. Drawn
into the figure is the scattering of an occupied state near kF into an empty
state near kF that does hardly change the energy of the system.
Figure 6.6: (kF , T ), drawn for various temperatures T as a function of wavevector. Illustration from [38]
159
tions for (2kF , T ) we can write the equation of motion for Qq as

"

21
21 #
2
2M
q
q
q = 2 + g
Q
(q, T )g
Qq
q
M~
~

2g 2 q
2
= q +
(q, T ) Qq ,
~
(6.34)
(6.35)
which makes it evident that we have a new, smaller phonon frequency

qnew
2
= q2 +
2g 2 q
(q, T ).
~
(6.36)
The effect is most prominent at q = 2kF , we have

new 2
2k
F
2
2k
F
2g 2 n(F )2kF
ln
~
1.140
kB T

.
(6.37)
This equation defines a (mean field) transition temperature where the phonon
mode 2kF freezes in,
~2kF
kB T M F = 1.140 e 2g2 n(F ) .

(6.38)
We have to add for completeness that the use of a one-dimensional model for
studying the transition to the CDW ground state is somewhat problematic. In a
one-dimensional world, fluctuations strongly suppress the transition to the CDW
ground state (see for example [39]), the phase transition measured in the laboratory is the locking in of the phase of neighbouring chains (see [38]).
Soliton Excitations in Polyacetylene
This section is based on [40] and [41].

Trans-Polyacetylene has a twofold-degenerate ground state, as can be seen in
Figure 6.7. Therefore, one can at least intuitively expect that there exist solitons, excitations corresponding to a moving domain wall between the two ground
states.
We use the following approximations to simplify the treatment of polyacetylene
neglect interactions between chains,
treat the electrons only in the form of a spring constant,
treat only displacements of CH groups along the chain direction,
treat the electrons in tight-binding approximation, expand nearest-neighbour
hopping to first order around the equilibrium position (see Figure 6.8),
160
Topology in Physics
Figure 6.7: The two ground states of trans-polyacetylene, un is the dimerisation

coordinate used in the model Hamiltonian. Above the A phase, below the B
phase. Image from [40].
neglect Coulomb interactions between electrons.
We have therefore for the bonding energy
E =
1X
K (un+1 un ) ,
2 n
(6.39)
where K is the effective spring constant. The chain has N atoms (we will refer to
the CH groups as atoms since their structure is ignored) on it, we use periodic
boundary conditions for simplicity.
We expand the hopping integral for the electrons as
tn+1 = t0 (un+1 un ) ,
(6.40)
where t0 is the hopping integral for the undimerised chain, is the electronlattice displacement coupling constant. The above equation is the standard form
for electron-phonon interaction in metals.
The kinetic energy of the CH groups, each with mass M , is given by
Ekin =
1X
M un .
2 n
(6.41)
We have therefore the complete model Hamiltonian

1X
X
1X
H=
tn+1,n cn+1,s cn,s + h.c. +
K (un+1 un ) +
M un , (6.42)
2 n
2 n
n,s
with cn,s (cn,s ) denoting the creation (annihilation) operator for a electron with
spin s on the CH group number n. Of course, c and c satisfy anticommutation
161
Figure 6.8: Nearest-neighbour hopping integral as a function of dimerisation

coordinate difference (from [40]).
relations whereas un and pn satisfy commutation relations. The system is spin
degenerate, therefore we will not write out sums over spin explicitly in the following.
Given a perfectly dimerised chain, means un = (1)n u, the model Hamiltonian
without the kinetic energy term reduces to

X
(6.43)
Hd (u) =
[t0 + (1)n 2u] cn+1 cn + h.c. + 2N Ku2 .
n
We make use of the 2a periodicity of the problem by introducing a reduced zone
scheme with a zone boundary 2a

and summing over two states per k value, one
in the valence band and one in the conduction band. Figure 6.9 shows the u = 0
bands where
Ek0,v = 2t0 cos(ka) = k ,
Ek0,c = 2t0 cos(ka) = k .
(6.44)
The operators cvk and cck for valence- and conduction band states are given by
1 X ikan
cvk =
e cn ,
N n
(6.45)
i X i(ka+)n
e
cn
cck =
N n
which enables us to write the Hamiltonian as
X

Hd =
k cck cck cvk cvk + 4u sin(ka) cck cvk + cvk cck + 2N Ku2 .
(6.46)
We arrive at the expression for the Hamiltonian in the k representation

Hd =
X
k

k cck cck cvk cvk + 4u sin(ka) cck cvk + cvk cck + 2N Ku2 .
(6.47)
162
Topology in Physics
Figure 6.9: Brillouin zone scheme for the electron band in the undimerised
state. The unit cell is assumed to have two CH groups here to account for the
dimerisation that will be included later, leading to the opening of gaps at the
zone boundary. Then, the band denoted C will be the conduction band of a
semiconductor, whereas the band denoted V will be the valence band. Taken
from [40].
We are going to diagonalise this Hamiltonian by an SU (2) transformation in
the operators, the rotated operators are
v
v
ak
k k
ck
=
, |k |2 + |k |2 = 1.
(6.48)
ack
k k
cck
We write the Hamiltonian in equation (6.46) in terms of the new operators,
using
v
v
ck
k k
ak
=
(6.49)
cck
k k
ack
to get
Hd =
k
k avk + k kc
+ 4u sin(ka)
+ 2N Ku2 .

k akv + k ack k avk + k ack (k avk + k ack )

k avk + k ack (k akv + k ack ) k avk + k ack (k avk + k ack )
(6.50)
We simplify the notation using the number operator n = a a giving us the
163
form
Hd =
+
+

nvk k |k |2 k |k |2 + 4u sin(ka) (k k ) + 4u sin(ka) (k k )
k
c
nk k |k |2 k |k |2
v c
ak ak 2k k k +

+ 4u sin(ka) [k k + k k ]

k2 k2 4u sin(ka) + h.c. + 2N Ku2 .
(6.51)
For having a diagonal Hamiltonian in the rotated creation and annihilation

operators we require 2k k k + 4u sin(ka) (k2 k2 ) = 0. We choose to have
k real and positive, giving us
k k k + 2u sin(ka)(k2 k2 ) = 0.
(6.52)
Inserting our choices k = |k |, k = |k |, and k2 = 1 k2 we have

2k k2 (1 k2 ) = 4k2 u2 sin2 (ka)(2k2 1)2 ,
(6.53)
2 2
2 4
2 2
2
4
2 2
2
2
2 2
2
k k k k = 16 u sin (ka)k 16 u sin (ka)k + 4 u sin (ka), (6.54)
1
k4 (2k 2k ) + k2 (2k + 2k ) 2k = 0,
(6.55)
4
where we have introduced k = 4u sin(ka). We solve the equation for k2 , giving
p
Ek2 Ek4 2k Ek2
2
(6.56)
k =
2Ek2
s
1
Ek4 2k Ek2
=
(6.57)
2
4Ek4

k
1
=
1
(6.58)
2
Ek
with Ek2 = 2k + 2k . Because of k2 + k2 = 1 the does only switch the roles of
k and k , we can choose

1
1
k
k
2
2
, k =
,
(6.59)
k =
1+
1
2
Ek
2
Ek
without loss of generality.
We have therefore k2 k2 = Ekk > 0. Using this inequality we see in equation
(6.52) that k has positive sign for k > 0, negative for k < 0, namely

12
1
k
k =
,
1+
2
Ek

21
1
k
k = sgn(k)
.
1
2
Ek
(6.60)
164
Topology in Physics
We write the nonvanishing part of equation (6.51) as

X

(nck nvk ) k (k2 k2 ) + 2k k k + 2N Ku2 .
(6.61)
q
q
2
2
Inserting k (k2 k2 ) = Ekk and 2k k k = k 1 Ek2 =
2k
k
q 4 2 2 2 2
k +k k k k
2
= Ekk we have the diagonalised Hamiltonian
E2
2k 2k
Ek2
Hd =
Ek (nck nvk ) + 2N Ku2 .
(6.62)
The ground state energy is obtained from equation (6.62) by setting nvk = 1
and nck = 0 (we have N atoms, the sum over k has N2 terms, and the system is
spin degenerate) giving
X
E0 (u) = 2
Ek + 2N Ku2 .
(6.63)
p
2k
2k ,
+
k = 2t0 cos(ka), k = 4u sin(ka) and we replace
We recall Ek =
the sum by an integral, the normalisation is
L
2
2a
2a
Na
dk =
2
2a
dk =
2a
X
,
(6.64)
where L = N a denotes the chain length. The integral expression for the ground
state energy as a function of the dimerization u is therefore
L
E0 (u) =
2a
2a
Ek dk + 2N Ku2 .
(6.65)
We make use of the inversion symmetry, Ek = Ek to get

Z
2L 2a
E0 (u) =
Ek dk + 2N Ku2
0
Z q
2L 2a
=
(2t0 cos(ka))2 + (4u sin(ka))2 dk + 2N Ku2 ,
0
where we substitute q = ak, k =
2a
q =
,
2a
(6.66)
dk = a1 dq to get
Z q
2L 2
E0 (u) =
(2t0 cos(q))2 + (4u sin(q))2 dq
a 0
Z q
4L 2
(t0 cos(q))2 + (2u sin(q))2 dq.
=
a 0
(6.67)
165
Figure 6.10: Energy per CH group as a function of dimerisation, the two stable
minima are the A and B phase. Illustration from [40]
We replace cosine by sine,
(t0 cos(q))2 + (2u sin(q))2 = t20 + sin2 (q)(t2o + 42 u2 )

42 u2
2
2
= t0 1 (1
) sin (q) ,
t20
and we introduce the shorthand z =
t1
t0
2u
to
t0
(6.68)
get to
Z q

4N t0 2
N Kt20 z 2
E0 (u) =
1 (1 z 2 ) sin2 (q) dq +
22
0
4N t0
N Kt20 z 2
=
E(1 z 2 ) +
22
(6.69)
where E denotes the complete elliptic integral of the first kind. If z is small (which
is simply restating the condition that the hopping integral can be linearised) we
can use the approximation

1 ln 4 1
2
E(1 z ) u 1 +
z2 + . . . ,
(6.70)
2 |z|
2
which shows us that the energy has a local maximum at u = 0 and two global
minima at u0 (see Figure 6.10). We can use equation (6.69) to calculate the
density of states per spin direction

|E|
N
|E| [, 2t0 ]
1
L
2
2
2 2 ) 2
(4t
E
)(E
[
]

0
,
(6.71)
0 (E) =
=
k
0 otherwise
2 dE
dk
= 4u = 2t .
where = 2a
0
1
166
Topology in Physics
Calculations Using Greens Functions

We are calculating various Greens functions which we are going to need later on
for our treatment of the soliton. We use the eigenfunction expansion
+
*

X | c i h c |
v
v
|
i
h
|

k
k
k
k
+
(6.72)
Gdn,m (E) = n
n ,

E + i0+ Ekc E + i0+ Ekv
k
to get
(1 P
Gdn,m (E)
N
1
N
2Eeika(nm)
n m even
k (E+i0+ )2 Ek2
P 2Ek [k +ik (1)n ]2 eika(nm)
n
k
(E+i0+ )2 Ek2
for the off diagonal elements and
iE
2
2
(4t E )(E 2 2 )] 2
[ 0 E
1
Gdn,n (E) =
[(4t20 E 2 )(2 E 2 )] 2
1
2 4t2 )(E 2 2 ) 2
(E
]
[
0
m odd,
(6.73)
|E| [, 2t0 ]
|E| (0, )
(6.74)
|E| (2t0 , )
for the diagonal elements ( = 4u0 ). The expression for the diagonal elements
can be checked using the density of states already calculated because we have
d (E) =
sgn E X
=Gdn,n (E).
(6.75)
Soliton Excitations
The classical ground state is twofold degenerate, therefore there exists an excitation corresponding to a soliton. To get rid of the sign of the displacement un , we
define the order parameter
n := (1)n un .
(6.76)
The two ground states are then
(
0n =
u0 A phase
u0 B phase
(6.77)
We are doing all of our calculations assuming a widely separated soliton-antisoliton

pair in a ring , with the order parameter as in figure 6.11. This circumvents
boundary effects caused by the shift of the order parameter at one end of the
chain. If we started with a chain in a perfect B phase and introduced a soliton,
one end of the chain would end up being in the A phase, needing energy for the
switch.
167
Figure 6.11: A soliton S and an antisoliton S in a ring of (CH)x . The order

parameter is plotted radially, A and B denote the two ground states of the
dimerised chain. Image from [40]
Because of the symmetry of soliton and antisoliton, we can just work with a chain
approaching perfect B phase as n 7 and perfect A phase as n 7 . We
are determining the minimum energy configuration for this chain by isolating a
segment from n = to n = of the chain surrounding the soliton (which is
centered at n = 0). We decompose our Born-Oppenheimer model Hamiltonian
as
H = H0 + V
(6.78)
where H0 describes a chain with perfect B phase displacements for n
and perfect A phase displacements for n . The hopping between groups in
n is defined to be zero. Written out, the hopping integrals defining
H0 are
t0 (1) t1 , n <
(6.79)
t0n+1,n = 0, n <
n
t0 + (1) t1 , n .
where is taken to be odd.
provides the missing hopping integrals, it is therefore defined
The perturbation V
by
n+1,n = t0 + (1)n (n+1 + n ), n <
V
(6.80)
where the n will be minimising the total energy.
We are using the formula for calculating the energy difference due to a localised
potential derived in the appendix, namely
Z

2 EF
0
= ln det 1 G V dE.
(6.81)
E =

168
Topology in Physics
Figure 6.12: Schematic representation of the isolating potential used to get the
Greens Function for the perfectly dimerised chain with neighbouring isolated
atoms from the Greens Function for the infinite perfectly dimerised chain.
For carrying out that calculation, we first need the Greens function G0 in the
which is close to the Greens function Gd for the perabsence of the potential V,
fectly dimerized chain but not identical. We break the chain into three segments
where G0 is block diagonal, these are
A n
S <n<
B n .
(6.82)
(6.83)
(6.84)
In the S segment, we have G0n,m (E) = E1 n,m because there is no coupling between
atoms. For the A segment, we decouple by introducing an artificial potential U
placed on the group 1 as depicted in Figure 6.12. We consider U 7 to
arrive at the decoupled case. We have in segment A the following equation for
G0 :
G0n,m = Gdn,m + Gdn,1 U G01,m
(6.85)
which gives for U 7 in the A segment
G0n,m
Gdn,m
Gdn,1 Gd1,m
Gdn,n
(6.86)
and in analogy for the B segment

d
G0n,m = G
n,m
d+1,m
dn,+1 G
G
dn,n
G
(6.87)
is G with t1 replaced by t1 .
where G
Now we have the integrand in (6.81), we can carry out the integration numerically
using the trial function
u0 , n
n = u0 tanh nl , < n <
(6.88)
u0 , n
169
Figure 6.13: Soliton energy E(l) as a function of an assumed half-width l. Shown

for various values of the energy gap Eg . Taken from [40].
where l denotes the half-width of the soliton. The energy as a function of the
half-width is plotted for three different gap energies in Figure 6.13. The halfwidth is determined by finding the minimum energy of the soliton.
The use of trial functions other than tanh does not result in considerable differences in the numerical results.
The Electronic Structure of the Soliton
We calculate the change in the density of states due to a soliton by using the
formula

1
0
E ln det 1 G V = (E) 0 (E),
(6.89)
R
depicted in Figure 6.14. Together with the conservation of electrons, (E)dE =
0 and the symmetry of (E) we can see that the valence band and the conduction band each have a deficit of one-half state for each spin. Therefore, since
the valence band is fully occupied whereas the conduction band is empty, the
missing electron occupies the state 0 in the middle of the gap. Therefore a neutral soliton has spin one-half. A low energy charged soliton state corresponds to
removing the unpaired soliton from 0 or adding a second electron (which might
for example stem from doping the material). Therefore a charged soliton has spin
zero. We have therefore the following unusual charge-spin relations:
Q0 = 0,
Q = e,
1
2
s = 0.
s0 =
(6.90)
(6.91)
170
Topology in Physics
Figure 6.14: Change of density of states due to the presence of a soliton. The
gap center state is a function of strength unity. Illustration of [40].
The missing electron density at each site is compensated by the electronic density
|0 (n)|2 . To see this we use the fact that the local density of states n,n (E) satisfies
the sum rule
Z
n,n (E)dE = 1
(6.92)
We have also the symmetry n,n (E) = n,n (E) which leads us to the local
compensation formula
Z
2
n,n (E)dE + |0 (n)|2 = 0.
(6.93)
Because the distribution of the electronic density is not influenced by the occupancy of the state, we have also that the charge distribution of the charged
soliton is |0 (n)|2 .
Numerical calculations for n,n (E) are shown for n = 0, 6, 12 and n = 1, 5, 11
in Figure 6.15 for a soliton centered at n = 0. We see that the density of states
vanishes for n odd.
This is an exact result and can be shown by considering E0 = 0 which reads,
written out,
..
..
0
tn+1,n
0
t
H0 = n1,n
tn,n+1
0
tn+2,n+1
0
...
...
0
0
..
.
0 (n)
= 0.
0 0 (n + 1)
..
...
.
(6.94)
In this form we can see that we have two uncoupled recursive equations for 0 (odd)
and 0 (even). For the center at even n, only 0 (even) is normalizable, and vice
171
Figure 6.15: Change of local density of states due to the presence of a soliton for
various distances from the soliton, even left, odd right. Both pictures from [40].
Figure 6.16: Shape of a soliton centered at n = 0, illustration from [40].

versa.
The shape of the soliton for an even center is

n
1
1
0 (n) u sech
cos
n ,
l
l
2
(6.95)
which is depicted in Figure 6.16.

Third-filled band case
This section draws from [42].
We assume now a third-filled band in polyacetylene, reusing the concepts from
above. The ground state of this system is threefold degenerate, shown diagrammatically in Figure 6.17. We have now two types of kinks, type I from A phase
to B phase (and from B to C or C to A respectively) and type II from B to A
(and from A to C or from C to B). We can calculate the charge of a type I kink
using the fact that the chain gets shorter by a type I kink (see Figure 6.18). Since
the phase shift is 2 in total a charge of +2|e| flows into the region where the
three kinks have been formed, therefore each type I kink has a charge of + 23 |e|
due to symmetry. This result can be checked by Greens function calculations
172
Topology in Physics
Figure 6.17: Three degenerate ground states for a trimerised chain, long bonds
are dotted. Image from [42]
Figure 6.18: Above: Standard trimerised chain. Below: Trimerised chain with
three type I kinks (each contracted to a single point for the illustration).
or numerical calculations. By the same argument as before one can see that a
type II kink has a charge of 23 |e|. Naturally, one identifies the type II kinks
as the antiparticles of the respective type I kinks. In surprising analogy with
hadron and quark structure, one can then see that either three (anti)solitons or
a soliton and an antisoliton can be created without disturbing the ground state
of the chain far away.
Appendix: Greens Functions Introduction and Some

Useful Formulas
This section derives some formulas used in [40], using [43] referenced therein.
Greens Function for t > 0
We set the starting time t0 = 0 and define the Greens function as
= ieiHt
G(t)
t>0
(6.96)
= 0 for t < 0.
and we set G(t)
This definition corresponds to a definition using time-ordered creation and annihilation operators. We take the momentum basis:
E
D

p1 (t)
ap2 (0) 0
(6.97)
p1 G(t) p2 = i 0 T a
173
where T is the time-ordering operator. Therefore for t > 0:

E
D

p1 G(t) p2 = i 0 eiHt ap1 eiHt ap2 0 = i p1 eiHt p2
(6.98)
and for t < 0:

i 0 ap2 eiHt ap1 eiHt 0 = i 0 ap2 eiHt ap1 0 = 0 eiHt 0 = 0
(6.99)
We define the Fourier transform as

Z
f () =
f (t)eit dt.
(6.100)
We recall that ~ = 1 in our units, in particular = E, therefore the Fourier

+ i), = 0+ is
transform of the Greens function G(E
Z
i(E+i)t
+ i) =
G(t)e
G(E
(6.101)
Z
=
iei(H(E+i))t dt
(6.102)
Z
=
iei(H(E+i))t dt
(6.103)
0

(6.104)
= [H (E + i)]1 ei(HE)t et 0
= [H (E + i)]1 .
(6.105)
We also
P note that if we write out the Hamiltonian in terms of eigenfunctions,
H = k Ek |k i hk | we immediately see that
X
+ i) = [H (E + i)]1 =
[Ek (E + i)]1 |k i hk | .
(6.106)
G(E
k
We make use of the distribution equation

1
= principal part
lim
70,>0 x + i

1
i(x)
x
(6.107)
to see that

X
1
+ i)) = 1 =
= Tr(G(E
[Ek (E + i)]1
!
=
(Ek E)
which is the one-electron density of states (E) for the Hamiltonian H.
(6.108)
174
Topology in Physics
Gap energy E (eq. 6.81)

We have the unperturbed Hamiltonian H0 with eigenvalues j . The perturbed
has eigenvalues Ej .
Hamiltonian H = H0 + V
We use the replacement of the energy E by E i, = 0+ .
The Greens Function of the unperturbed Hamiltonian H0 is defined as [E i H0 ]1
Y E i Ej
det [E i H0 ]1 [E i H] =
E i j
j

h
i
= det [E i H0 ]1 E i H0 V

0
= det 1 G (E)V
(6.109)
(6.110)
(6.111)
which gives us
"
#!

1
X
X
1
= E ln det 1 G0 (E)V
= = E
ln(E i Ej )
ln(E i j )
j
j
X
X
1
1
1
=
(E i Ej )
(E i j )
j
j
= (E) 0 (E)
(6.112)
(6.113)
(6.114)
where (E) is the density of states with the full Hamiltonian H, 0 (E) with the
unperturbed Hamiltonian H0 .
We choose the energy eigenvectors 0 of H0 as a basis, therefore G0 (E) will
does only have matrix elements between some eigenvectors,
be diagonal
V
if E
D and
0 0
has nondiagonal
means n V m 6= 0 iff n, m [nmin , nmax ], 1 G0 (E)V
elements only in the submatrix [nmin , nmax ], therefore the determinant is that of
the submatrix.
We assume conservation of electrons, therefore we have
Z EF
Z EF0
(E)dE =
0 (E)dE.
(6.115)
If we assume EF < EF0 (EF0 > EF is analogous), then we have

Z EF
Z EF0
((E) 0 (E)) dE =
0 dE (EF0 EF )0 (EF ),
(6.116)
EF
because we take the system to be large and the potential not too big, which
implies that EF EF0 will be small compared with EF .
175
We write down the energy difference, inserting a spin factor of 2,

E
=
2
EF
E (E)dE
0
EF
E 0 (E)dE
EF
E((E) 0 (E))dE
|
Z
EF
0
R EF
EF
E 0 (E)dE
0
EF
E 0 (E)dE
(6.117)
(6.118)
EF
0
EF
EF
E 0 (E)dE
{z
}
0 (E)dE=EF
(6.119)
R EF
((E)0 (E))dE
EF
=
Z
0
EF
E (E)dE
EF
=
Z
(E EF )((E) 0 (E))dE
(6.120)

1
dE,
(E EF ) = E ln det(1 G0 (E)V)
EF
(6.121)
where we integrate by parts to eliminate the partial derivative with respect to

energy, the boundary term vanishes at E = because G0 (E) is suppressed as
E 1 for large energy:
EF
Z EF

E
1
1
0

= (E EF ) = ln det(1 G (E)V)
= ln det(1 G0 (E)V)dE

2
(6.122)
Z EF
2
= ln det(1 G0 (E)V)dE.
(6.123)
E =

176
Topology in Physics
Vortices
Thomas Eggel
Supervisor: Stefan Fredenhagen
The topological structure of solutions of the Ginzburg-Landau
equations is the main concern of this report. A brief introduction
is to serve as a physical orientation of where the investigations
are to take place. The GL energy functions will be discussed
in the global and gauged theory. The topological possibilities
in both cases are examined and a scaling argument is shown to
provide useful information in the global theory. By the virtue of
the Principle of Symmetric Criticality invariant solutions under a
combined SO(2) action in the gauged theory are shown to exhibit
interesting behaviour. In the end interaction between vortices
and a possible extension to a dynamic theory are investigated.
Introduction
The Landau theory of second order phase transitions is based on the assumptions
that a phase transition can be characterized by some kind of order parameter and
by a simple postulated dependence of the free energy on this order parameter.
The crucial insight concerning superconductors was made by Ginzburg and Landau in 1950 [44] as they stated :
We shall start from the idea that (the order parameter) represents some
effective wave function of the superconducting electrons.
This implies that the order parameter is complex and is generally varying in
space.
The free energy was then postulated to be of the form
1
f (x) = fn + |(x)|2 + |(x)|4
2
to which a gradient term is added to account for energy arising due to the
177
Topology in Physics
Order Parameter
178
T
Tc
Figure 7.1: Order parameter close to Tc
spatial variation of the field, which reads

1
~2
f (x) = fn + |(x)|2 + |(x)|4 +
|(x)|2 ,
2
2m
see [45]. Vortices are now, heuristically speaking, soliton solutions of the field
equations obtained by variation of an energy functional in which the above form
of the free energy (1) is incorporated, which exhibit a certain behaviour. Namely
a field configuration is considered to contain a vortex at a point x if
x is an isolated point where vanishes
along a small circle enclosing x counterclockwise, the phase of increases
by 2N , N Z \ {0}.
N is then called the multiplicity of the vortex. To get a geometrical idea of what
a vortex field configuration looks like, one might find it useful to consider figure
7.2 as a first, heuristic indication of the form of a vortex.
The Ginzburg-Landau energy functions
Vortices are soliton solutions of the Ginzburg-Landau equations. These are the
equations one obtains by variation of the Ginzburg-Landau energy functions
Vortices
179
Figure 7.2: The winding of a field around a zero (vortex)
which are as a matter of fact functionals defined on the space of all field configurations. There are two types of field theories to be discussed: a global theory in
which there is only the scalar field (x) present and the local theory in which the
the scalar field (x) couples to an electromagnetic field with gauge group U (1).
The scalar field , which is a complex quantity, shall be written as
(x) = 1 (x) + i2 (x).
The Global Theory
In the global theory, the GL energy function is given by the following expression

Z
1
V V () =
+ U (.)
d2 x
(7.1)
2
where U denotes the part of (1) that depends on ||. For further investigation
it will prove useful to chose the following form for U
+ ()
2,
U = +
8
, , R.
(7.2)
180
Topology in Physics
The functional (7.1) is invariant under a global U (1) phase rotation
(x) 7 ei (x),
(7.3)
If we require that a lower bound for the energy exist, must be chosen such
that > 0. The coefficient can be adjusted such that the minimal value of U ,
Umin is zero. If furthermore is chosen < 0, equation (7.2) can be written in the
following form
2
U = (m2 )
(7.4)
8
and the GL energy function reads

Z
1
2 2
V =
+ (m ) . d2 x
(7.5)
2
8
We shall denote the vacuum manifold by V and thus, in the present case the
vacuum manifold is just the circle |(x)| = m. Thus
1 (V) = 1 (S 1 ) = Z,
(7.6)
which, we will find out, will give rise to the possibility of existence of vortices.
Variation of (7.1) with respect to yields the following field equation
2 + (m2 )
= 0.
2
(7.7)
The vacuum solutions are of the form (x) = mei and since the gradient
energy has to vanish, = constant and thus the choice of spontaneously
breaks the global U (1) symmetry.
The Gauged Theory
In the gauged theory, the scalar field couples to the electromagnetic field and the
GL energy function reads

Z
1
2 2 2
2
V =
B + Di Di + (m ) d x,
(7.8)
2
4
where B = f12 = 1 a2 2 a1 is the magnetic field and Di = i iai the
covariant derivatives.
The GL energy (7.8) is invariant under the following gauge transformation
(x)
7
ei(x) (x)
ai (x)
7
ai (x) + i (x).
(7.9)
(7.10)
ei(x) is a spatially varying phase rotation. Variation of (7.8) with respect to

,ai yields the following field equations
Vortices
181
Di Di + (m2 )
= 0
2
i
ij j B + (D
i Di ) = 0
2
(7.11)
(7.12)
We note that equation (7.12) has the form of Amp`eres equation in two di i Di ) being interpreted as the electric
mensions, with the quantity ji = 2i (D
current in the plane.
In order that a solution be a vacuum solution, the components of the integrand
of (7.8) need to vanish separately. Thus the energy is minimized if
|(x)| = m
B = 0
Di = 0.
(7.13)
(7.14)
(7.15)
(x) = mei(x) from (7.13)

ai (x) = i (x) from (7.15) .
(7.16)
(7.17)
(i i )ei
(7.18)
This requires that
Di vanishes if
i.e. (x) = (x) + constant . The vacuum is thus of the following form
(x) = mei(x) , ai (x) = i (x),
(7.19)
which, by the gauge transformation ei , can be transformed to the simple vacuum

= m, ai = 0.
Derricks Theorem
In [46] G.H. Derrick noted that in many theories that have an energy functional
of the form
Z
E() = ()2 + f ()
(7.20)
for any f (), the equations derived from the variation of the energy functional
have no stable, time-independent, localized solutions. By localized it is meant
that the integral (7.20) converges when taken over all space. This is shown
as follows: Suppose is a localized solution in 3 dimensions of the variation
E = 0. Define the scaled field Rconfiguration () (x)R = (x) with some
arbitrary constant and write E0 = f ()d3 x and E2 = ()2 d3 x and write
Z

1
1
(7.21)
e() =
( )2 + f ( ) d3 x = E2 + 3 E0
182
Topology in Physics
where the second identity is obtained by a simple change of variables. Therefore

de()
= E2 3E0
(7.22)
d =1

d2 e()
= 2E2 + 12E0 .
(7.23)
d2
=1
Now since is a solution of the variation E = 0 for = 1, it follows that

de()
= 0
(7.24)
d =1
1
E0 = E2
(7.25)
3

d2 e()
= 2E2 < 0
(7.26)
d2 =1
Hence () is not stable with respect to rescaling since the second variation
is always negative and thus is unstable. If, as it occurs in our case, f () 0,
then the variation E = 0 has no nontrivial localized solutions, be it stable or
unstable, at all, for in this case both quantities E0 and E2 are non-negative and
thus have to vanish separately which then yields 0.
Application of Derricks Theorem
In this subsection we will apply Derricks Rescaling Theorem to the global and
the local theory. We will see that in the global theory, Derricks Theorem provides
us with useful nonexistence information whereas in the local theory, the theorem
will not prove so useful. Note that we are working in two spatial dimension which
is different to the framework in which we have introduced the theorem, but this
will not give rise to any significant changes.
The usefulness of Derricks Theorem depends crucially on the way the rescaled
quantities are defined. For the scalar field we define
() (x) = (x),
(7.27)
which means that the gradient takes the following form

() (x) = ((x)) = (x).
(7.28)
For the gauge potential we will chose a form that ensures that the covariant
derivative behaves similarly under spatial rescaling as the gradient. Thus we
define
A() (x) = A(x)
(7.29)
Vortices
and therefore
183
()
DA () (x) = DA (x)
(7.30)
and the field strength tensor acquires a second order term

F () (x) = 2 F (x).
(7.31)
Derricks Theorem in the Global Theory

We have seen that the vacuum solutions are of the form (x) = ei with constant. The question that is now arising is if there is other finite energy solutions
not identical to the vacuum. The answer will be no and this will be shown using
Derricks Theorem.
Consider equation (7.5) and write
Z
2 d2 x
E0 =
(m2 )
(7.32)
8
and
Z
E2 =

1
d2 x
2
and thus
e() = E2 +
1
E0 .
2
(7.33)
(7.34)
If finite energy non-vacuum solutions existed, they would need to satisfy

Z
2 d2 x = 0,
E0 =
(m2 )
(7.35)
8
so || = m everywhere. Let = mei and substitution into (7.7) yields
2 = 0 and = 0,
(7.36)
therefore = constant and thus is a vacuum.

Derricks Theorem in the Local Theory
In the gauged theory, Derricks Theorem does not rule out finite energy solutions
other than the vacuum. Write (7.8) as

Z
1
2 2 2
2
V =
B + Di D + (m ) d x = E4 + E2 + E0 .
(7.37)
2
4
Hence the energy as a function of reads
e() = 2 E4 + E2 + 2 E0 .
(7.38)
184
Topology in Physics
Differentiation with respect to yields
d
e() = 2E4 23 E0 .
d
Taking this expression at = 1 yields
Z
Z
1
2 2
2 d2 x,
E0 = E4
B d x=
(m2 )
2
8
(7.39)
(7.40)
the two contributions to the energy are equal. But this is by no means as restricting a condition as (7.35).
Topological Structures
Both in the global and local theory topological structure can be found. In the
global theory the requirement of finite energy will show to be too strict for vortices
to be allowed, whereas in the local theory there will be vortex solutions that are
of finite energy.
Topology in the Global Theory
It is convenient to express (7.5) in polar coordinates x1 = cos and x2 = sin
:

Z Z
1 2
1
2 2
V =
+ 2 + (m ) dd.
(7.41)
2 0
4
0
Consider a field configuration whose energy density approaches 0 rapidly as
|x| . Thus
|| m and 0 as
(7.42)
Assume the limit lim (, ) exists and denote it by
()
() = mei
(7.43)
The map
: S S 1
(7.44)
is a map from the circle at infinity to S 1 . Since S 1 = V, is a map from

the circle at infinity to the vacuum manifold. Since we assume to be single
valued, must satisfy
(2) = (0) + 2N
N N
(7.45)
Let us now have a closer look at the contribution of to the energy density. The
is O( 12 ) provided () is differentiable. Let 0 be sufficiently
term in 12
large, so that (, ) can be approximated by
(, ) = mei .
(7.46)
Vortices
185
Since = i m ei and = m2 ( )2 the contribution to

the energy outside of circle of radius 0 , denoted by E,0 is given by
Z Z
1 2 2 1
E,0 = m
( )2 dd.
(7.47)
2
0
0
The integrals in E,0 separate and the radial integral is logarithmically divergent unless = 0. In this case the field configuration would have a limit
that is of the form
() = mei
with = const.
(7.48)
For this field configuration with constant phase the winding number satisfies
N = 0. Thus the requirement of finite energy does not allow for solutions that
are topologically distinct to the vacuum. We would like to mention that if we
were prepared to relax the condition of finite energy and allow solutions with
logarithmically divergent energies, the radial part in (7.47) would not have to
vanish and neither would N . Solutions with N 6= 0 do have a topological structure
that is distinct to the topological structure of the vacuum and thus allowing for
solutions with logarithmically divergent energies in the global theory leads to
vortex solutions.
Topology in the Local Theory
Again we consider the GL energy in polar coordinates x1 = cos and x2 = sin
and (7.8) becomes

Z Z
1 2 1 2
1
2 2
V =
f + D D + 2 D D + (m ) dd.
2 0
2
4
0
(7.49)
The covariant derivatives are given by
D = ia
D = ia .
(7.50)
(7.51)
We will now seek to find a limiting form of an arbitrary finite energy field configuration . Let {(x), ai (x)} be a finite energy field configuration. The requirement
of finite energy, V < , imposes the boundary condition
|| m as |x| .
(7.52)
The different terms in the integrand of (7.49) have to vanish separately, so it

follows that
D 0 as |x| .
(7.53)
186
Topology in Physics
Let us now fix an angle and move out on a radial line 0 and suppose
is large. Asymptotically = mei and thus
D = im ( a )ei = 0.
(7.54)
The term in parentheses has to vanish and therefore a = . We now transform the field to the radial gauge a = 0 by a suitable gauge transformation.21
In the new gauge, 0 as |x| , so
()
lim (, ) = () = mei
(7.55)
which then defines on the circle at infinity i.e. a limiting form of along each
radial line. Furthermore, the two quantities
Z Z
1
c1 =
D D dd
(7.56)
2
Z Z
1 2
c2 =
f dd
(7.57)
2
need to satisfy c1 < and c2 < . From the boundedness of c1 it follows that
D 0 as
(7.58)
and from the boundedness of c2 it follows that

f 0 as
(7.59)
Since f = a a , it follows from (7.59) that a = a and since we

are working in the radial gauge this is equivalent to a = 0, so a has a limit
lim a (, ) = a
().
(7.60)
This defines the gauge potential on the circle at infinity. Equation (7.58)
implies
a
= 0.
(7.61)
So since the gauge potential is not necessarily vanishing on the circle at infinity,
the derivative of the phase at infinity is not necessarily vanishing i.e. = (),
not necessarily constant. This is a tremendous difference to the global theory in
which under the requirement of finite energy, the phases of the field configurations
need to be constant.
Again
1
: S
S 1
(7.62)
21
Such a gauge Rtransformation

can always be found: if a 6= 0 then the gauge transformation
g(, ) = exp i 0 a (0 , )d0 is an appropriate transformation.
Vortices
187
is a map from the circle at infinity to the vacuum manifold V = S 1 and the
winding number N is an integer given by
1
( (2) (0))
(7.63)
2
It is possible to uncover another nature of the quantity N which is often called
the topological charge of the field configuration {(x), ai (x)}. The first Chern
number c1 is, in the framework of our theory, defined to be the following quantity
Z Z
1
c1 =
f12 dx1 dx2
(7.64)
2
N=
where f12 = 1 a2 2 a1 is the magnetic field in the plane. Now this can be
written by Stokes theorem for differential forms as a line integral along the circle
at infinity
Z 2
Z 2
1
c1 =
a d|= =
a ()d.
(7.65)
2 0
0
By the fundamental theorem of calculus, (7.63) can be written as
Z 2
Z 2
1
N=
()d =
a ()d
2 0
0
(7.66)
where in the last identity equation (7.61) was used. From (7.65) it also becomes
obvious that 2N is the total flux of the magnetic field that traverses the whole
plane. Thus we see that the magnetic flux is quantized in units of 2.
There is also a third topological characterization of the winding number N .
It can be regarded as the total vortex number in the plane, i.e. the total amount
of points in the plane where the field vanishes with the respective multiplicities of
the zeros taken into account. To see this we assume that the set of points where
the field vanishes,
{x|(x) = 0} {A, B, C . . .}
(7.67)
is finite and that these zeros are isolated. We denote by {nA , nB , nC , . . .} the
multiplicities of the zeros. By the geometric argument provided by figure 7.3, it
is obvious that the winding number of the field along the circle at infinity is
the sum of the multiplicities
N = nA + nB + nC + . . . .
Vortex Solutions
All known finite energy, static solutions of the field equations have
circular symmetry ( 6= 1) about some point and
(7.68)
188
Topology in Physics
A
C
A
C
A
AB
B
Figure 7.3: Zeros of and different paths

a reflection symmetry .
In order to obtain information about the behaviour and the form of the fields and
the energy density about a zero of , we will consider solutions that are invariant
under a particular SO(2) action. To ensure that the fields obtained in this way
are really solutions, we will invoke the Principle of Symmetric Criticality.
Principle of Symmetric Criticality
The following principle allows one to conclude that if a variational principle is
invariant under some symmetry group G, then to test if a field configuration
that is invariant under a subgroup of the symmetry group is a stationary value
it suffices to investigate the vanishing of the first variation of the action with
respect to variations that are also symmetric, see [47] We will state it following
[48]:
Vortices
189
Let be the fields of some Lagrangian field theory with action S() and
let C denote the space of all field configurations . Let K be a subgroup of the
symmetry group, and let CK C denote the configuration space of all K-invariant
fields, ie. fields such that
k() = k K
(7.69)
Let denote SK () denote the action of the theory restricted to CK and let 0 (x)
CK be a field configuration that extremizes SK .
Then the Principle of Symmetric Criticality states that 0 (x) is automatically
a stationary point of the full action S
Solutions Invariant under Combined SO(2) Action
We will construct solutions that are invariant under a certain combined SO(2)
action. Consider the SO(2) action by the combined rotations and phase rotations
)), N Z i.e.
(R(), R(N
R()
: 7 ei
(7.70)
)) is a phase roand R() the usual rotation. Fix m = 1. Thus (R(), R(N
iN
tation e
accompanying a rotation by , which is a global transformation as
it is independent of , . This combined SO(2) action leaves the field invariant,
provided
(, ) = eiN ()
a (, ) =
a ()
(7.71)
a (, ) =
a ()
We now gauge transform a 7 0. Because of the reflection symmetry (, ) =
) we have that () is real.
(,
The following boundary conditions ensure that at = 0 is single valued and
the gauge-potential non-singular and that at = we have
D = ia = (iN iN )eiN = 0.
(7.72)
Since (0) = 0 there is a vortex 22 located at the origin. Furthermore because

of (7.71) the phase of increases by 2N and since a
= N the total magnetic
flux is given by
Z
Z
2
a d =
0
N d = 2N
(7.73)
We thus obtain a GL energy of the following form

!

2 2
Z

1 da
d
1
2
V =
+
+ 2 (N a )2 2 +
1 2
d,
2
d
d
4
0
(7.74)
22
or multivortex
190
Topology in Physics
Figure 7.4: () (solid curves), a () (dashed curves) for N = 1 with =

0.5, 1.0, 2.0. From [48].
which gives rise to the field equations

d2 1 d
1
+
2 (N a )2 + (1 2 ) = 0
2
d
d
2
2
d a 1 da
+ (N a )2 = 0.
2
d
d
(7.75)
(7.76)
These equations have been solved numerically. Figure 7.4 shows the scalar
field and the gauge field a as functions of for different values of . Figure 7.5
shows the energy density and the magnetic field for a N = 1 vortex with = 1.
Note how the energy density as well as the magnetic field B are peaked about
the location of a vortex. The existence of solutions of (7.75) and (7.76) has been
proven for all N 6= 0 and all > 0.
Forces between Gauged Vortices
First we will give some general results concerning the forces that act between
gauged vortices and then we will give an analytical expression for the forces
between gauged vortices at large separation. We will assume in this subsection
that is of the order of 1.
General Results
Generally one says that there is an attractive force acting between two vortices
if the energy of the two-vortex field configuration increases as the separation between the vortices increases. Similarly, one says that there is a repelling force
acting between two vortices if the energy of the two-vortex field configuration
Vortices
191
Figure 7.5: Energy density (solid curve) and magnetic field B (dashed curve) for
the N = 1 vortex with = 1. From [48].
decreases as the separation between the vortices increases. From all the different ways of investigating the energy of multi-vortex configurations the following
results emerge: For
< 1 Vortices attract
> 1 Vortices repel
= 1 Vortices are in a neutral equilibrium
In [49] the vortex interaction energy was calculated numerically for a twovortex configuration as a function of the separation of the vortices. The results
for = 1.3, 1.0, 0.7 are shown on figure 7.6
The different values for can be associated to the physical behaviour of a superconductor. Recall the magnetization curves of a type I/II superconductor as
shown on figure . In a type I superconductor, the magnetic field is expelled until
it reaches some critical value Hc . From this exterior field strength on, the magnetic field completely penetrates the superconductor, which leads to the dramatic
discontinuity of the magnetization curve at H = Hc . This type of superconductor
corresponds to < 1. The vortices attract and the overlapping of the vortices
cause the magnetization to break down. A type II superconductor exhibits a
different behaviour. Until the exterior field strength has attained some first critical value Hc1 , a type II superconductor behaves like a type I superconductor.
Once this first threshold value is surpassed, the magnetic field starts to penetrate
the superconductor in flux quanta. This corresponds to > 1, i.e. to the case
where the vortices repel. They arrange in a regular lattice since experimental
superconductors are of finite size.
192
Topology in Physics
Figure 7.6: Energy of a two-vortex field configuration as a function of the vortex

separation. From [49].
Forces between Gauged Vortices at large Separation

This case is interesting because there exist an analytic form for the asymptotic
behaviour of the two vortex configuration. Let Eint (s) denote the interaction
energy between two N = 1 vortices at a separation s >> 1. This is given by the
total energy minus twice the energy of infinitely separated vortices
Eint (s) = Etot 2E1 .
(7.77)
There is two contributions to the energy: A contribution due to the scalar field
and a contribution coming from the magnetic field. The result is given by
A2s
A2
K0 ( s) + m K0 (s),
(7.78)
2
2
K0 denoting the zeroth order modified Bessel function of the second kind and Am
and As coefficient that correspond to the magnetic and scalar field respectively.
Figures 7.8 and 7.9 show plots of (7.78) for two different values of .
Eint (s) =
Vortices
193
<1
>1
M
Hc
Hc1
Hc2
Figure 7.7: Magnetization curve of type I and II superconductors
Dynamics - Scalar Electrodynamics Model
The static GL equations determine how to energy of a configuration of several

vortices depends on their separations. But this does not determine how the
vortices will move. We will discuss a possible extension of the static gauged
theory by the well known Abelian Higgs or Scalar Electrodynamics model. The
Lagrangian of the Abelian Higgs model is given by

Z
1
L=
(7.79)
f f + D D (1 ) d2 x.
4
2
8
We note that in addition to the gauged static GL energy as given in (7.8),
with m = 1, we now consider also a kinetic energy part
Z

1
e21 + e22 + D0 D0 d2 x
T =
(7.80)
2
with e1 , e2 components of the electric field and D0 the time covariant derivative.
The Euler-Lagrange field equations are given by
D D (1 )
= 0
2
i
D ) = 0.
f + (D
2
(7.81)
(7.82)
The conserved quantities of the field configurations satisfying these field equations are the topological charge or winding number N and the conserved geometrical Noether charges. These include the momentum, the angular momentum and
the energy. We will only discuss the momentum, for the other quantities we refer
the reader to [48]. Momentum is associated with the translational invariance of
the Lagrangian. An infinitesimal translation in the xi direction combined with
194
Topology in Physics
0.25
Eint
0.2
0.15
2Eint
0.1
0.05
0
0
4
s
Figure 7.8: Asymptotic behaviour for = 1.5
an infinitesimal gauge transformation with parameter ai yields the following

gauge variations
= i iai = Di
(7.83)
= i + iai = Di
aj = i aj j ai = ij B.
This yields the following form for the conserved momentum

Z
L
L
L
2
Pi =
+
+ (0 aj ) aj d x
(0 )
(0 )

Z
1
1
=
D0 Di + D0 Di + ij ej B d2 x
2
2
(7.84)
(7.85)
where L denotes the Lagrangian density. The conserved electric charge is given
by
Z
i
0 D0 )d2 x.
(D
Q=
(7.86)
2
Withing this framework, an N = 1 vortex behaves like a particle whose rest-mass
is given by the static GL energy. Solutions can be Lorentz boosted in order to
obtain vortices moving at arbitrary speeds up to the speed of light. It is also
possible to compose a superposition of such solutions and set vortices to collide.
The geometrical quantities involved in a two vortex collision are indicated in
figure 7.10
The 90o scattering is the most interesting case. It shows that vortices are in a
fundamental way non-Newtonian particles. Since vortices are indistinguishable,
Vortices
195
0.05
Eint
0
-0.05
-0.1
2Eint -0.15
-0.2
-0.25
-0.3
-0.35
0
4
s
Figure 7.9: Asymptotic behaviour for = 0.5
because the fields are the fundamental objects and the vortex locations are simply
the zeros of the scalar field, they can not be labeled. Thus two vortices can be
followed until they collide but as the zeros emerge from the scattering location, it
is no longer possible to tell which vortex, or which part of the outgoing vortices
corresponds to a certain incoming vortex. For further information see [48].
196
Topology in Physics
11
00
00
11
00
11
a
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
1111111111111
0000000000000
Figure 7.10: A two-vortex collision
11
00
00
11
00
11
8 Berezinskii Kosterlitz
Thouless transition in spin
systems
Lucas Lombriser
Supervisor: Munehisa Matsumoto
Spontaneous continuous symmetry breaking in two dimensions
is not possible for non-zero temperatures. Therefore, common phase transitions are not expected in a two-dimensional
xy-model. The appearance of vortices, however, induces the
Berezinski-Kosterlitz-Thouless transition. The motivation of this
chapter is to derive the arguments for such a transition and investigate some of its properties.
Introduction
Usually, one associates a phase transition to undergo spontaneous symmetry

breaking when taking its disordered phase to the less symmetric ordered phase.
We expect the several separate orientations of the order parameter below the critical temperature to be related by a symmetry operation with respect to the order
parameter. The emergence of an exclusive orientation characterizes the spontaneous symmetry breaking. There, however, is a theorem by Mermin and Wagner,
which states that there cannot be any spontaneous continuous symmetry breaking in dimension d 2, provided a non-zero temperature. This implies absence
of long-range order in a two-dimensional system with continuous symmetry. In
a spin system described by a xy-model, which is of continuous symmetry, we,
however, observe a phase transition which is known as the Berezinski-KosterlitzThouless transition.
As we will see, at low temperatures, the spin interaction between nearest
neighboring sites is subsiding algebraically, thus it is of quasi-long-range order.
197
198
Topology in Physics
For high temperatures, we, however, expect an exponential decay of the correlation, hence the spin interaction is short-ranged. Now, how can one explain this
phenomenon without contradiction to the Mermin-Wagner theorem? As we will
see, topological defects called vortices will make the phase transition possible,
and thus we can elude a continuous symmetry breaking. But before we will get
to that point, we will have to behold some more general concepts, such as the
basics of statistical mechanics and renormalization group theory.
Statistical Mechanics
A classical system of N particles which is coupled to a reservoir of fixed temperature T , while its energy can fluctuate, is described by its canonical partition
function
X
Z=
eH() ,
(8.1)
where H is the Hamiltonian and indicates a possible state. Further, we have

= kB1T , with kB the Boltzmann constant and T the temperature. The exponential eH is also referred to as the Boltzmann weight. The probability that
one particular state is occupied is then
P =
eH()
,
Z
which is also called the canonical ensemble. This defines the entropy

1
S = kB ln
,
P
(8.2)
(8.3)
where P is the probability of occurrence of a state when equally probable. S

can, however, also be derived by thermodynamic means. All thermodynamic
quantities can be determined by Eq. (8.1), starting with the free energy
1
F = ln Z.
(8.4)
Furthermore, we obtain for the internal energy of a system

U = T 2
(F/T )
T
(8.5)
and, by Legendre transformation of U with respect to S, we receive F , thus

F = U T S,
(8.6)
where the potential F is called the Helmholtz free energy. It describes a minimum
in equilibrium, provided the temperature and the volume of the system are held
Berezinskii Kosterlitz Thouless transition in spin systems

constant. The expectation value of a fluctuating quantity A is given by
P
H()
X
A()e
hAi =
A()P = P H() .
e
199
(8.7)
For the discussion of the Berezinski-Kosterlitz-Thouless transition, we have

to adopt some general concepts about phase transitions. First, we introduce the
order parameter, which is a non-zero value in the ordered phase. In the disordered
phase, however, it is identically zero. The point at which the value of the order
parameter becomes zero is the critical point, which defines the transition going
from the ordered to the disordered phase. In a ferromagnet the order parameter
is identified by the spontaneous magnetization. When a quantity exhibits
correlation, the order of a system is indicated by the correlation function
G(x) = h(x)(0)i,
(8.8)
where x is the spatial distance. The order parameter will become zero for infinite
system size, provided the correlation function tends to zero.
Another, general way of writing the correlation function is
G(x)
er/(T )
,
rd2+
(8.9)
where is a critical exponent and (T ) is the correlation length, which is a

measure of the length scale at which the behavior of a material begins to differ
distinctly from its macroscopic properties.
For systems at thermal equilibrium, we observe a first-order transition when at
the critical temperature Tc the first derivatives of the free energy with respect to a
thermodynamic variable are discontinuous. If they are continuous, but the second
derivatives exhibit discontinuity, it is termed a second-order phase transition.
They are sometimes also called discontinuous and continuous phase transitions,
respectively. In general, the correlation length of a first-order transition is finite,
whereas for a second-order transition we expect infinite correlation length.
A very well known concept is the Ising model, which considers a lattice with
classical spins si = 1 attached to each lattice site i. The Hamiltonian is given
by
X
1X
H(s) =
Jik si sk h
si ,
(8.10)
2 i,k
i
where h indicates an external homogeneous field and the constant J describes
the nearest neighbor interactions. Requiring J > 0, we obtain a model for a
ferromagnet. For an Ising model with dimension d dl , where dl = 1 is the lower
critical dimension for discrete symmetries, there cannot be any phase transition.
200
Topology in Physics
The Renormalization Group

In renormalization group procedures, parameters defining the problem are reexpressed in terms of other parameters, without changing the physics in behind.
This results in renormalization group flows in the parameter space, whose equations describe the physical problem and are the essence of renormalization group
theory.
Consider real space renormalization in lattice spin systems, e.g. block spin
transformations. Let us assume a two-dimensional Ising model which is not
exposed to any magnetic field. Now we group the spins into squares of 3 3 spins
and assign a new variable s0 = 1 to these blocks, where s0 = +1 or s0 = 1 if
there are predominantly up spins or down spins, respectively, at hand. At the
critical point, we observe a scale invariance, thus the configuration of the new
variables s0 and the one for the spins s are statistically the same. Starting above
the critical point, the statistics change, but we find, assuming nearest neighbor
interaction, that the dominant interactions will remain short-ranged, independent
on the count of iterations of the transformation. Furthermore, the partition
function, and thus thermodynamics, is not affected by the renormalization group
procedure. Large distance behavior remains the same.
The renormalization group transformation can be understood as acting on
the space of all possible couplings {K}, which is
{K 0 } = R{K},
(8.11)
where R depends on the length rescaling parameter b. Suppose there is a fixedpoint K = K at which R is differentiable. Thus, by linearizing about the
fixed-point, we obtain
X
Tab (Kb Kb ),
Ka0 Ka
(8.12)
b
where Tab = Ka0 /Kb |K=K . Furthermore, retrieving the eigenvalues i and the
left eigenvectors {ei } of T , we have
X
eia Tab = i eib .
(8.13)
a
We assume the eigenvalues to be real. Near the fixed point, we observe

X
X
eia Tab (Kb Kb )
eia (Ka0 Ka ) =
u0i =
a
a,b
i eib (Kb Kb ) = i ui ,
(8.14)
where we have defined the scaling variables

X
ui
eia (Ka Ka ),
a
(8.15)
201
which describe linear combinations of the deviations from the fixed point, hence
Ka Ka . Furthermore, we can define the renormalization group eigenvalues yi
by i = byi . The scaling variables ui are called relevant if yi > 0 and irrelevant if
yi < 0. When yi = 0, the scaling variables are marginal. A scaling variable that
is relevant will, by repeated renormalization group iterations, drift off the fixed
point, while in the irrelevant case, it will move toward zero, provided starting
sufficiently close to the fixed point. When ui is marginal, there can be made no
predictions whether the fixed point is approached or ui is driven away from it.
From such renormalization group flows, we retrieve the physics of the system.
Suppose a n0 -dimensional space describing the neighborhood of a fixed point
with n relevant eigenvalues. Thus, ignoring marginal eigenvalues, there are
(n0 n) irrelevant eigenvalues and therefore an (n0 n)-dimensional hypersurface, whose points are attracted toward the fixed point. By continuity, we can
expand this hypersurface to a finite region, which is called the critical surface
and exhibits infinite correlation length. This surface can act as a separatrix
when dividing points which tend to large values of Ka from those flowing toward
Ka = 0.
The renormalization group is universal, thus possesses a universality class,
which consists of all those critical models which flow into the same fixed point.
The xy-Model
Consider rotations in a two-dimensional plane of continuous symmetry, i.e. U (1)
or O(2). We define as the angle indicating the direction of the order parameter, which is either a two-dimensional vector hsi or a complex number
hi. The occurrence of an exclusive orientation of the order parameter is responsible for symmetry breaking. A two-dimensional lattice with a planar spin
s(x) = s(cos (x), sin (x)) attached on each site x obeys the O(2) symmetry.
This is the so-called xy-model, where spin interactions are of nearest neighbor
order. It can also be written in complex form, where the order parameter is
replaced by hi = |hi|ei .
The Free Energy in the xy-Model
Considering Ginzburg-Landau theory, we find that, in the ordered phase of the
xy-model, the free energy F has the shape of the bottom of a champagne bottle
(Fig. 8.1). Its minimum is achieved on a circle on the base of the bottle. Points on
this circle are depicted by the angle and the radius specified by the magnitude of
the order parameter. Rotations around the circle correspond to spatially uniform
changes in , and thus do not alter the free energy. However, spatially nonuniform changes in increase the free energy.
As long as there is no evidence for the contrary, we can assume the elastic free
energy Fel = F [(x)] F [ = const.] to be analytic in . Thus, the simplest
202
Topology in Physics
6z

@

@

hsi

Figure 8.1: Free energy of the ordered phase. The horizontal is described by
the order parameter of the xy-model. At temperatures lower than the critical
temperature, the free energy takes the shape of a wine bottle. Sometimes it is
also referred to as the Mexican hat potential.
form for Fel is
Z
1
dd xs [(x)]2
(8.16)
2
when employing the absence of terms of linear dependence of in the expansion
of Fel , which follows from the fact that the uniform state is minimal with respect
to any variations of the order parameter. Excitations of the order parameter
are, in magnetism, called spin-waves. Such degrees of freedom are Goldstone
modes that occur whenever a continuous symmetry is broken. The coefficient s
is referred to as the spin-wave stiffness, helicity modulus, or simply as the rigidity.
It can be thought of as the coefficient describing a parabolic dispersion relation.
Lowest state in free energy is achieved when applying spatially uniform changes
in . However, when imposing boundary conditions, we might want to consider
non-uniform solutions for , e.g.
z
= 0 ,
(8.17)
L
1
Fel =
s Ld2 2 ,
(8.18)
2
when requiring = 0 at z = 0 and = 0 at z = L. Therefore, we can define the
spin-wave stiffness in the following way:
s = lim 2L2d
L
F [0 ] F [0]
,
02
(8.19)
203
where the constraint for F [0] is = 0 at both z = 0, L and F [0 ] satisfies the

conditions mentioned previously. In d = 2 dimensions, s has the units of energy
and therefore is the characteristic temperature scale for the sample. At low
temperatures, s is of order Tc . Eq. (8.16) is the simplest form consistent with
the constraint of charge conservation and thus the invariance of the energy under
continuous global U (1) transformations. This can be perceived when taking into
account the commutativity of the microscopic Hamiltonian H of a system with
particle number N , i.e.
[H, N ] = 0,
X
,
N
(8.20)
(8.21)
where the complex order parameter is

||ei(x) h i.
(8.22)
The unitary transformation U U (1), with

U eN ,
(8.23)
leaves the Hamiltonian invariant, i.e.

UHU = H.
(8.24)
It can be shown that, in general, gauge transformations that leave the Hamiltonian invariant must conserve particle number, i.e. gauge invariance and charge
conservation are equivalent. [50]
The free energy in Eq. (8.16) can also be derived from the Hamiltonian of a
d-dimensional hypercubic lattice
X
X
si sj = Js2
cos(i j ),
(8.25)
H = J
hiji
hiji
where si and sj are classical spins of magnitude s lying in the xy-plane. Their
orientation with respect to the x-axis is indicated by i , where 0 i < 2
and hiji indicates the summation over nearest neighbors. H exhibits a continuous
symmetry, i.e. it is invariant under the transformation i i + 0 . We suppose
|i j | << 2 at sufficiently low temperatures. Thus, we can use the Taylor
expansion of the cosine,
1
cos(i j ) = 1 (i j )2 + O[(i j )4 ],
2
to yield
H = Js2
X
hiji
X
1
1 + Js2
(i j )2 .
2
hiji
(8.26)
(8.27)
204
Topology in Physics
Replacing the first term by the ground state energy E0 and writing the second
term as the sum over a that runs over all nearest neighbors with lattice sites x,
we obtain
Js2 X
H = E0 +
[(x + a) (x)]2 .
(8.28)
4 x,a
Furthermore, we assume (x) to vary little with x. Hence, using Taylor expansion
and the continuum model, we get
Z
Js2
H = E0 + d2 dd x[(x)]2 ,
(8.29)
2a
with a being the distance between nearest neighbors. The second term of the
2
expression above is equivalent to Eq. (8.16) as s = aJs
d2 in the d-dimensional
hypercubic lattice.
Suppose now a spatially uniform external field hx (x) aligned along the x-axis.
Then the additional term obtained in the Hamiltonian is
Z
Z
d
Fext = d x hx hs(x)i = dd x|s(x)| hx cos (x).
(8.30)
The minimum energy state imposes = 0 and hsi = sex , i.e. the magnetization
points also in x-direction. Thus, we observe a continuous symmetry breaking
associated with the application of an external field.
Correlation and Order
The spin correlation function, following Eq. (8.8), is
G0 (x, 0) = hs(x) s(0)i = s2 hcos[(x) (0)]i
= s2 Rehei((x)(0)) i = s2 eRe(g(x)) .
(8.31)
Note that is the phase of the fluctuating field, whereas indicated the phase of
the average order parameter. Furthermore, q is a point of the reciprocal lattice
and
Z
dd q 1 eiqx
g(x) = T
.
(8.32)
(2)d s q 2
This expression can be derived in spin-wave approximation. We impose periodic
boundary conditions and use Fourier representation to write
1 X
q eiqx .
(x) =
N q
(8.33)
205
Eq. (8.29) becomes

Js2
H = E0 + d2
2a
1
d x
N
d
X
q,q0
qq 0 q q0 ei(q+q )x
0
Z
1 Js X 0
0
= E0
qq q q0 dd xei(q+q )x
d2
N 2a
q,q0

d X
Js2 a2 2
1
qq 0 q q0 (d) (q + q 0 ),
= E0
2
a
N q,q0
2
(8.34)
where, in Eq. (8.34), we have used the Dirac Delta function,

(d)
(q) =
1
2
d Z
dd xeiqx .
(8.35)
Furthermore, we transform
a d Z
1 X
dd q 0
N q0
2
(8.36)
and we obtain
Js2 a2 X
H = E0
N
qq
2
q
Z
dd q 0 (d) (q + q 0 )q 0 q0
Js2 a2 X 2
q q q
2
q
X0
= E0 + Js2 a2
q 2 (q2 + q2 ),
= E0 +
(8.37)
P0
where q = q + iq = (q ) and
is the sum over half the Brillouin zone.
Now we can evaluate the expectation value with methods of statistical mechanics,
i.e. Eq. (8.7), explicitly
P H i((x)(0))
e
q e
i((x)(0))
P H
he
i=
,
(8.38)
q e
where = kB1T and H is determined by Eq. (8.37). Furthermore, we define
= Js2 a2 and consider the transformation given in Eq. (8.36) as well as Gauss
integration over the decoupled wave number space variables. Thus, for the de-
206
Topology in Physics
nominator, we find
X P 2 2
X
eH = eE0
e 2 q q |q |
q
= eE0
XY
q
YZ
2
|
e 2 q
q |2
q 2 |q |2
2
d
e
q
N0 q
Z
Z
1 Y
2 | |2
q
q 2 |q |2
q
2
2
d
e
=
d
e
q
q
0
N q

1 Y 2
=
,
N 0 q q 2
=
(8.39)
where N 0 is a constant, containing eE0 and the normalization factor of the

continuity transform. The numerator is determined as follows:
X
X 2 2 i 1 P (eiqx 1)
eH ei((x)(0)) = eE0
e 2 q |q | e N q q
q
= eE0
P0
2
|
q |2 ei N
2
|
q |2 ei 2N
e 2 q
= eE0
e 2 q
q [q (e
iqx 1)+ (eiqx 1)]
iqx 1)+ (eiqx 1)]
q [q (e
Z
2
1 Y
2 i 1 (cos q x1)
=
dq e 2 q |q | e N q
0
N q
Z
2
2 i 1 sin q x
dq e 2 q |q | e N q

1
1 Y 2
12 N
[(cos q x1)2 +(sin q x)2 ]
2q
=
e
0
2
N q q

1 P 1cos q x
1 Y 2
N
q
q2
=
e
.
N 0 q q 2
(8.40)
Thus, Eq. (8.38) becomes

hei((x)(0)) i = e
kB T
N Js2 a2
1cos q x
q2
Therefore, using Eqs. (8.31) and (8.36), we get

Z
kB T ad2
dd q 1 cos(q x)
Re(g(x)) =
,
Js2
(2)d
q2
(8.41)
(8.42)
207
which corresponds nicely to Eq. (8.32) as s = aJs

d2 and the Boltzmann constant
is set kB = 1. In d = 2 dimensions, g(x) becomes
Z 2
T
d Q 1 eiQx
g(x) =
s
(2)2
Q2
Z
T
dq d 1 eiq|x| cos
(8.43)
=
,
2s
2
q
where we have used Q = Rq, with q = (q cos , q sin ), R SO(2), and RT x =
(|x|, 0), hence Q x = (Rq) x = q (RT x) = q|x| cos . Thus, Eq. (8.32) in
d = 2 is written as
T
I(|x|),
2s
Z 2

Z

dq 1
iq|x| cos
I(|x|) =
d 1 e
,
q 2 0
0
g(x) =
(8.44)
(8.45)
where is the high wave number cutoff. Using the Bessel function of order 0
Z
Z 2
1
1
iq|x| sin
J0 (q|x|) =
e
=
eiq|x| cos ,
(8.46)
2
2 0
Eq. (8.45) becomes
Z
I(|x|) =
1 J0 (q|x|)
dq,
q
(8.47)
which can be written in terms of u = q|x| and is broken into three parts,
Z 1
Z |x|
Z |x|
1 J0 (u)
du
J0 (u)
I(|x|) =
du +
du
u
u
u
0
1
1

= ln |x| + + O (|x|)3/2 .
(8.48)
For large |x|, becomes
Z 1
Z
1 J0 (u)
J0 (u)
=
du
du 0.1159.
u
u
0
1
(8.49)
This quantity can be derived by evaluation of the following sums:

k
Z 1
1X
1
1 J0 (u)
k!
du =
0.1212,
(8.50)
3
u
8 k=0 [(1 + k)!]
4
0
!
k
Z
m
X
1
1
J0 (u)
1X
k!
du = lim
ln m
3
m
u
k
8 k=0 [(1 + k)!]
4
1
k=1
ln 2
0.5772 0.1212 0.6931,
(8.51)
208
Topology in Physics
where the Euler-Mascheroni constant, lim
Pm
1
k=1 k

ln m 0.5772, is also
denoted as . In the case of a square lattice, = + 21 ln 8, thus 1.6169. For

d = 2 and |x| large, using Eqs. (8.44) and (8.48),
T
T
I(|x|) =
[ln (|x|) + ]
2s
2s
T
T
ln (e |x|) =
ln (|x|),
=
2s
2s
g(x) =
(8.52)
= e . We perceive that g(x) exhibits logarithmic divergence. Thus,

where
Eq. (8.31) becomes
G0 (x, 0) = s2 eRe(g(x)) = s2 e 2s ln (|x|)
T
2
s.
= s2 (|x|)
T
(8.53)
(8.54)
This can be written as
G0 (x, 0) = s2 (|x|)
,
T
=
,
2s
(8.55)
(8.56)
which describes an algebraically decaying correlation for |x| large. Such powerlaw decay in the order parameter correlation functions is referred to as quasilong-range order (QLRO). Eq. (8.55) is in agreement with the Mermin-Wagner
theorem and we observe that for T > 0 the system appears to be in a critical
phase. At high temperatures, however, we expect that
G(x)
|x|
s
2T
e|x|/ ,
(8.57)
where 1 = ln(2T /s ).
Thus, there must be a phase transition, despite the Mermin-Wagner theorem.
It is carried out by topological defects called vortices. These are another kind
of excitations, different from the spin-waves used above. They produce a phase
transition that is independent of spontaneous symmetry breaking.
Vortices
Assume a d-dimensional space and consider the magnitude of the order parameter
hs(x)i = s(cos (x), sin (x)) to be periodic in (x). We may encounter situations
in which hs(x)i is continuous everywhere but in a subspace of dimension ds ,
with ds < d, e.g. we suppose a singularity at the origin. This singularity can be
removed by no longer defining the angle at this point and having the magnitude
of the order parameter tend toward zero at the origin.
209
Figure 8.2: Vortex of unit strength in the xy-model. changes by 2 when

encircling the singularity counterclockwise with a large closed curve. A closed
curve, not containing the vortex core, exhibits no change in .
In d = 2, we can set (x) = k + 0 , 0 being a constant, and we write
x = (r, ) in polar coordinates. Hence, hs(x)i will be continuous and we have
a finite = k/r everywhere but in the singularity. Therefore, one can extract
the following conditions:
I
(x) dl = 2k,
(8.58)
where l is encircling the singularity, and
I
(x) dl = 0,
(8.59)
where l is not encircling the singularity. In the two-dimensional xy-model, one

can think of this as having the angle varying very little between neighboring
sites. Following its behavior around a large closed curve, we find a change of by
a non-zero multiple of 2 (Fig. 8.2). In spin configurations, such singularities are
called vortices, and their winding number k is referred to as the vortex strength or
vorticity. More generally speaking, such vortices belong to the class of topological
defects. They have in common that they do not disappear under whatever continuous transformation of the order parameter. In d = 3, the vortices appear as
lines, in analogy with charged wires. For reasons of simplicity, we shall predominantly suppose d = 2, as we are considering the Berezinski-Kosterlitz-Thouless
transition in the xy-model.
Vortex Energy
The total energy of a vortex Ev in the two-dimensional xy-model with winding
number k can be divided into two parts, the core energy Ec , arising with the
destruction of the order parameter at the vortex core, and the elastic energy Eel ,
hence
Ev = Ec + Eel .
(8.60)
210
Topology in Physics
The free energy, Eq. (8.16), thus the elastic energy when neglecting entropy
contributions, applied to vortex excitations satisfies
I
d = 2k,
(8.61)
s 2 = 0,
(8.62)
where Eq. (8.61) is equivalent to Eq. (8.58). Requiring Fel = 0, corresponding

to ground state configuration of the spin system, produces Eq. (8.62),
Z
Z
s
s
d
2
Fel =
d x[] =
dd x
[]2 ()
2
2
()
Z
= s dd x 2 = 0,
(8.63)
when using partial integration. A solution to these constraints, excluding the
origin, is the field
= k,
v s =
(8.64)
k
e ,
r
(8.65)
with = tan1 (y/x), r = (x2 + y 2 )1/2 , and longitudinal superfluid velocity v s ,

since it is proportional to the gradient of a scalar function. Thus, the elastic
energy, following Eq. (8.16), is
2
Z
Z
k
1
1
2
2
2
s d xv s = s d x
Eel =
2
2
r
Z R
rdr
1
dr 2 = k 2 s ln(R/a),
=
s 2k 2
(8.66)
2
r
a
with core radius a and linear dimension of the system R. The same result can
be attained by a different calculation. Assuming a discontinuity of 2k of the
variable along a cut with surfaces and + , while = 0 on and = 2k
on + , brings forth
Z
Z

1
1
2
2
Eel =
d xs () =
d2 xs (()) (2 )
2
2
Z
Z

1
1
2
2
=
d xs (()) ( ) =
d2 xs (v s ) (8.67)
2
2
Z
Z
1
1
=
s vs d =
hs d
(8.68)
2
2
Z

1
=
+ hs d+ + hs d
2
Z R
1 +
( )
drs |v s | = k 2 s ln(R/a),
(8.69)
=
2
a
211
where we have used Eq. (8.62) in Eq. (8.67), Gauss and hs = s v s in Eq. (8.68).
Furthermore, e and e are the normals to + and , respectively. In the
thermodynamic limit, Eq. (8.66) becomes infinite, thus we do not expect single
vortices to appear. Since it is impossible to distort the spin configuration of
a vortex into a completely aligned state, vortices are considered topologically
stable, although single vortex excitations are located above ground state.
The core energy in two dimensions is depending linearly on the area of the
defect and the condensation energy fcond of the ordered state, which describes
the increase in free energy per unit area originating from the destruction of the
order parameter, hence
Ec = Aa2 fcond ,
(8.70)
A being a numerical constant and a the core radius. We assume the total vortex
energy in Eq. (8.60) to be minimized with respect to the parameter a. Therefore,
k 2 s
= 2aAfcond = 0,
a
k 2 s
a2 =
,
2 Afcond
(8.71)
(8.72)
and we obtain
s k 2 .
(8.73)
2
Consider two vortices which are separated by the distance r and at positions x1 and x2 , respectively. Eq. (8.62) demands (x) = (1) + (2) , with
(i) = tan1 [(y yi )/(x xi )]. Thus,
Z
1
(2) 2
Eel =
d2 xs (v (1)
s + vs )
2
Z
1
(2)
(2)
(1)
= E1 + E2 +
d2 x(h(1)
s v s + hs v s )
2
Z R
Z R
k1
1 +
k2
1 +
s dr + (1 1 )
s dr
= E1 + E2 + (2 2 )
2
r
2
r
r
r
(8.74)
= E1 + E2 + 2s k1 k2 ln(R/r),
Ec =
(i)
where v s = (i) and Ei are the energies of the isolated vortices,

Z
1
2
d2 xs (v (i)
Ei =
s ) ,
2
(8.75)
separated by the distance r. Hence, using ln(R/r) ln(R/a) = ln(a/r),

Eel = s (k12 + k22 ) ln(R/a) + 2s k1 k2 ln(R/r)
= s (k1 + k2 )2 ln(R/a) + 2s k1 k2 ln(a/r).
(8.76)
212
Topology in Physics
Figure 8.3: Multiple vortices in a finite xy-model. Vortices whose strengths have
the same sign are repelling each other, whereas opposite signs lead to an attractive
interaction.
Taking
this to the general case of vortices (Fig. 8.3), one observes that for
P
k = 0 the elastic energy does not diverge for infinite sample size. Such
states will occur at T > 0 as thermal excitations. A pair of vortices with opposite
strengths is topologically equivalent to the uniform state, i.e. they describe ground
state excitations. The tightly bound configuration in which the vortices of such
a pair are nearest neighbors is the state of minimal energy. Furthermore, one
identifies easily the repulsive interaction of vortices of same sign, an attractive
interaction for opposite signs, respectively. In the case of two vortices, we obtain
F 21 = 2 Ev = 2s k1 k2 (ln r)
(x2 x1 )
= 2s k1 k2
,
|x2 x1 |2
(8.77)
which is the force exerted by vortex 1 on vortex 2.

Analogy in Magnetism
When considering loops of constant currents, we encounter a striking analogy
between vortex interactions and magnetic interactions. We find that we can
213
apply the theory of magnetism to vortex theory in d = 3 when replacing the

magnetic intensity H with v s and the magnetic induction B with hs . These
perceptions immediately impose further analogies, as
Z
Z
v s dl = 2k
H dl = I,
(8.78)
s 2 = hs = 0 B = H = 0.
(8.79)
Thus, it follows that the current I carried by the enclosed wire I 2k and the
magnetic permeability s . Now we introduce the field m as the analogy of
the current density J . For encircling vortex lines, we obtain
I
Z
Z
X
v s dl = v s dS = 2
k = m dS.
(8.80)
Therefore, we have
vs = m
(8.81)
and we call m the vortex density. In case of a single vortex, we can write, in
analogy with the current density of a wire,
m(x) = 2kez (2) (x x),
(8.82)
where the vortex line at the core position x = (x , 0) is parallel to ez . In case

of vortex lines with cores at x = (x , 0), Eq. (8.82) becomes
X
m(x) = 2ez
k (2) (x x),
(8.83)
which for vortex loops with core position x (l), as a function of arc length l, is
XZ
dx (l)
m(x) = 2
dl
k (3) (x (l) x).
(8.84)
dl
Furthermore, we have
m = ( v s )
= (v s ) 2 v s = 2 v s ,
since 2 = 0 by Eq. (8.62). This is compatible with
Z
v s = dd xG(x x0 )m(x0 ),
(8.85)
(8.86)
where G(x x0 ) is the Laplacian Green function defined by

2 G(x x0 ) = (d) (x x0 ),
(8.87)
214
Topology in Physics
which can also be written in the following form:

Z
0
dd q eiq(xx )
0
G(x x ) =
.
(2)d
q2
(8.88)
Moreover,
m = ( v s ) = ijk i j vs,k = 0.
(8.89)
Therefore, Eq. (8.16) becomes

Z
Z
Z
1
1
0
d
2
d
d 0
0
Eel =
s d xv s = s d x d x G(x x )m(x) v s
2
2
Z
Z
1
0
d
s d x( v s ) dd x0 G(x x0 )m(x)
=
2
Z
Z
1
d
=
s d x dd x0 m(x)G(x x0 )m(x0 )
2
Z
Z
Z
dd q 1
1
0
d
iq x
s
d x m(x)e
dd x0 m(x0 )eiqx
=
d
2
2
(2) q
Z
1
dd q m(q)m(q)
=
s
.
(8.90)
2
(2)d
q2
Berezinski-Kosterlitz-Thouless Transition
Free Energy Argument

Suppose a two-dimensional lattice of linear dimension R with lattice parameter
a. The vortex core can now be at (R/a)2 different positions. Thus, in statistical
mechanics, i.e. Eq. (8.3), the entropy of the system S = ln (R/a)2 = 2 ln (R/a)
when setting kB = 1. The energy of a single vortex of unit strength (k = 1) is,
using Eq. (8.66), E = s ln(R/a). Therefore, a xy-system containing a single
vortex of unit strength accomplishes a change in free energy, i.e.
F = E T S = (s 2T ) ln(R/a),
(8.91)
where we have used Eq. (8.6). One can clearly identify the critical temperature
Tc = s /2. For temperatures lower than the critical temperature, T < Tc , the
free energy finds its minimum only in case of absence of any isolated vortices. The
contrary is true when T > Tc , and single vortices of unit strength will proliferate.
Thus, Tc indicates the transition temperature from the algebraically ordered to
the disordered phase.
Reduction in Spin-Wave Stiffness
In Sec. 2, we have seen that at the vortex core the magnitude of the order parameter tends to zero. But not only does this amplitude fluctuation reduce the
215
average magnitude of the order parameter, it diminishes the spin-wave stiffness

as well. To see that, we first impose boundary conditions to (x), i.e. vanishing
(x) at the edges of the sample. Hence, we want (x) to be spatially uniform.
We expect the gradient of (x) averaged over the volume to be zero, since
Z
Z
1
1
d
d xh(x)i =
dSh(x)i = 0,
(8.92)
S being the surface and the volume of the sample. Now, let (x) = 0 (x)+vx,
with 0 = 0 at the boundaries. Then we obtain a spatially uniform
Z
1
=
dd xh0 (x) + vi = v.
(8.93)
Furthermore, using Eq. (8.19), we write the renormalized spin-wave stiffness as

R
s =
2
v2
[F (v) F (0)].
(8.94)
We decompose into an analytic part a and a singular part sing caused by

vortices. The Hamiltonian of the xy-model at low temperatures is a function
||
of v s = (x), which can be split into a longitudinal part v s = a and a
||
transverse part v
s = sing . We have v s = 0 and v s = 0. The
requirement of a spatially uniform gradient of the macroscopic phase leads to
||
vs = vs + v + v
s , with a = 0 on the boundaries. Thus, by Eq. (8.4),

H(v)
F (v) = T ln tr exp
,
(8.95)
T
R
||
2
0
0
where H(v) = (s /2) dd x(v s + v + v
s ) + H and H is independent of v s .
When only taking into account the first part of H(v), we find by Eq. (8.95) and
Taylor expansion,
s v 2
F (v) =
2

Z
H(v = 0)
s
d
T ln tr exp
exp
d x[v v s (x)]
T
T
Z
s v 2
2s
=
dd xdd x0 hvsi (x)vsj (x0 ivi vj + F (0) + O(v 4 ). (8.96)

2
2T
From Eqs. (8.94) and (8.96), one derives the following expression:
Z
2s
R
s = s
dd xhv
s (x) v s (0)i.
(d 1)T
(8.97)
216
Topology in Physics
Thus, we have a reduction in the macroscopic spin-wave stiffness R

s , caused by
vortices. Rewriting this in terms of the vortex source function m, we have

2
1
R
(8.98)
s = s
lim 2s hm(q) m(q)i,
d 1 q0 q T
which also corresponds to

Z
1
1
R
d
||
||
s =
d x hg (x) g (0)i
hg (x) g (0)i ,
T
d1
||
||
using hv s (q) v s (q)i =
T
s
(8.99)
and g(x) s v s (x).
Renormalization Equations
We have contemplated, in analogy to the magnetic case, how vortex lines induce
a vortex density, i.e. Eq. (8.83). The source function is normal to the xy-plane,
m(x) = 2ez nv (x), where nv (x) defines a scalar vortex density. In two dimensions, the correlation function will yield the expression given in Eq. (8.48).
Similarly, we can now calculate the Laplacian Green function, Eq. (8.88), in two
dimensions, hence
Z
Z
dq
d2 q eiqx
1
G(x) =
=
J0 (q|x|).
(8.100)
(2)2 q 2
2
q
When implying I(|x|) from Eq. (8.47), we can simplify the calculation and write
1
1
ln(R/a)
I(|x|).
2
2
Thus, using = 2/a, we obtain
G(x) =
G(x) =
1
1
ln(R/a)
ln(|x|/a) + const.
2
2
(8.101)
(8.102)
By utilization of Eqs. (8.83) and (8.90), the contribution to Eel from the first
term can be written as
!2
Z
2
X
s
2
(8.103)
2 ln(R/a)
d xnv (x) ln (R/a)
k .
2
For an infinite sample size, the total vorticity must equal zero. Else, Eq. (8.103)
would lead to an infinite energy contribution. The Hamiltonian of the system is
H = HSW + HV ,
(8.104)
217
where HSW is the spin-wave part and HV is the contribution from vortices, thus
Z
HSW
1
=
K d2 x(a )2 ,
(8.105)
T
2

Z
HV
|x x0 |
2
2 0
0
= K
d xd x nv (x)nv (x ) ln
T
a
|xx0 |>a
X
Ec
(8.106)
+
k2 ,
T
where K = s /T is the reduced spin-wave stiffness and HSW is emerging from
the longitudinal part of v s . In Eq. (8.106), we integrate only over positions which
are separated by a distance greater than the short distance cutoff a, such that no
two vortices can occupy the same point in space. We can impose the vortices to
lie on a lattice and consider a as lattice parameter, thus Eq. (8.106) becomes

X
HV
Ec X 2
|Rl Rl0 |
k ,
(8.107)
= K
+
kl kl0 ln
T
a
T l l
0
l,l
where l indicates a vortex position and Rl,l0 are vectors in the lattice.
Often, it is useful to consider the analog case of a two-dimensional Coulomb
gas with Hamiltonian HC . Thus, HV is identical to HC up to core contributions.
Therefore, the condition of having the total vorticity vanish is analog to the
constraint of charge neutrality.
Furthermore, we rewrite Eq. (8.98) as
hnv (q)nv (q)i
q0
q2
KR = K (2)2 K 2 lim
(8.108)
lim KR (q),
(8.109)
q0
where KR = R
s /T is the renormalized spin rigidity. When considering low temperatures, thus Ec >> T , we can apply Taylor series expansion to the fugacity
y = eEc /T
(8.110)
and, since constraining charge neutrality, which imposes lim nv (q) = 0, we obtain
q0
for the vortex density correlation

hnv (q)nv (q)i = q 2 C2 + O(q 4 ),
where
C2
Z
1
d2 xd2 x0 hnv (x)nv (x0 )i(x x0 )2
= lim
q0 4
1 X
=
(Rl Rl0 )2 hkl kl0 i.
4 0
l,l
(8.111)
(8.112)
218
Topology in Physics
To lowest order in y, we have

hkl kl0 i = 2y
|Rl Rl0 |
a
2K
.
Considering only lowest order in K 1 , we obtain

Z
dr r 32K
1
1
3 2
KR = K + 4 y
.
a a
a
(8.113)
(8.114)
The integral converges for 3 2K < 1, hence K > 2/, corresponding to low
temperatures, while for K 2/ it diverges. Thus, perturbation theory does not
hold at high temperatures. One relies upon renormalization group arguments to
master this problem. The integral is divided into two components, where the
first, converging component is incorporated into K 1 ,
Z
Z ael Z
+
,
(8.115)
ael
3 2
Z
0 1
(K )
=K
ael
+ 4 y
Therefore,
KR1
dr r 32K
.
a a
dr r 32K
a
ael a
Z
dr r 32K
0 1
3 2 l l(32K)
= (K ) + 4 y e e
a a
a
Z

32K
dr r
= (K 0 )1 + 4 3 (y 0 )2
,
a a
a
0 1
= (K )
3 2
(8.116)
+ 4 y
(8.117)
where Eq. (8.117) was attained through rescaling the cutoff parameter,
ael a,
(8.118)
y 0 = e(2K)l y.
(8.119)
and defining the rescaled fugacity
Taking into consideration y 4 corrections, Eq. (8.117) can be rewritten as

Z
dr r 32K 0
1
0 1
3 0 2
KR = (K ) + 4 (y )
,
(8.120)
a a
a
leading to a replacement of K by the shifted parameter K 0 . When rewriting
Eq. (8.116) as
Z ael
(K 0 )1 K 1
dr r 32K
3 2 1
= 4 y
,
(8.121)
l
l a
a a
219
evaluating the integral,

Z
ael
a
dr r 32K
1 e(42K)l
=
,
a a
2K 4
(8.122)
and letting l 0, we obtain

(K 0 )1 K 1
1 e(42K)l
= 4 3 y 2 lim
l0
l0 (2K 4)l
l
3 2
= 4 y lim e(42K)l
lim
l0
dK 1
= 4 3 y 2 ,
dl
(8.123)
(8.124)
where in Eq. (8.123) we have used de lHopitals rule. The same procedure applied
to Eq. (8.119) yields
y0 y
e(2K)l 1
= lim
y
l0
l0
l
l
= lim (2 K)e(2K)l y
lim
l0
dy
= (2 K)y.
dl
(8.125)
Thus, we have derived the differential renormalization equations, up to higher

order correction terms,
dK 1
= 4 3 y 2 (l) + O[y 4 (l)],
dl
dy(l)
= [2 K(l)]y(l) + O[y 3 (l)].
dl
(8.126)
(8.127)
These two equations are often referred to as the Kosterlitz-Thouless recursion

relations.
Vortex Unbinding
For K(l) > 2/, one observes, by examination of the Kosterlitz-Thouless recursion relations, a decrease in the fugacity y(l) with increasing arc length l. In
case of K(l) < 2/, the fugacity increases with growing l. This behavior is in
perfect accordance to the occurrence of unbound vortices above the critical temperature Tc and their absence below Tc , as preliminarily deduced from an energy
contemplation.
Another way of explaining this observation is derived from real space renormalization group discussed in Sec. 1. By an increase of the lattice parameter
a, which also describes the minimum distance between vortices, closely bound
220
Topology in Physics
vortex-antivortex pairs separated by a distance less than a will disappear. The

vortex density thus decreases. K = K(l = 0) is a measure of the energy of angle variation between nearest neighboring sites at a distance a from each other.
Therefore, the stiffness also decreases and K 1 (l) will increase with increasing l.
In contrary, when vortices are unbound, with rescaling of a, vortices persist.
In the analogy of the Coulomb gas, when replacing vortices with free charges,
one can identify this transition with the unbinding of molecules at a condensed
phase, while attaining a conducting plasma. The constraint of vanishing vorticity
holds, hence charge neutrality of the plasma is required.
Integration of the Kosterlitz-Thouless Recursion Relations
Set K (1 x(l)) = K, where K = K and y = y describe the fixed point of the
recursion relations, Eqs. (8.126) and (8.127), which is K = 2/, y = 0. For
small deviations from the fixed point, one can consider only lowest order in x.
Thus,
d 1
d
K
=
(1 x)1
dl
dl 2
dx
=
(1 x)2
2
dl
= 4 3 y 2 ,

d
2
y = 2 (1 x) y.
dl
(8.128)
(8.129)
(8.130)
Hence, to lowest order in x,

dx
= 8 2 y 2 (1 x)2 = 8 2 y 2 ,
dl
dy
= 2xy,
dl
(8.131)
(8.132)
and
dx2
dx
= 2x
= 16 2 xy 2 ,
dl
dl
dy
dy 2
= 2y
= 4xy 2 ,
dl
dl
dx2
dx2 dl
=
= 4 2 .
dy 2
dl dy 2
(8.133)
(8.134)
(8.135)
Solving the differential Eq. (8.135), one receives the hyperbolae

y2 =
1
(x2 + C),
4 2
(8.136)

y
221
C>0
HH
C=0
H
HH
C<0
HH
HH
H

HH
x
x

y = 2
HH y = 2

HH

HH

HH
q
q
|C|
|C|
Figure 8.4: Hyperbolic solutions to Eq. (8.135). In case of C = 0, one obtains the
asymptotes of the hyperbolae. The critical line y = x/(2) ends in the critical
fixed point x = 0, y = 0.
with C being a constant (Fig. 8.4). It is a measure for the distance from the
critical point and can be written as
C = b2 (T Tc ).
(8.137)
Converting Eq. (8.131) in terms of x, one attains

dx
= 2(x2 + C).
dl
(8.138)
We see that, for x 6= 0, x(l) increases with increasing l, i.e. x flows to the right.
p In
case of low temperature, T < Tc , thus C < 0, when replacing u(l) = x(l)/ |C|,
we obtain
p
du
=
2
|C|dl.
(8.139)
u2 1
Integration of the left hand side achieves

Z u(l)
Z
1 u(l)
1
du
1
=
+
du
2
2 u(0)
u+1 u1
u(0) u 1

1
u(l) + 1
u(l) 1
1
= ln
+ ln
2
u(0) + 1
2
u(0) 1

1 u(l) 1 + u(0)
1
ln
(8.140)
=
.
2
1 + u(l) 1 u(0)
The right hand side integrated is
Z l p
p
2 |C|dl = 2 |C|l.
0
(8.141)
222
Topology in Physics
Thus, one can write u(l) as follows:
u(l) =
1 D0 e
|C|l
,
1 + D0 e4 |C|l
(8.142)
where D0 = [1 + u(0)]/[1 u(0)], while
4 |C|l
p
p
1 D0 e
lim x(l) = lim |C|

= |C|.
l
l
1 + D0 e4 |C|l
(8.143)
Furthermore, for R
s (Tc )/Tc = 2/,
KR = lim K(l) = lim K (1 x(l)) = K (1 +
l
l
p
p
2
= K (1 + b T Tc ) = [1 + b T Tc ]
R
p
(Tc )
= s
[1 + b T Tc ],
Tc
p
R
R
T Tc ].
s (T ) = s (Tc )[1 + b
|C|)
(8.144)
(8.145)
Considering high temperatures, T > Tc , hence C > 0, the solution to Eq. (8.138)
is

Z x(l)
dx
1
1 x(l)
1 x(0)
=
tan tan
= 2l.
(8.146)
2
C
C
C
x(0) x + C
For T >pTc , in the vicinity of the p
critical point, x(0) is negative,
and when
1
T Tc , |C| 0, while |x(0)| >>
|C|. Hence, tan (x(0)/ C) /2.
Let x(l ) be positive, thus tan (x(l )/ C) /2, and Eq. (8.146) becomes
2l = ,
C
1
b0
l =
=
,
2 b T Tc
T Tc
(8.147)
(8.148)
where bb0 = /2 is universal, although the coefficients b and b0 are not. Thus, the
correlation length in the neighborhood of the critical point, with the condition
T > Tc , is
0
= el = eb / T Tc .
(8.149)
a
Rewriting Eqs. (8.55) and (8.56) in terms of the renormalized parameters,
one obtains
G(x)
|x|(T ) ,
1
(T ) =
.
2KR (T )
(8.150)
(8.151)
223
Thus, in the limit T Tc , the exponent (T ) becomes

lim (T ) = lim
T Tc
T Tc
1
1
=
2KR (T )
4
(8.152)
when taking into consideration

lim KR (T ) =
T Tc
2
.
(8.153)
At exactly T = Tc , hence C = 0, Eq. (8.138) becomes

dx
= 2x2 ,
dl
which integrated delivers
x(l) =
x(0)
.
1 2lx(0)
(8.154)
(8.155)
For l large, we have x(l) 2l1 , and thus KR1 = 2 [1 + x(l)] to leading order in y.
Applying the Josephson scaling relation to the reduced spin-wave stiffness, leads
to
KR (q) = e(d2)l KR (el q).
(8.156)
Therefore, when setting el q = 1 or x(l ) = (ln q 1 )/2,

Z
d2 q 1 eiqx
g(x) =
(2)2 KR (q)q 2

Z
dq
1
1
1
4 |x|1 q
2 ln q 1
1
1
ln |x| ln(ln |x|),

4
8
and
G(x) = eg(x)
(ln |x|)1/8
.
|x|1/4
(8.157)
(8.158)
At high temperatures, T > Tc , we observe that

G(x)
with 1 = ln(2/K).
K
2
|x|
e|x|/ ,
(8.159)
224
Topology in Physics
6
y(l)
^
R
*
1
1
r
-
/2
(l)
Figure 8.5: Renormalization flows of the recursion relations. The thin line represents initial conditions. The line originating at the dot applies to a temperature
T > Tc .
Renormalization Analysis
One can now investigate the renormalization flows (Fig. 8.5) of the recursion
relations, Eqs. (8.126) and (8.127). We find a separatrix, which passes through
the critical point y(l) = 0, K 1 = /2, indicated with ys (l), Ks1 (l). The initial
conditions of a flow satisfy the fugacity term
y = eEc /T = eEc K/s ,
(8.160)
and thus determine the transition temperature, which arises from intersection
of the fugacity term with the separatrix ys , Ks1 . Furthermore, small y, with
K 1 < /2, i.e. points beneath the separatrix, flow toward y(l) = 0, with K 1
exhibiting a finite value. Thus, in the limit
KR = lim K(l),
l
(8.161)
the renormalized rigidity will be finite, which induces the absence of any unbound
vortices. Points above the separatrix will, however, tend to large y, with K 1
large, thus aim at a phase with unbound vortices and vanishing rigidity. For
points that lie on the separatrix itself, having K 1 < /2, the renormalization
flow is toward the critical point.
225
Conclusion
Two-dimensional systems that exhibit O(2) or U (1) symmetry cannot undergo
a phase transition based on spontaneous symmetry breaking. This is substantiated by the Mermin-Wagner theorem. However, a Berezinski-Kosterlitz-Thouless
transition, which we have characterized as vortex unbinding phenomenon, is possible. It carries an algebraically ordered phase with power-law correlations to a
disordered phase with exponential correlations.
More precisely, we have seen that for T < Tc spin-wave theory applies and
the system is of quasi-long-range order, namely
G(x) |x|(T ) ,
(8.162)
where (T ) = 1/[2KR (T )]. Located at the left-hand-side of the separatrix, the

system aspires a state of vanishing fugacity, hence suppressed vortices, and finite
rigidity. At T = Tc , vortices unbind and the correlation function becomes
G(x)
[ln(|x|/a)]1/8
.
|x|1/4
(8.163)
The renormalization flow is toward the critical point. In the case of T > Tc ,
vortices are unbound and the system exhibits exponential correlations, i.e.
G(x) e|x| ,
1
(8.164)
where the inverse correlation length is 1 = ln(2/K). Its flows head toward a
state of vanishing rigidity.
Example in Condensed Matter Physics
Superfluid helium films exhibit proper xy-symmetry, thus one expects a BerezinskiKosterlitz-Thouless transition, taking the superfluid to the normal fluid state.
However, in this case we define
vs
~
,
m
(8.165)
which has units of velocity. We therefore call v s the superfluid velocity. The free
elastic energy is then
2 Z
Z
1
~
1
d
2
d xs vs = s
dd x()2 .
(8.166)
Fel =
2
2
m
Thus, the rigidity s is the mass density. When also replacing
2
~
s
s ,
m
(8.167)
226
Topology in Physics
we can directly apply the results of Sec. 3 to two-dimensional superfluid helium

films. Thus, for m being the mass of a helium core, we obtain
lim K(T ) =
T Tc
lim
T Tc
2
s
(m2 /~)2 R
s
= lim
Tc T Tc
Tc
2m kB
3.4913 109 g cm2 K1
2
~
(8.168)
when considering Eq. (8.153) and reintroducing the Boltzmann constant kB . Measurements of superfluid densities extrapolated to zero frequency yield for this
value about 3.35 109 g cm2 K1 . [51]
Furthermore, one can also observe a large non-universal peak in the specific
heat above Tc that is associated with the entropy liberated by the vortex unbinding and is characteristic for a Berezinski-Kosterlitz-Thouless transition.
9 Berry phase I: Mathematical

concepts
Simon Wood
Supervisor: Sebastian Pilgram
Berrys geometric phase is a phenomenon encountered in periodic quantum-mechanical systems whos period is large. This
report aims at giving a basic understanding of this phase and
the mathematical tools required to determine it.
Introduction
In the quantum mechanical description of a physical system one has a finite or

infinite dimensional Hilbert space H of state vectors and a set of linear operators
acting on these state vectors. In a non isolated environment these operators depend on a set of (external) parameters R = (R1 , R2 , . . . ). Each R characterises
a particular configuration of the environment and uniquely determines the operators. In particular, a changing environment is described by time-dependent
parameters, R = R(t).
For a quantum system in a classical environment, the parameters R label
the points of a smooth manifold M . Every change of the environment is then
described by a curve C : [0, T ] M , with points R(t) C. The Manifold
M is called the parameter space. The Hamiltonian operator is assumed to be a
smooth single valued function of R M . Here smoothness of the Hamiltonians
means that its eigenvalues and eigenvectors are smooth functions of R.
In a quantal environment the parameters are (generalised) eigenvalues of the
operators of the of the quantal environment.
An example of a quantum mechanical system in a classical environment, that
we will be looking at in detail later on is a magnetic moment m in a rotating
(classical) magnetic field B of constant magnitude B := |B|. The parameter-
227
228
Topology in Physics
dependent Hamiltonian is given by

e
= H0 + bR
J
H = H0 m B = H0 Bg
RJ
2mc
(9.1)
The parameter space of this Hamiltonian is

R3 | |R|
= 1}.
S 2 = {R
In general the evolution of the pure states of the quantum system in the
external environment is described by the Schrodinger equation
d(t)
= H(R)(t),
dt
(9.2)
d(t)
= H(R(t))(t)
dt
(9.3)
i
or
i
for a changing environment. The general (mixed) states are described by the
statistical operators (t) = |(t)i h(t)| whose evolution is given by the Lioville
von Neumann equation
i
d(t)
= [H(R(t)), (t)] .
dt
(9.4)
The space of physical states does not only contain the solutions of (9.3) for
one given fixed value of the parameters R or for one given environmental process
t 7 R(t), but for all R M . This means that there is a single space of physical
state vectors H for all values of R. For any given value of R, one may choose an
orthonormal basis of eigenvectors |n, Ri of the parameter-dependent Hamiltonian
H(R) |n, Ri = En (R) |n, Ri
hm, R|n, Ri = mn
and write the Hamiltonian to its spectral resolution
X
H(R) =
En (R) |n, Ri hn, R| .
(9.5)
(9.6)
Given an environmental process along a time parametrisation R(t), one can define
a time dependent Hamiltonian as H(t) := H(R(t)) and the spectral resolution
(9.6) changes to
X
(9.7)
H(t) := H(R(t)) =
En (R(t)) |n, R(t)i hn, R(t)| .
n
Berry phase I: Mathematical concepts
229
In general the projection operators

n (R(t)) := |n, R(t)i hn, R(t)| ,
(9.8)
corresponding to the eigenstates of H(t) change with time, so the eigenstates

themselves will also change.
We assume that the observables are single-valued functions of R over the whole
parameter space of the environment. Single-valuedness of the observable means
that if the same value of R occurs more than once (i.e., at different times) during
a process, then the observables are the same at each occurrence. In particular, if
the environmental process is periodic, i.e., if the environmental parameters R(t)
traverses a closed path C in M and return, after some period T , to their original
values,
C : R(0) R(t) R(T ) = R(0),
(9.9)
then the Hamiltonian, its eigenvalues and the projection operators (9.8), which
are uniquely defined by (9.6), are the same at R(T ) as they are at R(0),
H(R(T )) = H(R(0)),
En (R(T )) = En (R(0)),
|n, R(T )i hn, R(T )| = |n, R(0)i hn, R(0)| .
(9.10)
(9.11)
(9.12)
It is important to note that though the observables are single-valued functions of

R the basis vectors |n, Ri themselves will in general not be single valued over the
whole parameter space, but only on sufficiently small open subsets of M (called
coordinate patches). In other words, in general it will not be possible to define
smooth single-valued |n, Ri for all R M . Thus (9.12) does not imply
|n, R(T )i = |n, R(0)i
for R(T ) = R(0),
(9.13)
but only
|n, R(T )i = ein |n, R(0)i
for R(T ) = R(0),
(9.14)
where ein is a phase factor. This is a consequence of the Schrodinger equation (9.3) only defining its
eigenstates up to a complex phase ein .
normalised

Therefore the eigenstates |n, Ri0 gained from a phase transformation
|n, Ri |n, Ri0 = ein (R) |n, Ri
(9.15)
form just as valid a basis of eigenvectors of the Hamiltonian as {|n, Ri} (where
n (R) are arbitrary real phase angles).
Phase transformations for which the phase factors ein (R(t)) are single-valued
functions are called gauge transformations. In general, if we go from one patch
O1 M of the parameter space to a neighbouring patch O2 M with a different
parametrisation, then eigenvectors of H(R) in the overlap region R O1 O2
will be related by phase transformations of the form (9.15).
230
Topology in Physics
Berrys Geometric Phase and the Adiabatic Approximation
In this section we discuss an approximation for solving the Schrodinger equation

(9.3). For a Hamiltonian H(R(t)) whose parameters change in time, the interaction with the environment can cause the physical system to evolve from the n-th
eigenstate |n, R(0)i hn, R(0)| at t = 0 into any other eigenstate |m, R(t)i hm, R(t)|,
at a later time t. A very particular situation arises if this does not happen, i.e.,
when the state remains an eigenstate of H(R(t)) at all times t with the same
energy quantum number n. The adiabatic approximation assumes that this is
the case. This means that |(t)i h(t)| changes in such a way that at all times t
|(t)i h(t)|
adiabatic
|n, R(t)i hn, R(t)| = n (R(t)).
(9.16)
This time development is called adiabatic time development.

The dynamical equations (9.3),(9.4) and the equation for adiabatic time development (9.16) impose two separate conditions on the state |(t)i h(t)| and need
not be generally compatible with each other. To see when they are compatible
we insert (9.16) into (9.4). This yields
i
d(t)
= [H(R(t)), (t)] = [H(R(t)), n (R(t))]
dt
X
=
Em (R(t)) [m (R(t)), n (R(t))]
m
= En (R(t)) [n (R(t)), n (R(t))] = 0.
(9.17)
Equation (9.17) means that (t) does not change in time,

W (t) = |(t)i h(t)| = |(0)i h(0)| = (0),
t.
(9.18)
Hence any adiabatically evolving state (9.16) (which obeys (9.4)) must be stationary. In particular it cannot have a non-trivial cyclic evolution
(0) (t) (T ) = (0),
(9.19)
in which (t) changes in time. Therefore exact adiabatic cyclic evolutions do not
adiabatic
exist. The adiabatic equality = in (9.16) can only be an approximation.
Since we have shown that adiabatic development can only be an approximation, we must derive a condition of validity for the adiabatic approximation. First
we express the evolving state vector (t) in the basis {|n, R(t)i},
X
(9.20)
(t) =
cm (t) |m, R(t)i .
m
231
Suppose that initially (t) is an eigenvector of the initial Hamiltonian H(0), then
the adiabatic approximation is a valid approximation if and only if we can ignore
all the coefficients cm (t) in (9.20) except cn (t), i.e.,
(t)
adiabatic
cn (t) |n, R(t)i ,
(9.21)
with cn (0) = 1. The solutions of the Schrodinger equation (9.3) evolve unitarily,
thus the coefficient cn (t) must be a phase factor (i.e., have modulus 1). Substituting the expression (9.21) for (t) in the Schrodinger equation (9.3) and using
(9.5), we find

d
d
cn (t) |n, R(t)i + cn (t) |n, R(t)i = iEn (R(t))cn (t) |n, R(t)i
dt
dt

d
d
adiabatic
cn (t) + iEn (R(t))cn (t) |n, R(t)i = cn (t) |n, R(t)i .
dt
dt
(9.22)
The inner product of (9.22) with |m, R(t)i for m 6= n in view of the orthogonality
of the basis vectors yields

d
d
cn (t) hm, R(t)| |n, R(t)i = hm, R(t)|
cn (t) + iEn (R(t))cn (t) |n, R(t)i
dt
dt

d
cn (t) + iEn (R(t))cn (t) mn = 0
=
dt
hm, R(t)|
d
adiabatic
|n, R(t)i = 0,
dt
m 6= n.
(9.23)
This is the necessary and sufficient condition for the validity of the adiabatic
approximation. We can express the left-hand side of this equation in terms of
the matrix elements of the time derivative of the Hamiltonian. In order to show
this, we take the differential of both sides of (9.5)
dH(R) |n, Ri + H(R)d |n, Ri = dEn (R) |n, Ri + En (R)d |n, Ri ,
and compute the inner product of both sides of this equation with |m, Ri for
m 6= n. The left-hand side yields
hm, R| dH(R) |n, Ri + hm, R| H(R)d |n, Ri = hm, R| dH(R) |n, Ri
+ Em (R) hm, R| d |n, Ri
and the right-hand side
hm, R| dEn (R) |n, Ri + hm, R| En (R)d |n, Ri =dEn (R)mn
+ En hm, R| d |n, Ri
=En (R) hm, R| d |n, Ri
232
Topology in Physics
hm, R| dH(R) |n, Ri + Em (R) hm, R| d |n, Ri = En (R) hm, R| d |n, Ri

hm, R| dH(R) |n, Ri
hm, R| d |n, Ri =
(9.24)
En (R) Em (R)
For a given environmental process, R = R(t), we can turn the differentials appearing in (9.24) into total time derivatives by dividing by dt. This leads to
hm, R(t)| dtd H(R) |n, R(t)i
d
hm, R(t)| |n, R(t)i =
.
dt
En (R(t)) Em (R(t))
(9.25)
From the equation (9.25) combined with the condition (9.23) we can re-express
the validity of the approximation in the form
hm, R(t)| dtd H(R(t)) |n, R(t)i
En (R(t)) Em (R(t))
adiabatic
0,
m 6= n.
(9.26)
The adiabatic approximation is a valid approximation if and only if the left-hand

side of (9.26) can be neglected. It is important to note that the left-hand side of
(9.26) has the dimension of frequency. This means that in order to decide whether
it can be neglected, one must have an intrinsic frequency (or energy) scale for
the quantum system. For our example of a magnetic moment interacting with a
rotating magnetic field (9.1) the intrinsic frequency scale is given by b.
Calculating Berrys Geometric Phase
In the preceding section we showed that for a periodic Hamiltonian H(t), an
adiabatically evolving state (t) = |(t)i h(t)| with the initial condition (0) =
n (R(0)) traverses a closed path C in the state space P(H). This does not imply
that the normalised state vector (t) which fulfils the Schrodinger equation (9.3),
also traverses a closed path in H. In general the path
C : [0, T ] (t) H,
with h(t)|(t)i = 1,
(9.27)
is not closed in H, but satisfies

C(T ) = (T ) = ei (0).
(9.28)
For the case of a time-independent Hamiltonian, the phase factor is

ei = eiEn T , or ei = ei
RT
0
dt0 En (t0 )
(9.29)
This is called the dynamical phase factor. For a general time-dependent Hamiltonian H(R(t)), there is an additional phase factor which is called the the geometric
phase or Berry phase.
If the adiabatic approximation is valid, we can express the evolving state
vector according to (9.21), where the coefficient cn satisfies (9.22). We can obtain
233
the explicit formula for cn by calculating the inner product of both sides of (9.22)
with |n, R(t)i. This yields

d
d
cn (t) + iEn (R(t))cn (t) nn = cn (t) hn, R(t)| |n, R(t)i ,
dt
dt

d
d
cn (t) = cn (t) iEn (t) + hn, R(t)| |n, R(t)i ,
(9.30)
dt
dt
where En (t) := En (R(t)). This equation can be integrated to obtain

d
dcn
= iEn (t) + hn, R(t)| |n, R(t)i dt,
cn
dt
Z t
Z t
Z cn (t)
d
dcn
= i
En (t0 )dt0
hn, R(t0 )| 0 |n, R(t0 )i dt0 .
dt
cn (0) cn
0
0
(9.31)
In view of cn (0) being equal to 1, this leads to

Z t
cn (t)
En (t0 )dt0 + in (t),
ln cn |cn (0) = i
i
cn (t) = e
0
Rt
0
En (t0 )dt0 in (t)
(9.32)
where
ein (t) := ei
Rt
0
ihn,R(t0 )| dtd0 |n,R(t0 )idt0
(9.33)
Because the coefficient cn (t) in (9.22) has modulus 1 n (t) is a real phase angle.23
It is important to note that n (t) is only defined up to an integer multiple of 2.
A remarkable property of the phase angle n (t) is that it does not depend
on the time dependence of the integrand in (9.33) but only on the path traced
by R(t) in the parameter space. In fact, it can be directly defined in terms of a
curve integral
Z t
d
n (t) =
i hn, R(t0 )| 0 |n, R(t0 )i dt0
dt
0
Z R(t)
|n, Ri dRi
=
i hn, R|
i
R
R(0)
Z R(t)
=
Ani (R)dRi
(9.34)
R(0)
over a vector-valued function

An := i hn, R| |n, Ri
23
A more analytic proof will be supplied later in this section.
(9.35)
234
Topology in Physics
This vector-valued function is called the Mead-Berry vector potential. As can

be seen in (9.35) it depends on single-valued basis eigenvectors |n, Ri of H(R).
Since a smooth single-valued basis vector may in general not be found on the
whole parameter space M , but only on its patches, the same is true for An . This
problem will be discussed in detail for the example (9.1) in section 3. Here we
will assume that the curve C lies in a single patch over which a complete set of
smooth and single-valued basis vectors |n, Ri exists.
The Mead-Berry vector potential (9.35) may also be expressed as a (local)
differential one-form,
An = Ani dRi := hn, R|
|n, Ri dRi = i hn, R| d |n, Ri ,

Ri
(9.36)
which is defined on the same patch as |n, Ri. Here the differentials dRi are the
basis differential one-forms (covariant vectors), and d is the exterior derivative
operator. The one-form An is called the Mead-Berry connection one-form. It can
be used to yield the following expression for the phase angle n (T ) of (9.34),
Z
R(T )
i hn, R |d| n, Ri =
n (T ) =
R(0)
An ,
(9.37)
where C is the curve traced by the parameters R in the parameter space M .

We shall now prove that n (t) is real. To do this, it suffices to show that
hn, R| d |n, Ri is purely imaginary. The proof requires the following identity
d hn, R|m, Ri = d nm
(d hn, R|) |m, Ri + hn, R| d |m, Ri = 0
(d hn, R|) |m, Ri = hn, R| d |m, Ri
(9.38)
Because the scalar product is a hermitian two-form and the identity (9.38) we
receive
hn, R| d |n, Ri = (d hn, R|) |n, Ri
= hn, R| d |n, Ri.
(9.39)
So hn, R| d |n, Ri is equal to the negative of its complex conjugate which can only
be the case if it is purely imaginary.
Returning to equation (9.21) we now see that the state can be expressed as
(t)
adiabatic
ei
Rt
0
En (t0 )dt0 in (t)
|n, R(t)i .
(9.40)
As can be seen in this equation in addition to the dynamical phase factor, there
exists another phase factor which is given in terms of the eigenvectors |n, Ri of
H(R) according to (9.37).
235
As we know from (9.15) the basis eigenvectors |n, R(t)i are only determined up
to a phase factor. Thus, the additional phase factor in (9.40) can be transformed
away by a phase transformation for non cyclic changes. In the following we will
show that in general this is not possible for cyclic changes.
From the definition of the Mead-Berry vector potential (9.35), An (R) transforms according to

n
An (R) A0 (R) = hn, R|0 |n, Ri0

= i hn, R| ein (R) ein (R) |n, Ri

= i hn, R| |n, Ri + iein (R) ein (R) .
= An (R) n (R).
(9.41)
Alternatively we have the connection one-form
n
An (R) A0 (R) = An (R) dn (R).
(9.42)
In view of (9.41), under a gauge transformation the phase angle n (t) transforms
according to
Z R(t)
n
0
n (t) n (t) =
A0 (R)dR
R(0)
= n (t) n (R(t)) + n (R(0)).
(9.43)
If we do the calculations that led to (9.40) using |n, Ri0 in place of |n, Ri, we
obtain (9.40) with the primed quantities on the right-hand side. Using (9.21) we
obtain for the primed quantities
0
ein (t) |n, R(t)i0 = ein (t) ein (R(t)) |n, R(t)i .
(9.44)
If n (R(t)) is an arbitrary single-valued function modulo 2, we can choose it

0
such that n (R(t)) = n (t), thus the phase factor ein (t) ein (R(t)) becomes unity
and we find in place of (9.40)
(t) = ei
Rt
0
En (t0 )dt0
|n, R(t)i .
(9.45)
This also fulfils the initial condition

(0) = |n, R(0)i hn, R(0)| .
(9.46)
Since |n, Ri0 is just as a basis eigenvector as |n, Ri, we can use it to describe
the time development of the state vector, i.e., use (9.45) which only involves the
dynamical phase factor.
The above arguments made use of the fact that n (R(t)) was arbitrary. If
after some period T the environmental parameters return to their original values
236
Topology in Physics
as described by the closed path C of (9.9), then one cannot choose n (R(T ))
freely to remove n (T ).
For R(T ) = R(0), the single valuedness of ein (R) implies
ein (R(T )) = ein (R(0)) , or n (R(T )) = n (R(0)) + 2k
k Z.
(9.47)
Therefore according to (9.43)

I
n (T )
n0 (T )
=
=
IC
C
A0 (R)dR
A(R)dR 2k
= n (t) 2k
kZ
k Z.
(9.48)
Thus n (T ) - which is only defined modulo 2 - is invariant under the gauge

transformation (9.15) and cannot be removed. We therefore have
(T ) = ei
RT
0
En (t0 )dt0 +in (T )
|n, R(T )i
with n (T ) given by the loop integral over the closed path C of (9.9)
I
n (C) := n (T ) =
An dR
modulo 2
IC
=
An
modulo 2.
C
(9.49)
(9.50)
Inserting the initial conditions we obtain

(T ) = ei
RT
0
En (t)dt in (C )
(0).
(9.51)
The phase angle n (C) is called the Berry phase angle, and ein (C ) is called the
Berry phase factor.
We have shown that for a closed path the extra phase factor cannot be transformed away. This does not mean that n (C) cannot be zero. Should this be the
case, the vector potential An (R) (and the connection one-form An (R)) need not
be zero but will be trivial, which means
An (R) = (R)
An (R) = d(R).
(9.52)
(9.53)
where (R) is a well-defined function of R.24 The cases that we are interested
in are those for which the Berry phase angle is different from zero (or an integer
24
Note that we still require the curve C to lie on a single patch of the parameter space. If
this is not the case, then (9.52) may be satisfied on individual patches but since the curve C
does not in general lie in one patch, the corresponding Berry phase may still be non-trivial.
Such a phase is called a topological phase and subject of the Berry phase II: The Aharonov
Bohm Effect presentation.
237
multiple of 2). The Hamiltonian (9.1) provides an example of such a case. A

detailed discussion of this system will be presented in the next section.
From (9.41) we see that the Mead-Berry vector potential satisfies the same
gauge (phase) transformation rule, (9.41), as the vector potential of electromagnetism. The set of phase factors ein (R) form the group U(1) of unitary 1 1
matrices. We therefore have a gauge theory with gauge (symmetry) group U(1)
and gauge potential An (R). This is the reason for which we call the phase
transformation (9.15) a gauge transformation. Whereas An (R) is not an invariant quantity with respect to a gauge transformation (it transforms according to
(9.41)), the Berry phase is gauge invariant. If the parameter space is three dimensional and the parameter R is a three dimensional vector R = (R1 , R2 , R3 ), we
have a complete analogy with electrodynamics. However, the physical meaning of
the quantities associated with the Berry phase is different. The gauge potential
(9.35) is defined in terms of the eigenvectors of the Hamiltonian and has nothing
to do with electromagnetism. For an m-dimensional parameter space we again
have a U(1)-gauge theory, but the gauge transformations and gauge potentials
now depend on m parameters R = (R1 , . . . , Rm ) and An (R) consists of m components Ani (R), i = 1, . . . , m. In this case we have a U(1) gauge theory over an
m-dimensional parameter space.
In analogy to electrodynamics we can define a gauge field strength tensor F n
with the components
Fijn :=
n
n
Aj
A ,
i
R
Rj i
i, j = 1, 2, . . . , m.
(9.54)
This is an antisymmetric covariant tensor field of rank two, i.e., it is a differential

two-form
Anj
1 n i
j
F = Fij dR dR =
dRi dRj = dAn .
2
Ri
n
(9.55)
Here the antisymmetry of the wedge product was used, and dAn stands for the
exterior derivative of the Mead-Berry connection one-form An given by (9.36).
The two-form F n is also called the Mead-Berry curvature two-form. It is given
by
F n = d (i hn, R| d |n, Ri)
= i (d hn, R|) d |n, Ri + i hn, R| d2 |n, Ri
= i (d hn, R|) d |n, Ri
(9.56)
where we have used the identity d2 = 0 of the exterior derivative.

An important property of the curvature two-form is that unlike the connection
one-form, it is a gauge-invariant quantity. This follows directly from (9.42),
(9.55), and the identity d2 = 0,
n
F n F 0 = dA0 = dAn d2 n = dAn = dF n .
(9.57)
238
Topology in Physics
The gauge invariance of F n has two important consequences: F n is a globally

defined object over M , and it may be used to yield a direct formula for the Berry
phase.
In order to obtain this formula, we assume that the curve C bounds a surface
S M , and use Stokes theorem to convert the loop integral in (9.50) to a surface
integral over S. The result is
I
Z
Z
n
n
n (C) =
A =
dA =
F n modulo 2.
(9.58)
C
The surface S can arbitrarily chosen as long as it is bounded by the closed curve
C.
Next we wish to use the manifestly gauge-invariant expression of the Berry
phase angle (9.58), to investigate the consequences of a possible degeneracy of
the energy eigenvalues. In order to see what happens when two energy levels
become degenerate for some values of the parameters, we express the curvature
two-form F n in terms of the eigenvalues En (R). Using the completeness of the
basis {|m, Ri},
X
1=
|m, Ri hm, R| ,
(9.59)
m
we can express (9.56) in the form

X
[(d hn, R|) |m, Ri] [hm, R| d |n, Ri]
Fn = i
m
= i
[hn, R| d |m, Ri] [hm, R| d |n, Ri]
=i
[hm, R| d |n, Ri] [hn, R| d |m, Ri] .
(9.60)
m6=n
In the second equality we used the identity (9.38) and the third equality follows
from the antisymmetry of the wedge product of two one-forms. In particular,
note that due to this antisymmetry the m = n term in the sum is identically
zero.
In order to show the dependence of F n on the difference of the energy eigenvalues, we substitute (9.25) in the right-hand side of (9.60). This yields
Fn = i
X hn, R| [dH(R)] |m, Ri hm, R| [dH(R)] |n, Ri

[En (R) Em (R)]2
m6=n
(9.61)
The form (9.61) does not depend on the phase factor of the basis vectors
|m, Ri. Hence unlike the connection one-form (9.36), which is expressed in terms
of the smooth single-valued |n, Ri and can therefore be defined only on a patch
239
in the parameter space M where the latter exists, the curvature two-form (9.61)
is globally defined on M . Therefore, (9.61) can be used to compute the geometric
phase angle even for the cases where the curve C lies in a region in the parameter
space where smooth single-valued |n, Ri do not exist. The formula (9.61) also
shows that the singularities of F n occur at those values of R = R0 where the
eigenvalues are degenerate En (R0 ) = Em (R0 ).
From the above formulas we see that the Berry phase angle n (C) is independent of how the closed loop C is traversed (provided that the condition (9.26) for
the validity of the adiabatic approximation is satisfied). This means that it is not
sensitive to the details of the dynamics of the quantum system, it only depends
upon the path C and is thus a geometric quantity.
Berrys Geometric Phase applied to a Precessing Magnetic Field
In this section we will take a closer look at the quantum system described by the
equation (9.1), that of a quantum particle with magnetic moment m = B gJ
in an external magnetic field B(t) = B R(t)

whose direction R(t)
is changing
periodically. In particular we will consider the case in which the direction of the
magnetic field precesses around a fixed axis which we take as the 3-axis (z-axis)
of our coordinate frame in space (R3 ). If the direction rotates slowly (adiabatically) this system provides an application of the general ideas developed in
Section 2. The Schrodinger equation for a magnetic moment in a precessing magnetic field has been solved exactly. Therefore we need not restrict ourselves to
the adiabatic approximation and will also compute the non-adiabatic geometric
phase. The latter is also known as the Aharonov-Anandan phase. With the help
of this example we will then discover the non-adiabatic geometric phase for a
general cyclic evolution.
In our example we are not interested in the dynamics of the particle in position space, but only in the dynamics of the spin in the magnetic field. So our
Hamiltonian (9.1) reduces to
H(R(t))
=
Bge
R(t) J = bR(t)
J
2mc
(9.62)
where J is the angular momentum operator of the quantum system, R(t)

is the
Bge
parameter that describes the changing environment, and b = 2mc is a constant.
The Parameterisation of the Basis Vectors
For the system described by the Hamiltonian (9.62), the parameter space of the
in R3 . Therefore the parameter space
environment is the set of all unit vectors R
of this system may be identified with the unit sphere S 2 embedded in R3 . We
240
Topology in Physics
parametrise the points of S 2 by the polar and azimuthal angles (, ) according

to
sin cos
= R(,
) = sin sin ,
(9.63)
R
cos
where
0 , and 0 < 2.
This parametrisation associates unique values of the pair (, ) to each unit vector
except for the unit vector
R

0
e3 := 0
(9.64)
1
of the north pole N and the unit vector e3 of the south pole S. e3 and e3
are given respectively, by = 0 and = for all values of ; the value of is
= e3 .
not determined when R
The special case in which the magnetic field precesses uniformly about the
3-axis is described by
t),
B(t) = B(sin cos t, sin sin t, cos ) = B R(,
(9.65)
with B, and = /t being constants. The precession of such a magnetic field

is shown in Fig. 9.1.
The eigenvectors (9.5) for this example are defined by
J |k, Ri = bk |k, Ri .
H(R) |k, Ri = bR
(9.66)
J . The place of the energy quantum

They are eigenvectors of the operator R(t)
number n in (9.5) is taken here by k which is the quantum number for the
component of angular momentum along the (changing) direction of the external
magnetic field.
For this example, the eigenvalues Ek (R(t)) = bk of the time-dependent Hamiltonian H(R(t)) are constant, but the eigenvectors |k, R(t)i and the eigenprojectors k (R(t)) = |k, Ri hk, R| change in time. The eigenvalue of the observable is
constant but its physical interpretation changes in time; it is the eigenvalue of the
J where the direction R(t)
observable R
changes with respect to the laboratory
frame.
or by the polar
The vectors of (9.66) are parameterised by the unit vector R
and azimuthal angles (, ). They can be obtained by applying (, )-dependent
rotations to an eigenvector |k, e3 i of the component of angular momentum in
direction of the north pole, J3 = e3 J .
241
e3
B(0)
B(t)
e2
t
e1
Figure 9.1: A quantal magnetic moment in an external magnetic field precesses
uniformly around a cone of semi-angle with e3 . (, ) are the polar and azimuthal angles for the rotation of the external magnetic field.
242
Topology in Physics
There are many (, )-dependent rotations R(, ) SO(3) which when ap ). We choose the following product of
plied to e3 give the unit vector R(,
rotations.
R(, ) = R3 ()R2 ()R3 ()e3 = R3 ()R2 ()e3
0) = R(,
),
= R3 ()R(,
(9.67)
where R3 () does nothing to the unit vector e3 and where R2 () produces the
0) which lies in the 1-3 plane at an angle with respect to the
unit vector R(,
e3 -axis. The rotation R3 () has been included in the definition of R(, ) in
order for the rotation R(0, ) to be independent of .
Rotations, like any other continuous transformations, are represented in the
space of quantum physical states by unitary operators (representing the group
SU(2)). The unitary operators that represent the rotations R3 () and R2 () are
given by
U3 () = eiJ3 ,
U2 () = eiJ2 .
(9.68)
The product of two or more rotations is represented by the product of the corresponding operators. The rotation R(, ) of (9.67) is thus represented by the
operator
U (, ) = U3 ()U2 ()U3 () = eiJ3 eiJ2 eiJ3 .
(9.69)
We now choose a fixed normalised eigenvector |k, e3 i of J3 = e3 J = R(0,

0)J
and transform it using the unitary operator U (, ). The resulting state vector,
3 i = eiJ3 eiJ2 eiJ3 |k, e
3i ,
|k, , i := U (, ) |k, e
(9.70)
) J with eigenvalue k,
is an eigenvector of the operator R(,
) J |k, i = k |k, , i .
R(,
(9.71)
In order to prove (9.71), we use the following transformation property of the

angular momentum operators Ji which follows from their commutation relations25
eiJ2 J3 eiJ2 = J3 cos + J1 sin ,
(9.72)
eiJ3 J1 eiJ3 = J1 cos + J2 sin ,
(9.73)
iJ3
e
25
iJ3
J2 e
= J2 cos J1 sin .
(9.74)
A
A
These
=
P properties can be verified using the Backer-Campbell-Hausdorff formula, e Be
B + n=1 Bn /n!, B0 := B, Bn := [A, Bn1 ] and the commutation relations of the angular
momentum operator.
243
Together with the fact that J3 commutes with eiJ3 , we receive
R(,
) J |k, , i = (sin (J1 cos + J2 sin ) + J3 cos ) |k, , i

= eiJ3 J1 eiJ3 sin + J3 cos |k, , i

= eiJ3 J1 eiJ3 sin + eiJ3 J3 eiJ3 cos |k, , i

= eiJ3 (J1 sin + J3 cos ) eiJ3 |k, , i

3i
= eiJ3 eiJ2 J3 eiJ2 eiJ3 eiJ3 eiJ2 eiJ3 |k, e
3 i = k |k, , i
= eiJ3 eiJ2 eiJ3 k |k, e
(9.75)
The state vector (9.70) is a smooth vector-valued function of (, ). This

except for the south
vector-valued function gives a unique state vector for all R
pole R = e3 , where = and (9.72) becomes

eiJ2 J3 eiJ2 = J3 .
(9.76)
eiJ2 eiJ3 eiJ2 = eiJ3 ,
(9.77)
This implies
iJ3 iJ2 iJ3
=e
iJ2
(9.78)
which may be used to obtain

3 i
|k, , i = eiJ3 eiJ2 eiJ3 |k, e
3 i
= eiJ2 e2iJ3 |k, e
3 i .
= eiJ2 e2ik |k, e
(9.79)
This shows that at the south pole different normalised state vectors are obtained
as varies in the range 0 < 2. also varies in the range 0 < 2 at the
north pole but |k, , i is single valued at the north pole because as a result of the
inclusion of the rotation R3 () in the definition of R(, ), |k, 0, i does not
depend on . It is therefore a smooth single-valued vector function everywhere
on S 2 except at the south pole.
A smooth vector-valued function which is well defined at the south pole but
not at the north pole is obtained by a gauge transformation (9.15), namely
|k, , i0 = ei2k |k, , i = eiJ3 eiJ2 eiJ3 ei2k |k, 0, 0i
= eiJ3 eiJ2 eiJ3 |k, 0, 0i .
(9.80)
(9.81)
The new state vector |k, , i0 differs from |k, , i by the phase factor eik (,) =
ei2k . At the south pole |k, , i0 evaluates to a single vector
3 i
|k, , i = eiJ2 |k, e
(9.82)
244
Topology in Physics
but at the north pole it evaluates to many vectors

3 i .
|k, 0, i0 = e2ik |k, e
(9.83)
Hence |k, , i can be used everywhere on S 2 except at the north pole. Either
vector (9.70) or (9.80) can be used in the overlap region O1 O2 of the two open
patches
O1 := S 2 \ {S} and O2 := S 2 \ {N },
(9.84)
of S 2 .
The state vector |k, , i0 is an eigenvector of J3 with eigenvalue k and an
eigenvector of R(,
) J = e3 J = J3 with eigenvalue k.
are needed. HowWe thus see that two different parameterisations of |k, Ri
0
ever, since |k, , i and |k, , i differ only by the phase factor ei2k the projection operators and corresponding subspaces rays in the Hilbert space coincide:
|k, , i hk, , | = |k, , i0 hk, , |0
(9.85)
Calculating the Mead-Berry Connection and the Berry Phase for Adiabatic Evolutions
We next calculate the Mead-Berry connection one-form Ak for adiabatic evolution of the Hamiltonian (9.62) using the previously derived formula (9.36). By
definition
Ak (R) = Aki dRi = i hk, R|
|k, Ri dRi ,
Ri
(9.86)
where i = 1, 2 correspond to the coordinates and of either of the patches O1

or O2 of the sphere S 2 .
On the patch O1 in which (9.70) holds, we have
|k, , i
3 | iU (, ) U (, ) |k, e
3 i =: rA ,
= hk, e
Ak = i hk, , |
|k, , i
3 | iU (, ) U (, ) |k, e
3 i := r sin A ,
= hk, e
Ak = i hk, , |
(9.87)
(9.88)
245
where A and A are defined for future use. Next we compute

iU (, )
U (, ) = ieiJ3 eiJ2 eiJ3 eiJ3 eiJ2 eiJ3
iJ3 iJ2 iJ3 iJ3

= ie
e e e
(iJ2 )eiJ2 eiJ3
= eiJ3 J2 eiJ3
= J2 cos J1 sin ,
iU (, ) U (, ) = ieiJ3 eiJ2 eiJ3 eiJ3 eiJ2 eiJ3
iJ3 iJ2 iJ3

= ie
e e (iJ3 )eiJ3 eiJ2 eiJ3
(9.89)
+ ieiJ3 eiJ2 eiJ3 eiJ3 eiJ2 (iJ3 )eiJ3

= eiJ3 (J3 cos J1 sin )eiJ3 J3
= sin (J1 cos + J2 sin ) + J3 (cos 1).
(9.90)
Substituting (9.89) and (9.90) in (9.87) and (9.88), we find

3 | (J2 cos J1 sin ) |k, e
3i ,
Ak (, ) = hk, e
(9.91)
3 | (J1 cos + J2 sin ) sin

Ak (, ) = hk, e
3 i .
+ J3 (cos 1) |k, e
(9.92)
Leading us to
Ak (, ) = 0,
Ak (, )
(9.93)
3 | J3 (cos 1) |k, e
3i
= hk, e
= k(1 cos ),
6= .
(9.94)
Next we repeat the same calculation in the patch O2 in which the basis vectors
|k, , i0 of (9.80) are single valued. This leads to
k
3 | (J2 cos J1 sin ) |k, e

3i ,
A0 = hk, e
(9.95)
3 | (J1 cos + J2 sin ) sin

A0 = hk, e
3i .
+ J3 (cos 1) + 2k |k, e
(9.96)
The components of the Mead-Berry connection one-form are then given by

k
A0 = 0,
k
A0
= k(cos + 1),
(9.97)
6= 0.
(9.98)
According to the general theory of Section 2 we expect Ak to transform according

to (9.42). So with = 2k we receive
k
A0 Ak = d(, ) = d(2k) = 2kd.
(9.99)
246
Topology in Physics
Having obtained the expression for the Mead-Berry connection one-form, we

next compute the corresponding curvature two-form. According to (9.55) this is
given by
F k = dAk =
Ak
Ak
d d +
d d
(9.100)
In view of (9.93) and (9.94) we obtain

k
F k = F
d d = k sin d d.
(9.101)
We can now use the formula (9.58) to calculate the Berry phase angle for a closed
path C,
Z
Z
Z
k
(9.102)
k (C) =
F = k sin d d = k d mod 2,
S
where S is any surface, which has the closed curve C as its boundary, and where
d is the element of solid angle. Here we take the direction in which C is
traversed, to be right handed with respect to the surface normal of S (see figure
9.2).
dS = d d
C
C
dS
S2
C
S
Figure 9.2: The difference of the line integrals of A and A0 can transformed, using
Stokes theorem, into an integral over a closed 2-surface S S 0 .
Denoting by (C) the solid angle subtended by C, i.e.,
Z
(C) :=
sin d d,
S
(9.103)
247
we can write (9.102) in the form

k (C) = k(C)
mod 2.
(9.104)
As explained in section 2, unlike the single-valued basis vectors |k, , i and

the connection one-form Ak , the curvature two-form F k is independent of the
choice of local coordinates. Therefore in (9.102) we can use a surface S O1
or a surface S 0 O2 , as long as both S and S 0 have C as their boundaries, i.e.,
S = C = S 0 (see figure 9.2). In order to investigate the consequences of this
property of the curvature two-form, we next compute the Berry phase angle using
S 0 . This yields
Z
Z
Z
k
sin d d = k
sin d d,
(C) =
F = k
(9.105)
S 2 \S
S0
S0
Z

Z
=k
sin d d sin d d .
(9.106)
S2
where the direction of the normal of S 0 and the direction in which C is traversed
are again given by the right-hand rule. This means that the normal of S 0 points
into the sphere and the normal of S points out of the sphere. As the integral of
d over the whole unit sphere is 4 and the integral over S is given by (9.103)
we obtain
k (C) = k(4 (C))
modulo 2.
(9.107)
Comparing (9.107) and (9.104), we find that

k(C) = 4k k(C)
modulo 2,
(9.108)
which can only be satisfied if

1
3
k = 0, , 1, , 2, . . .
2
2
(9.109)
Thus we conclude that k must be an integer or a half-integer.

If the basis vectors (9.70) and (9.80) are also eigenvectors of J 2 as they happen
to be in our example, then the possible values of k must automatically be (9.109)
But if these vectors are not J 2 eigenvectors - as is the case for a molecule - then
(9.109) still holds.
So far C could have been any closed path on the unit sphere. We will now
calculate the Berry phase for the special path C 1 : [0, T ] S 2
C 1 (t) := R((t), (t)) = R( = const, = t),
(9.110)
yielding the phase

Z
k (C 1 ) = k
0
sin 0 d0 d = 2k(1 cos ).
(9.111)
248
Topology in Physics
Substituting (9.111) and (9.66) in (9.40) we obtain the expression for the adiabatically evolving state vector,
adiabatic
eibtk eik (t) |k, , ti ,

t
2
=
, k (t) = k (C).
T
T
(t)
(9.112)
(9.113)
In particular for t = T , we obtain from (9.51)

(T )
adiabatic
ei2 k eik (C ) (0),

b
(0) = |k, , 0i .
(9.114)
Using the fact that S 2 is a sub-manifold of R3 we re-express our results in terms

of the Mead-Berry (three-)vector potential Ak (9.35) and the curvature (three)vector of the field strength F k . To compute Ak , we may view the connection
one-form Ak as a one-form in R3 , i.e.,
Ak = Akr dr + Ak d + Ak d,
where Ak and Ak are given by (9.93) and (9.94) and Akr = 0. Next we write Ak
in a new (Cartesian) basis (dr, rd, r sin d):
Ak = Akr dr + Ak rd + Ak r sin d.
The new components Akr = Akr = 0, Ak = 0, and Ak are called the spherical
components of Ak . They yield the components of the Mead-Berry vector potential
in the spherical coordinates,
r + Ak e
+ Ak e
,
Ak = Akr e
(9.115)
r , e
and e
are the unit vectors in R3 along the r, and directions,
where e
respectively. Leading us to
Ak (, ) =
k(cos 1)
,
e
r sin
6= .
(9.116)
Repeating the same computations in the coordinate patch O2 , we find

k
A0 =
k(cos + 1)
,
e
r sin
6= 0.
(9.117)
2k
,
e
r sin
(9.118)
A0 is related to Ak by the gauge transformation

k
A0 Ak = =
where = 2k.
249
More interestingly, we can compute the curvature three-vector,

k
k
r = 2 R(,
e
).
(9.119)
2
r
r
This can be obtained either by taking the curl of the vector potential or using
the identity
Fk =
k
F k = F
d d = k sin d d = Frk (rd)(r sin d) = F k dS,
to read off F k directly.

As seen from (9.119) the curvature three-vector (field strength) is directed
). It has the familiar form of the magnetic
r = R(,
along the radial direction e
field of a monopole.
The Exact Solution of the Schr
odinger Equation
In the preceding section we employed the results of section 2 to compute the Berry
phase. Hence we used the adiabatic approximation in our computations without
checking for the validity. In this section we shall explore the exact time-evolution
of the state vector (t) which satisfies the Schrodinger equation
t) J .
(t) = H(t)(t), with H(t) = bR(,
(9.120)
t
In particular we are interested in the solutions of (9.120) which describe cyclic
evolutions (9.18):
i
|( )i h( )| = |(0)i h(0)| .
(9.121)
In the adiabatic approximation the period is given by the period of the precession of the magnetic field, = T = 2
.
The expression (9.112) for the state vector is, as we showed in section 2, incompatible with the Schrodinger equation (9.120) and can only be approximations.
For the example considered here, (9.120) can be exactly solved. In particular the
cyclic solutions are known.
In order to find the exact solution of the Schrodinger equation (9.120), we proceed in the following way. The magnetic field (9.65) rotates counter-clockwise by
an angle t about the e3 axis. Instead of rotating the observables (Hamiltonian,
etc.) counterclockwise and keeping the state vector the same (in the laboratory
frame) we can rotate the state vector (t) clockwise (by an angle t) about
the e3 axis and keep the observables the same (view the magnetic field from a
frame which rotates with the field, such that the field appears stationary). For
the observable quantities (expectation values) these two points of view give the
same result, e.g.,
h(t)| H(t) |(t)i = h(t)| eitJ3 H0 eitJ3 |(t)i

= eitJ3 (t) H0 eitJ3 (t) .
(9.122)
250
Topology in Physics
This suggests the following unitary transformation of the state vector (t).
(t) 0 (t) := U (t)(t) = eitJ3 (t).
(9.123)
Inserting (t) = U (t) 0 (t) into the Schrodinger equation (9.120) for (t), we
obtain the Schrodinger equation for 0 (t),
itJ3 0
e
(t) = H(t)eitJ3 0 (t)
t
eitJ3 i 0 (t) = (J3 eitJ3 + H(t)eiwtJ3 ) 0 (t)

t
i 0 (t) = (J3 + eitJ3 H(t)eitJ3 ) 0 (t) := H 0 (t) 0 (t).

t
(9.124)
The state vector 0 (t) evolves in time according to the transformed Hamiltonian
U
H(t) := U H U iU
= eitJ3 H(t)eitJ3 J3 .
t
However the operator H 0 does not depend on time, for

H 0 = H0 J3 = b cos J3 + sin J1 J3
b
(9.125)
(9.126)
This was to be expected since in the rotating frame B does not change. Because
H 0 is time independent, (9.124) can be immediately integrated. The result is
0
0 (t) = eitH 0 (0).
(9.127)
The operator H 0 can also be related expressed as

H 0 = e J ,
(9.128)
where
b

b
cos
e3 + sin e1 ,
r
b

:= b 1 +
2 cos .
b b
e :=
(9.129)
(9.130)
The unit vector e lies in the 1-3 plane. Therefore it can be written in the form
3 + sin e
1 = R(
0),
,
e = cos e
where is the angle between e and e3 , i.e.

cos
b
cos
=
cos :=
,
b
1 2 cos + 2
sin
b
sin = sin =
,
1 2 cos + 2
(9.131)
(9.132)
251
and is a dimensionless parameter given by

:=
.
b
(9.133)
In view of (9.128), the time development given by (9.127) represents a rotation of the state vector 0 (0) by an angle t about the e-axis,
0 (t) = eiteJ 0 (0).
(9.134)
Having obtained the exact expression for 0 (t) we now transform back to the
laboratory frame. Using (9.123) and (9.127), we find
0
(t) = eitJ3 eitH (0).
(9.135)
Therefore the solution of the Schrodinger Equation (9.120) for arbitrary initial
state vector (0) is given by
(t) = eitJ3 eiteJ (0) =: U (t)(0).
(9.136)
This means that the evolving state (t) is obtained from (0) by transforming
(0) by the unitary time-evolution operator U (t) representing a rotation by t
along e followed by a rotation by t along e3 .
), i.e., (9.132), involves the physical paThe relation between (, ) and (,
rameter . This relation can be expressed in terms of a function F : M = S 2
S 2 which (on the patch O1 including the north pole) is given by

cos
F (R(, )) = F (, ) = arccos
,
2 2 cos + 1
) = R(
).
,
= (,
(9.137)
A global expression for F may be obtained by viewing the sphere S 2 as embedded
in R3 . It is given by
F (x1 , x2 , x3 ) = p
(x1 , x2 , x3 )
(x1 )2 + (x2 )2 + (x3 )2
(9.138)
where (x1 , x2 , x3 ) R3 . For < 1, the function F is a smooth diffeomorphism

of S 2 and consists of a translation in the e3 direction followed by a projection
along R onto S 2 .
Examining the condition (9.26) of validity of the adiabatic approximation it
is not difficult to see that for our example (9.62), the left-hand side of (9.26) is
proportional to the frequency of the precession of the magnetic field. since the
intrinsic frequency scale of the system is given by b, the adiabatic approximation
is valid if and only if = /b 1.
252
Topology in Physics
The frequency b is the angular velocity with which the state vector (t) rotates
0). For = 2/T = 0 ( = 0),
about the direction of the magnetic field R(,
(t) = eitbR(,0)J (0).
(9.139)
This is most evident if we use (9.129) and (9.133) to rewrite (9.136) in the form
(t) = eitJ3 eitR(,0)J (0),

= eitbJ3 eitb(R(,0)J J3 ) (0),
(9.140)
and then specialise to the case = 0. Equation (9.139) gives the time-evolution
of a state vector (0) for the time-independent Hamiltonian
0) J ,
H = bR(,
= constant.
(9.141)
Having obtained the exact solution (9.140) of the Schrodinger equation (9.120)
we can identify the cyclic solutions, i.e., the solutions which satisfy (9.121). By
definition the cyclic solutions with period are obtained by choosing the initial
state vector (0) to be an eigenvector of the evolution operator
U ( ) = ei J3 ei R(,0)J .

(9.142)
The corresponding eigenvalue ei is the total phase factor, for

( ) = U ( )(0) = ei (0).
(9.143)
In general, in order to find the cyclic states for arbitrary , one must find the
(non-stationary) eigenstates of the evolution operator (9.142). This is quite easy
if is chosen in such a way that the initial state is an eigenstate of both of the
operators appearing on the right-hand side of (9.142). In this case
h
i

ei J3 , ei R(,0)J |(0)i = 0.
(9.144)
It is not difficult to see that this condition is satisfied for = T = 2/. There
are other possible values for , but were merely interested in = T since that is
the period of our adiabatic cyclic solutions.
Cyclic solutions with a period = T are obtained by finding the simultaneous

eigenvectors of the operators ei2J3 , eiT R(,0)J , and therefore the evolution
operator
U (T ) = ei2J3 eiT R(,0)J .

(9.145)
These eigenvectors which we shall denote by k are consequently also eigenvectors

of the operator
0) := bR(
0) J ,
,
H(,
(9.146)
253
i.e., they satisfy

0) |k i = bk |k i ;
H(,
e J |k i = k |k i .
(9.147)
From (9.70) and (9.71) with replaced by and by 0 it follows that k is (up
to a phase factor) given by
2
0i = U (,
0) |k, e3 i = eiJ
|k i = |k, ,
|k, e3 i.
(9.148)
to obtain eigenvalues of ei2J3 we proceed using (9.74) with = 2. This yields

2
0i = ei2J3 eiJ
ei2J3 k = ei2J3 |k, ,
|k, 0, 0i
= eiJ2 ei2J3 |k, 0, 0i

0i.
= ei2k |k, ,
(9.149)
The initial state vectors which lead to a cyclic evolution with period T = 2/
are thus labelled by integers of half-integers k and given by (9.148),
0i,
|(0)i = |k i = |k, ,
(9.150)
(after an arbitrary phase factor has been fixed). The total phase of these cyclic
evolutions, i.e., the eigenvalues ei of (9.143) are also labelled by k and given
by
(T ) = U (T )(0) = eik (0) = ei2k ei2 k (0).
(9.151)
Thus, in this case, the possible initial states of an cyclic evolution are very similar
to the initial states of an adiabatic evolution. The only difference is that the initial
states of an adiabatic evolution are given by the eigenvectors |k, , 0i of the initial
0) J , whereas the initial states of the exact cyclic
Hamiltonian H(R(0)) = bR(,
evolutions are given by the eigenvectors (9.148) of the operator
0) = bR(
0) J ,
,
0) := H(,
H(,
(9.152)
which is different from H(R(0)).

The exact cyclic states are the states with the component k of angular mo 0) rather than R(,
,
0). For large values of
mentum along the direction e = R(
( b) these to directions can be very different, but for = /b 0 they
coincide.
We will next determine the analogue of the single-valued eigenvectors |k, R(t)i =
|k, (t), (t)i which are used in the calculation of the Mead-Berry connection oneform and Berry phase. These vectors k (t) lie on a curve above the closed curve
C : t |(t)ih(t)|;
|(T )ih(T )| = |(0)ih(0)|.
(9.153)
254
Topology in Physics
in the state space P(H). This means that they fulfil

k (t) = phase factor (t), and k (T ) = k (0).
(9.154)
There are many such vectors all differing by a phase transformation,

k (t) 0k (t) = eik (t) k (t),
k (0) = k (T )
modulo 2.
(9.155)
One and by far the most obvious choice for k (t) is

t)|k, e3 i
k (t) := U (,
ti.
= eitJ3 eiJ2 eitk |k, e3 i =: |k, ,
(9.156)
= ti are eigenvectors of the operator

These single-valued basis vectors |k, ,
) = H(F (, )) = bR(
) J ,
,
) := H(,
H(,
(9.157)
where F is the function given by (9.137). The operator H(R)

= H(F (R)) is
not the Hamiltonian but another parameter-dependent operator which can as
well serve to define an orthonormal basis of the Hilbert space H. In terms of the
i, the cyclic solution of the Schrodinger equation (9.120) is
known vectors |k, ,
given by
ti.
(t) = eitk eitk |k, ,
(9.158)
This relation expresses the cyclically evolving state vector (t) in terms of the
ti. It is the generalisation of the adiabatic
known quantities , and |k, ,
equality (9.112). Similarly, (9.151) is the generalisation of (9.114). However
the phase factor in (9.158) and (9.151) is not in the form of a product of the
dynamical and the geometrical parts. Thus, in order to obtain an expression for
each part, we must calculate at least on of them independently.
Dynamical and Geometrical Phase Factors for Non-Adiabatic Evolution
The splitting of the phase factor into a dynamical and a geometrical part can be
performed in two different ways. Either one gives an argument why a certain part
of the total phase is geometrical and obtains the dynamical part as the difference
between the total and the geometrical part, or one defines the dynamical part and
obtains the geometrical part as the difference between the total and the dynamical
part. We will pursue the latter approach, namely define the dynamical phase
angle kdyn and obtain the geometrical phase angle kgeom as a derived quantity,
kgeom = k kdyn .
(9.159)

The dynamical phase (angle) for general cyclic evolution is defined as
Z T
dyn
k :=
h(t)|H(t)|(t)idt,
255
(9.160)
the phase angle for the evolution of a stationary state as in (9.29).

We can easily calculate the dynamical phase (9.160) for the Hamiltonian
(9.62),

dyn
k = k2
+ cos .
(9.161)
Using (9.159) and (9.151), we then find

kgeom

= k2(1 cos ) = k2 1 + cos .

(9.162)
This has the same form as (9.111) for the adiabatic case except that is replaced
by of (9.132).
In the adiabatic approximation, = /b 1, we can expand the expressions
(9.130) and (9.132) with respect to . Up to first order in , we have
1 cos ,
(9.163)
b
(9.164)
sin sin + sin 2,
2
(9.165)
cos cos + (cos 2 1)
2
From this we conclude that the adiabatic approximation of the geometrical phase
angle for a general cyclic evolution (9.162) is identical with the Berry phase angle,
kgeom 2k(1 cos ) = kBerry .
(9.166)
These results justify the choice of (9.159) as a definition of the geometrical phase
for the general cyclic evolution. Next, we will show that the non-adiabatic geometric phase angle can be obtained from a connection one-form in the same way
the Berry phase angle is obtained from the Mead-Berry connection one-form.
In analogy to (9.35) and (9.36), we define the following connection one-form
Ak := ihk |d|k i,
(9.167)
and compute
|d|k, ,
i
Ak = hk, ,
id + hk, ,
| |k, ,
id
| |k, ,
= hk, ,
| |k ,
| |k, ,
id.
i d + hk, ,
= hk, ,
(9.168)
256
Topology in Physics
Here we used the basis vectors given by (9.156), i.e., the eigenvectors of the
operator H(R).
Comparing (9.168) with (9.86) and using (9.93) and (9.94), we
conclude

b
A = k(1 cos )d = k 1 cos +

d.
(9.169)
The curvature two-form (field strength) that follows from this connection oneform is given by
1
F k = dA = k h
1
cos
cos +
i3/2
2
b
sin d d.
(9.170)
The component
k
F
= k h
1
cos
cos +
i3/2
2
b
sin
(9.171)
of F k can be used to define the field strength (three-)vector

F k =
1 b cos
k
).
h
i3/2 R(,

r2
2
1 b cos + b
(9.172)
This is again a monopole type field similar to (9.119) but with modified monopole
strength. The connection one-form (9.169) has the same form as (9.94) for the
Mead-Berry connection, except that here it depends on the angle which differs
from the angle .
We can now use the analogue of (9.58) for the adiabatic approximation and
define a geometric phase angle also for the general cyclic evolution by
I
I
k
(9.173)
k :=
A = ihk |d|k i.
C
Then using (9.169), we obtain

Z
k = k
(1 cos )dt
= 2k(1 cos )

b
= 2k 1 cos +
.
This agrees with the result (9.166) obtained from (9.159) and (9.160).
(9.174)
257
With these results we can rewrite the cyclic solution (9.158) of the Schrodinger
equation in a form which completely resembles the adiabatic approximation, i.e.,
(9.112),
dyn
(t) = eik
(t) ik (t)
i,
|k, ,
(9.175)
where
Z
kdyn
:=
h(t0 )|H(t0 )|(t0 )idt0 = tk

+ cos
tk
,
ZT
t
d
ihk (t0 )| 0 |k (t0 )idt0
dt
0
Z t
0
= tk(1 cos )
= k (1 cos )dt
k (t) :=
tk
=
.
T
The concepts introduced in this section for the cyclic evolution, (9.153), are
the analogues of the eigenvectors |k, , i of the Hamiltonian. The connections
Ak of (9.169) are generalisations of the Mead-Berry phase (9.48).
The distinction between (9.175) and (9.112) is that in (9.175) we have an
equality not an approximate one. This means that the pure state corresponding
to (t) is indeed an exact cyclic evolution. If the frequency of the precession of the
magnetic field is small enough, i.e., much smaller than the frequency b, then
the curve (9.112) is close enough to (9.175), in order to provide an acceptable
approximation for the geometric phase. Nevertheless an exact adiabatic cyclic
evolution does not exist.
The geometrical phase (9.174) is not purely geometrical in the way that the
Berry phase (9.166) is. Unlike the Berry phase that only depends on and is
solely given by the path in the parameter space, the geometrical phase, for general
cyclic evolution, also depends on the parameter = /b of the Hamiltonian.
Experimental Verification of Berrys Phase
In 1986 Akira Tomita and Raymond Chiao [52] reported the first experimental verification of Berrys phase. The experiment consisted of a helically wound
optical fibre inside which a linearly polarised photon could be adiabatically transported around a closed path in momentum space, with the polarisation of the
photon being a direct measure for the Berry phase.
258
Topology in Physics
Figure 9.3: Measured angle of rotation of linearly polarised light vs calculated

solid angle in momentum space. Open circles, the data for uniform helices;
squares and triangle, nonuniform helices; solid circles, arbitrary planar paths.
The solid line is the theoretical prediction.
259
The first Experiment

The parameter space of this system is the photons momentum space, so in order
to form a closed path in momentum or k space, the propagation directions of
the input and output of the fibre were kept identical and the fibre was wound
into a uniform helix, i.e., the pitch angle26 was kept constant throughout the
entire helix. Since the modulus of the photons momentum k doesnt change as
it propagates along the fibre, the parameter space reduces to the sphere Sk2 with
radius k. The constant pitch angle then corresponds to the photon tracing a
k
k
k
k
Figure 9.4: The momentum vector k of the photon traces a circle of constant
longitude on the sphere Sk2 in momentum space.
circle of constant longitude. So we have a situation that is very similar to what
we encountered in section 3. Allowing us to reuse the formulae we computed.
Formula (9.104) then go over to
(C) = (C) = 2(1 cos ),
(9.176)
where = 1 is the helicity quantum-number of the photon. Our previously

formulated theory then predicts that (C) is the Berry phase for = 1, which
corresponds to the angle of rotation of linear polarisation. As can be seen in
the figure 9.3 the theory predicts the experimental results very accurately.
The second Experiment
In the second experiment the topological aspects27 of the Berry phase were tested.
The pitch angle was no longer kept constant. But as we can see in figure 9.3
these deviations from the uniform helix do not change the berry phase as long as
the average pitch angle is the same as before.
26
27
The angle between the local waveguide axis and the axis of the helix.
These will be discussed in detail in the Berry Phase II talk.
260
Topology in Physics
The Aharonov-Anandan Lift of Cyclic States
In section 3 we examined the phenomenon of the geometric phase for a class

of non-adiabatic cyclic evolution for the specific example of the time-dependent
Hamiltonian (9.62). In this section we shall uncover the general pattern that
underlies the results obtained in section 3.
In general we consider a time-dependent Hamiltonian H(t) = H(R(t)) whose
time-dependence is given by a path C in a parameter space M
C : R(0) R(t) R( ).
(9.177)
The time evolution of every pure physical state defines a curve in the space P(H)
of all such states. In particular, a cyclic state W (t) = |(t)ih(t)| with period
corresponds to a closed curve in P(H):
C : W (0) W (t) W ( ) = W (0).
(9.178)
In our example (9.62) the cyclic states are labelled by integers or half-integers k.
They correspond to the closed paths
Ck : Wk (0) Wk (t) = |k (t)ihk (t)| Wk (T ) = Wk (0),
(9.179)
in the state space P(H), where k (t) is given by (9.156).

Consider an arbitrary cyclic evolution with period . Then associated with
the corresponding closed curve C in P(H) are three different curves in the Hilbert
space H:
1. The curve
C : |(0)i |(t)i |( )i = ei |(0)i,
(9.180)
where |(t)i is the solution of the Schrodinger equation with the initial
state vector |(0)i being cyclic.
2. The closed curve
C closed : |(R(0))i |(R(t))i |(R(T ))i = |(R)i,
(9.181)
where |(R(t))i is the generalisation of |k (t)i of (9.156), i.e., |(R)i is

a smooth single-valued function with values in H which has the property
W (t) = |(R(t))ih(R(t))|. |(R)i is determined up to gauge transformations:
|(R(t))i |0 (R(t))i = ei(R(t)) |(R(t))i,
with ei(R( )) = ei(0) .
(9.182)
{
{
dyn
(T )
)
(T
(0)
261
(t)
(t)
(t)
(T ) = (0)
Figure 9.5: Closed path in the space of physical states and its lifts.
262
Topology in Physics
In our example (9.62) the closed curve in H associated with the curve Ck
of (9.179) is given by
ti |k (T )i = |k (0)i,
Ckclosed : |k (0)i |k (t)i = |k, ,
(9.183)
where the vectors |k (t)i are those of (9.156) and the gauge transformation
(9.182) is the one in (9.155). The curve (9.180) in our example is given by
the vectors (9.175).
3. The curve
C : |(0)i
|(t)i
:= ei
Rt
0 h(t
0 )|H(t0 )|(t0 )idt0
|(t)i |( )i,
(9.184)
where |(t)i is the solution of the Schrodinger equation.

These three curves C, C closed and C have the property that under the projection of H onto P(H) (state vectors onto states) they project onto the closed
curve C of (9.178), i.e.,
|(t)ih(t)| = |(t)ih(t)| = |(t)ih

(t)|
= W (t) P(H).
(9.185)
For this reason we call these three curves lifts of C. Figure 9.5 offers a schematic
illustration of the situation. The curves C, C closed and C will be called the dynamical lift, the closed lift and the Aharonov-Anandan (A-A) lift of the closed curve
C, respectively. An important property of the A-A lift C is that unlike the closed
lift C closed and the dynamical lift C, the A-A lift C is uniquely determined by
C. The non-uniqueness of the closed lift is due to the fact that the single-valued
state vectors |(R(t))i are only defined up to gauge transformations (9.182). We
shall discuss the non-uniqueness of the dynamical lift and the uniqueness of the
A-A lift directly. The A-A lift is also called the horizontal lift.
In our example (9.62) the A-A lift is given by
|(t)i
:= ei
Rt
0 h(t
0 )|H(t0 )|(t0 )idt0
|(t)i = eik(1+ cos )t |(t)i.
(9.186)
It is not a closed curve in H, but has the property
)i = eik (T ) |(0)i,
Ck : |(0)i
|(t)i
|(T
(9.187)
where k (T ) = kgeom is the geometrical phase (9.162) or (9.174).

The relation between the A-A lift, the dynamical lift and the closed lift in our
example is expressed by
dyn
k i = eik |(t)i = eik (t) |k, ,

(t)i,
|(t)
where
Z
kdyn (t)
:=
0
h(t0 )|H(t0 )|(t0 )idt0 .
(9.188)
(9.189)
263
Provided that (t) is a solution of the Schrodinger equation (9.3), it follows that
i
Rt
d
0
0
0
0 d
|(t)i = h(t)|H(t)|(t)i|(t)i
+ iei 0 h(t )|H(t )|(t )idt |(t)i
dt
dt
Rt
0 )|H(t0 )|(t0 )idt0
i
h(t
= h(t)|H(t)|(t)i|(t)i
+e 0
H(t)|(t)i
= h(t)|H(t)|(t)i|(t)i
+ H(t)|(t)i
Thus the A-A lift
|(t)i
:= ei
Rt
0 h(t
0 )|H(t0 )|(t0 )idt0
|(t)i
(9.190)
satisfies the following equation and initial condition:

i
|(t)i = [H(t) h(t)|H(t)|(t)i1]|(t)i,

dt
|(0)i
= |(0)i.
(9.191)
or
Taking the inner product of the right-hand side of (9.191) with |(t)i
|(t)i, we obtain
d
|(t)i = 0
dt
d
h(t)| |(t)i = 0
dt
h(t)|
(9.192)
of |(t)i
is orthogonal (in the Hilbert
This means that the tangent vector dtd |(t)i
space sense) to both |(t)i and |(t)i.

Although the dynamical lift C is uniquely determined by the Hamiltonian
H(t), it is not uniquely determined by the physical problem. For example, the
simple substitution
H(t) H 0 (t) = H(t) (t)1
with (t) R
(9.193)
leads to a new Hamiltonian H 0 (t) which describes the same physics, i.e., it has the
same (closed) curves of physical states t W (t) = |(t)ih(t)| in the projective
Hilbert space P(H) as H(t) does. But H(t) and H 0 (t) define different dynamical
lifts t (t) and t 0 (t). Moreover, two Hamiltonians H(t) and H 0 (t) which
have the same curves of physical states t |(t)ih(t)| and t0 | 0 (t)ih 0 (t)|
differ by a multiple (t)1 of the unit operator 1. Thus the dynamical lift does not
uniquely correspond to the physical problem. A lift that is uniquely associated
with the closed curve C : t |(t)ih(t)| in P(H), and therefore reflects the
physical effects, is the A-A lift C : t (t).

We will now discuss the generalisations of (9.183), i.e., closed lifts (9.181) of
the curve C. As seen from (9.173), |k (t)i can be used to calculate the geometric
phase, if one does not have a solution |(t)i of the Schrodinger equation. There
264
Topology in Physics
are various ways to obtain these single-valued |(t)i H. In general they are
curves of local sections. A local section is a smooth function of an open patch O
of P(H) into H. We will assume that our curve C lies in such an open patch O.
Then for any closed curve C : t W (t) one has the closed (single-valued) lift
t |(t)i = |(W (t))i,
(9.194)
W (t) = |(t)ih(t)|, and |( )i = |(0)i.
(9.195)
with
This is however determined only up to a gauge transformation (9.182). The

importance of the local sections |i or alternatively the closed lifts C closed is due
to their utility in the calculation of the geometric phase.
Since t |(t)i, t |(t)i

and any t |(t)i are lifts of the same closed
curve t W (t), there must exist a phase factor (t) such that
|(t)i
= (t)|(t)i.
(9.196)
Next we calculate (t). Differentiating both sides of (9.196), we obtain

d
d(t)
d
|(t)i =
|(t)i + (t) |(t)i.
dt
dt
dt
(9.197)
Taking the scalar product of this expression with |(t)i,

we have
(t)
d(t)
d
d
+ |(t)|2 h(t)| |(t)i = h(t)|

|(t)i = 0,
dt
dt
dt
(9.198)
where in the last equality we have used (9.192). Next we write (9.198) in the
form
1 d(t)
d
= h(t)| |(t)i,
(t) dt
dt
which can be integrated to yield
Rt
d
(t)
0
0
0
= e 0 h(t )| dt0 |(t )idt .
(0)
(9.199)
We shall denote the phase factor on the right-hand side of (9.199) by ei(t) and
write (9.196) as
|(t)i
= (0)ei
Rt
0
ih(t0 )| dtd0 |(t0 )idt0
|(t)i =: (0)ei(t) |(t)i.
(9.200)
If we chose the arbitrary constant phase (0) equal to 1 such that |(0)i
=
|(0)i = |(0)i, then we have
|(t)i
= ei(t) |(t)i.
(9.201)
265
This is the generalisation of (9.188). Combining (9.190) and (9.201), we find

|(t)i = ei
Rt
0 h(t
0 )|H(t0 )|(t0 )idt0
ei(t) |(t)i.
(9.202)
Equation (9.202) is the general relation between the solution of the Schrodinger
equation |(t)i and the closed lift |(t)i. Equation (9.40) is the adiabatic approximation of this relation, and (9.158) is its special case for the spinning quantum
system in a precessing external magnetic field. The adiabatic approximation uses
in place of the closed lifts |(t)i the eigenvectors of H(R(t)) which are more easily
accessible than |(t)i. According to (9.200), for the closed path C in P(H) the
phase angle ( ) is given by
I
I
d
( ) = (C) =
ih(t)| |(t)idt = ih|d|i modulo 2,
(9.203)
dt
0
where |(t)i corresponds to any of the closed lifts of C.
The phase angle ( ) is independent of the choice of the time parameterisation
of |(t)i, i.e., the speed with which |(t)i traverses its closed path. It is gauge
invariant. It is independent of the choice of the Hamiltonian as long as these
Hamiltonians describe the same closed path C in P(H). It depends only on the
closed curve C. It is therefore considered to be a geometric property of C, thus
has the name geometric phase.
The one-form appearing in the integrand of (9.203),
A := ih|d|i,
(9.204)
is called the Aharonov-Anandan connection one-form. It is the analogue of the

Mead-Berry connection one-form (9.36). It satisfies the following transformation
rule under gauge the transformation (9.182)
A A0 = A d.
(9.205)
The formula (9.204) for the A-A connection was obtained from the require is the A-A lift, i.e., the lift fulfilling (9.192) (which in turn followed
ment that (t)
from its definition (9.190)). This was the only possible definition of a lift which
depended solely on the physics of the problem.
Equations (9.201) and (9.194) lead to
)i = ei(C) |(0)i
|(
= ei(C) |(0)i.
(9.206)
Then, using (9.190) one obtains for the cyclic evolution of a state vector
|( )i = ei
R
0
h(t0 )|H(t0 )|(t0 )idt0 i(C)
|(0)i.
This relation is the non-adiabatic generalisation of (9.51).
(9.207)
266
Topology in Physics
Figure 9.5 shows a graphical representation of our description of the general

cyclic evolution. It shows a base space representing the space P(H) of physical
states and including a closed curve C. This curve represents the cyclic evolution of
a pure state W (t). Above C are shown the lifts C closed , C and C of C which belong
to the Hilbert space H (depicted in the figure by the three dimensional space).
Also shown is a fibre above the base point W (0) (depicted by the positive z-axis
with the z coordinate representing the phase angle modulo 2 or the element
ei(0) of the (gauge) group U(1). The fibre in this case (though it has been drawn
as a straight line) is a copy of the unit circle S 1 or the group U(1). We can
attach a copy of this S 1 not only to the point W (0) but to every one-dimensional
projection operator P(H). In this way we get a bundle of U(1) fibres attached
to each point of P(H).
The mathematical structure that we have encountered here is an example
of a principal fibre bundle (PFB). We shall refer to this PFB as the AharonovAnandan principle bundle.
Like any fibre bundle the A-A PFB consists of a base space, a set of fibres
associated with the points of the base space and forming a larger space called the
total or bundle space, and a structure group which acts on the fibres. As the total
space consists of fibres over points of the base space, we can define a projection
map from the total space to the base space. The fibres are then viewed as the
inverse images of points of the base space under this projection.
For the A-A PFB, the base space is the projective Hilbert space P(H) also
denoted by CP N 1 where N is the dimension of the Hilbert space H. The
total space S(H) is the set of all normalised state vectors in H, i.e.
S(H) := { H : h|i = 1} = S 2N 1 .
(9.208)
The projection map

(or : S 2N 1 CP N 1 )
: S(H) P(H)
(9.209)
is the obvious projection of the state vectors onto states,

(|i) = |ih|.
(9.210)
The fibres are normalised rays in the Hilbert space. They consist of all the
normalised state vectors associated with a given pure state and hence differing
by a phase factor. Therefore the fibres are copies of the unit circle S 1 . Finally
the structure group is the Abelian group U(1) which has the manifold structure
of S 1 as well. We shall denote the A-A PFB by
: U(1)
/ S(H)
or
U(1)
/ S 2N 1
P(H)
CP N 1 .
(9.211)
267
A PFB may be endowed with a geometric structure. The latter provides a

notion of parallel transportation or alternatively the notion of a horizontal lift. A
geometric structure on a PFB is also called a connection. For an Abelian PFB,
i.e., a PFB with U(1) as its structure group, a connection may be expressed by a
differential one-form satisfying a set of (gauge) transformations rules. The A-A
connection one-form A of (9.204) defines a particular connection on the A-A PFB
. The A-A phase also has a very well-known mathematical counterpart called
holonomy.
Exact Cyclic Evolution for Periodic Hamiltonians
If the time-dependent Hamiltonian is periodic, i.e.
H(t + T ) = H(T ),
(9.212)
for some time period T , then we can use the results of what is known as Floquet
theory in mathematics. In particular we next show that the evolution operator
for any periodic Hamiltonian has precisely the same form as the evolution operator (9.142) for the spin system (9.62). More precisely we prove that for any
T -periodic Hamiltonian, satisfying (9.212), there exists a time-independent Her and a T -periodic unitary operator Z(t) with Z(0) = 1, such
mitian operator h
that th time-evolution operator is given by
U (t) = Z(t)eith .
(9.213)
To see this, consider the unitary operator V (t) := U (t + T ). V satisfies the

following Schrodinger equation:
d
V (t) = iH(t)V (t)
dt
V (0) = U (T ) =: V0 ,
(9.214)
since
i
d
V (t)|(0)i = H(t)V (t)|(0)i,
dt
for all |(0)i.
The operator V 0 (t) := U V0 also satisfies (9.214). However we know from the
uniqueness theorem for initial value linear differential equations that solution of
(9.214) is unique. Therefore we have V (t) = V 0 (t), i.e.,
U (t + T ) = U (t)V0 = U (t)U (T ).
(9.215)
which satisfies
This relation is sufficient to construct a pair of operators (Z(t), h)
(9.213). To see this we first write t = nT + t0 for some integer n and t0 [0, T )
and then apply (9.215) repeatedly. This yields
U (t) = U (t0 )[U (T )]n = U (t0 )V0n .
(9.216)
268
Topology in Physics
Next we assume that the unitary operator V0 can be expressed as the exponential
of a Hermitian operator28 , i.e.,
0
V0 = eith .
In view of (9.216),(9.217) and n =
tt0
,
T
(9.217)
we have
U (t) = U (t0 )e
= U (t0 )e
= Z(t)e
i(tt0 ) 0
h
T
it0 0
h
T
ith
it 0
eT h
(9.218)
(9.219)
(9.220)
where we have defined

it0 0
Z(t) := U (t0 )e T
0
:= h
, and h
T
is hermitian and Z(t) is unitary. Furthermore, Z(t) satisfies

Clearly, h
Z(t + T ) = Z(t), and Z(0) = U (0) = 1.
(9.221)
For t = T , (9.218) reproduces the general result, namely
U (T ) = eiT h .
(9.222)
yields the exact cyclic states with period = T as its eigenstates.

In particular, h
so constructed, is not unique. The nonThe pair of operators (Z(t), h),
uniqueness of h is reminiscent of the non-uniqueness of the Hamiltonian H(t)

discussed previously.
If the Hamiltonian is parameter dependent (H = H(R)) and its time depen will
dence is realised by changing the parameters R in time, then the operator h
be a function of the initial parameter, R0 = R(t = 0). Thus in general it also
depends on a set of parameters belonging to the same parameter space M .
As in the case of the spin system of Section 3, the operator h(R)

generally
differs from the Hamiltonian H(R). In general, the dependence of h(R)

on the
Hamiltonian is not explicitly known. The only established fact is that for an
adiabatic evolution H(R)

can be approximated by H(R).
28
This can always be done if the Hilbert space is finite dimensional. For an infinite dimensional Hilbert space there are unitary operators which cannot be written as the exponent of (i
times) some Hermitian operator. However, the set of unitary operators which are exponentials
of Hermitian operators is a dense subset of the set of unitary operators with an appropriate
choice of topology on the latter set. This means that one can always find a sequence of unitary
0
operators {eihk : k = 0, 1, 2, . . . } which converges to V0 in this topology. Therefore, one can
0
0 := h
0 for some large value of k.
approximate V0 with eih where h
k
269
In the remainder of this section we briefly discuss the relation between the
and Z(t) of (9.213) and the geometric phase of Aharonov and Ananoperators h
dan.
Consider the evolution of a cyclic state vector |(0)i. By definition |(0)i is
an eigenvector of the evolution operator:
U ( )|(0)i = Z( )ei h |(0)i = ei( ) |(0)i,
(9.223)
where we have employed (9.213). On the other hand the evolution of |(0)i is
governed by the Schrodinger equation (9.3). In terms of the operators Z(t) and
these equations take the form
h,
|(t)i = Z(t)eith |(0)i,
(t),
H(t) = iZ(t)Z
(t) + Z(t)hZ
(9.224)
(9.225)
where we have used the unitarity of Z(t), and Z(t)

stands for the time-derivative
of Z(t).
Next we compute the dynamical phase angle. Substituting (9.224) and (9.225)
in the definition of the dynamical phase angle (9.160)
Z
dyn
( ) :=
h(t0 )|H(t0 )|(t0 )idt0 modulo 2,
(9.226)
0
we have
dyn
( ) =
0 )eit h |(0)idt0
ih(0)|eit h Z (t0 )Z(t
(9.227)
modulo 2.
+ h(0)|h|(0)i
(9.228)
For a cyclic state with the same period as the Hamiltonian ( = T ), the cyclic
Hence the second term on
state vector |(0)i is an eigenvector of the operator h.
the right-hand side of (9.227) is precisely the totally phase angle
h(0)|h|(0)i
= (T ).
(9.229)
In view of this equation we can directly express the geometric phase in terms of
the operator Z(t) and its time derivative:
(T ) = (C) := dyn (T ) (T ) modulo 2
Z T
0 )|(0)idt0 modulo 2.
=
ih(0)|Z (t0 )Z(t
(9.230)
Equation (9.230) is of some practical importance. Since both the Hamiltonian

H(t) and the operator Z(t) are periodic with the same period T , one can in
principle expand them in Fourier series. There are procedures to relate the Fourier
270
Topology in Physics
components of Z(t) to those of H(t) which may be used to yield a series expansion
of the geometric phase for arbitrary periodic systems.
The operator Z(t) is also of particular interest since it may be used to yield
single-valued vectors |(t)i of (9.194), namely
|(t)i = Z(t)|(0)i.
(9.231)
In view of this identification, we can write (9.230) in the form

Z
(T ) =
Z
ih(0)|Z (t0 )
0
T
=
0
ih(t0 )|
d
[Z(t0 )|(0)i]dt0
dt0
d
|(t0 )idt0
0
dt
modulo 2
modulo 2,
which is identical with (9.203). The non-uniqueness of Z(t), mentioned above,

corresponds to the gauge freedom of |(t)i.
Furthermore, we can use (9.231) to express the evolution of a cyclic state
vector according to
|(t)i = Z(t)eith |(0)i

= ei
(T )
t
T
(T )
i T t
=e
Z(t)|(0)i
|(t)i.
(9.232)
with eigenvalue
Here we have used the fact that |(0)i is an eigenvector of h
(T )/T .
The fact that the evolving state vector |(t)i is expressed as the product of
a phase factor and a periodic state vector |(t)i is a direct consequence of the
periodicity of the Hamiltonian. The state vectors of the form (9.232) are encountered in almost all physical systems with periodic features. The corresponding
wave functions are known as the Bloch wave functions.
10 Berry phase II: The

Aharonov-Bohm effect
Oliver Pf
affli
Interesting phenomena of the electromagnetic potentials in quantum systems will be discussed. In contrast to the conclusion of
classical mechanics, potentials cause effects on charged particles,
even in regions where all the forces on the particles vanish. The
Aharonov-Bohm effect and the Aharonov-Casher effect are examples for this fact. We will see that the A-B phase is a topological
phase as well as a Berry phase. That the Aharonov-Casher effect
is a variation of the A-B effect will, finally, be demonstrated.
Introduction
The time evolution of quantum states is totally different to the time evolution of
classical objects. This will be demonstrated with different experimental setups
which have in common, that the states do not undergo any forces but that they
live in a regime with non-zero potentials. The potentials cause physically relevant
effects even if all forces vanish. A well-known example for this is the AharonovBohm (A-B) effect which occurs if electrons move around an area with a magnetic
field. The quantum state of the electrons changes its phase when propagating even
if the magnetic field is confined to an area where the wave function of the electron
is zero. We will see, that the A-B phase is a topological phase and a Berry phase.
This means that the A-B phase is invariant under continuous deformations of the
electrons path and that the A-B effect can be described by Berrys formalism. His
idea was to parameterise the environment of a quantum state. He concluded that
under certain conditions, basically the adiabatic assumption, this dependence
on the environment results in additional phase shifts, which are caused by the
dynamics of the quantum state and the geometry of its path. In the case of the
271
272
Topology in Physics
A-B effect the area, to which the magnetic field is confined, is the environment of
the electron moving around it. Furthermore, the Aharonov-Casher (A-C) effect,
which is a variation of the A-B effect, will be described. Instead of electrons
we analyse magnetic moments, for example neutrons. Classically, Neutrons do
not interact with electric fields but, similar to the A-B effect, the phase of the
neutrons changes as they propagate in an area with an electric field. There are
also short descriptions of experiments which verify the A-B and the A-C effect.
These experiments measure interference phenomena, which prove the existence
of the change in the phase of quantum states.
Over long passages the text is very close to the references given at the end of
the article. Italicized parts relate to main ideas originating from primary sources.
The Aharonov-Bohm-Effect
Metal Tube Experiment

In this section I shall discuss the significance of the electromagnetic potentials
in quantum systems. These vector and scalar potentials were used in classical
electrodynamics as a mathematical aid for calculating the physical relevant fields.
However, in classical electrodynamics the equations of motion can be expressed
directly in terms of the fields, namely in the Maxwell equations. In quantum
mechanics this is not possible because the description of physical phenomenas requires the canonical formalism. The Hamilton operator occurs in the Schrodinger
equation, which is the equation of motion in quantum mechanics. In the Hamiltonian the electromagnetic potentials represent the electromagnetic interactions
between the particles described. Nevertheless, the Schrodinger equation of an
electromagnetic interacting particle is gauge invariant under the transformation
ei(x) ,
A A + .
(10.1)
Thus, if the electromagnetic fields vanish, it appears as if the potentials could be

gauged away and the potentials had no independent significance. That in general
this conclusion is incorrect can be shown with several physical systems, which
require a further interpretation of the potentials.
The first example is a charged particle inside a Faraday cage connected to an
external generator which causes the potential to alternate in time29 . This external
potential V (x, t) will cause an additional term in the Hamiltonian. V (x, t) is
constant on the surface of the metallic box in space. Moreover, it is constant in
space inside the box since 2 V (x, t) = 0. Hence, the potential is only a function
of time and the electric field vanishes. The Hamiltonian of the system is given
by H = H0 + V (t) whereas H0 is the Hamiltonian of the free particle confined to
the box. If 0 (x, t) is a solution of the Hamiltonian H0 , the solution for H will
29
See [55]
Berry phase II: The Aharonov-Bohm effect
Figure 10.2: A-B Experiment
Figure 10.1: Metal tube Experiment

be30
~,
iS/
= 0 e
273
S=
V (t)dt =
e(t)dt,
(10.2)
which derives from
i~
=
t
0
S
i~
+ 0
t
t

= H0 + V (t) = H.
The new solution remains the same as 0 up to a phase factor which causes
no change in the physical result.
The phase factor, however, becomes physically relevant, if we look at the
interference phenomena of two wave packets. Therefore, consider an experiment
in which a single incoming coherent electron beam is split up into two beams
passing through two different metal tubes (Figure 10.1). These metal tubes have
to be long in comparison to the width of the wave packets envelope function.
When the beams passed through the tubes they have to be recombined in order
to interfere coherently. The moment at which the wave packets are well inside
the tubes and far from their openings this system is equal to two Faraday cages
of the kind which was mentioned above. A problem occurs if the wave packets
come near to the opening of the tube. At this stage, the applied potential causes
electric fields. Thus, the potential has to be zero when the wave packets enter or
quit the tubes. But if the wave packets are inside the tubes and far from their
openings the potential can be changed in time. This arrangement will ensure that
the wave packets are never in a field. The potential are changed differently for
each tube. Hence, the two wave packets will get different phase factors according
to (10.2). The solution at the interference point is then given by the sum of the
two splitted electron waves
= 01 eiS1 /~ + 02 eiS2 /~ ,
whereas
Z
S1 = e
1 dt,
0
30
See [55]
S2 = e
2 dt.
0
(10.3)
274
Topology in Physics
The interference will depend on the Hphase difference (S1 S2 )/~, which can be
expressed through the integral e/~ dt around a closed curve in space-time
evaluating at the center of the wave packet. Thus, there is a physical effect of
the potentials even though no force is ever actually exerted on the electron 31 .
Up to now the considerations were non-relativistic. In order to get a covariant phase factor, we could try the following generalization. Replace by the
relativistic Vierervector (, A) and the dt by 1c (cdt, dx). This leads to the phase
factor

I
A
S = e
dt dx .
(10.4)
c
This generalization gives another example. Regard a long closely wound
solenoid, which is centered at the origin with the axis in the z direction. It
creates a magnetic field H, with A = H. The field H is confined within the
solenoid. However, the potential A is not, since the magnetic flux 0 through a
surface S containing the origin is constant
Z
I
0 =
Hds =
Adx.
(10.5)
S
Again, let a coherent electron beam, which points in the direction of the solenoid,
be splitted in two, so that each part goes along the opposing side of the solenoid
(Figure 10.2). However they must not touch the solenoid. In the end, they
have to be recombined at a point behind the solenoid. The Hamiltonian of this
configuration is
2
p (e/c)A
H =
(10.6)
2m
The electron beam, which is now not necessarily divided into wave packets, can
be considered as moving along a path in space only (at a constant time t). Hence,
equation (10.4) suggests that the associated phase shift of the electron beam is
I
e
e
S =
A dx = 0 .
(10.7)
c
c
Again, the wave function in this case splits up into two parts = 1 + 2
whereas i represents the beam passing the solenoid on the one or on the opposite
side. If 01 , 02 are the solutions with magnetic field H = 0, we can write
1 = 01 eiS1 /~ ,
2 = 02 eiS2 /~ ,
(10.8)
R
whereas S1 and S2 represent the phase factor (e/c) A dx along the paths of
the first respectively of the second beam. The interference between the two beams
will evidently depend on the phase difference 32 , (S1 S2 ) = S given in equation
(10.7).
31
32
See [55]
See [55]
275
This second experiment is called the Aharonov-Bohm effect (A-B effect) and
in the following sections we will establish, why the phase shift (10.7) is the correct
choice.
In 1985 the group of R.A. Webb et al.33 confirmed the Aharonov - Bohm
effect using minute Au rings. For the first time they measured the prescribed
period h/e of the magnetoresistance oscillation with respect to the magnetic flux.
The difficulty in this experiment was to manage two rivaling effects: Firstly, the
metallic ring has to be small enough in order to avoid randomisation by inelastic
scattering while the electrons traverse the ring. Secondly, the smaller the ring is
the more difficult it becomes to confine the magnetic field to the area bordered
by the ring. If the magnetic field enters the arms of the ring (and in practice
this is the case) the phase shift of the electron becomes path dependent as we
will see in the following section. Since the wire, which the ring is made of, has
an extension, the electrons can move on different paths in the wire. This could
destroy the interference effect if the magnetic field is high enough. Webbs group
used rings with an average diameter of 825 nm at a temperatures less than 1 K.
The oscillations persisted to rather high magnetic fields of around 8 T, which was
a quite surprising result.
Relation to Topology
One wishes to investigate a class of geometric phases which are topological in
nature. The generic example is the Aharonov-Bohm (A-B) phase. Aharonov and
Bohm analysed the quantum mechanics of an electron moving around but not
actually penetrating a magnetic flux line34 . This system has a non trivial topology
if we let the magnetic field be confined within an infinitely long solenoid. In this
section it will be shown, that the A-B phase is a particular case of the geometric
phases35 .
To start the analysis, let us consider a spinless particle of mass m and electric
charge e in free space. The coordinate and momentum operators xi and pi satisfy
the following commutation relations:
[
xi , xj ] = 0,
[
pi , pj ] = 0
[
xi , pj ] = iji 1,
33
(10.9)
See [56]
See [57], p. 26
35
for the definition of the geometric phase see [57] sect. 2.3. or Berry Phase I of this
Proseminar
34
276
Topology in Physics
here is 1 the identity operator acting on the Hilbert space H. Now we can find
the position representation of these operators which fulfills (10.9). For a state
vector H is
hx|
xi |i = xi hx|i = xi (x),
(10.10)

hx|
pi |i =
i i + i (x) (x),
x
(10.11)
whereas x = (x1 , . . . , xM ) are coordinates of the configuration space M and i =

i (x) are scalar functions of x. Since
hx|[
pi , pj ]|i

=

ii + i (x) ij + j (x) ij + j (x) ii + i (x) (x)

= i(x) j i i j = 0
holds for any (x), the following equation results:

i j j i = 0.
(10.12)
We will see later on, that since usually the configuration space M is an Euclidean
space RM , the functions i have no physical significance and thus are neglected.
The operator pi behaves well under coordinate transformations. This implies
that the i0 s transform the same way as the cotangent vectors36 x i do and we
can write37
:= i dxi ,
(10.13)
which is a one-form on M. If we use (10.12) we find that the exterior derivative
of vanishes (One says is a closed one-form):
d =
i j dxi dxj =
i,j
X1
i,j
(i j j i )dxi dxj = 0.
(10.14)
Here, the asymmetry of the wedge product dxi dxj = (dxj dxi ) was used. An
important result of algebraic topology is the Poincar
e lemma, which states: Every
closed form is exact provided that the manifold M on which the differential form
36
A vector
~ p at a point p of the manifold M is called a cotangent or covariant vector if its
xj
components behave under coordinates transformation as follows: (p0 )i = x
0i (p )j . Since for
j
the chainrule x0i = x

0i xj is valuable, this is indeed a cotangent vector. See Appendix
A.2. of [57]
37
The local basis vectors dxi of the cotangent space are defined by dxi := D(xp )i , whereas
D is the usual derivative and (xp )i := h~
p, e~i i the ind coordinate function.
xi
277
is defined is (topologically equivalent to) a star-shaped region of some Euclidean

space RM ,38 . Intuitively, star-shaped means there are no holes or cuts in M. An
exact one-form is a one-form which can be written as
= df = i f (x) dxi ,
| {z }
(10.15)
whereas f = f (x) is a function on M. Particularly if M = RM , the equation

(10.15) is valid. Moreover the i0 s may be gauged away in the following way
0 = eif (x) ,
(10.16)
since then

hx|
pi | 0 i =
ii + i (x) 0 (x)

= i (x) i f (x) ii 0
= ii 0 .
However, if M is not equivalent to a star-shaped region and in fact the infinitely
long solenoid in three dimensional space is a physically relevant example of such
an M, then Poincar
es lemma is violated and the i0 s must not be neglected in
(10.11). In order to understand what the influence of the functions i on (x) is,
an infinitesimal spacial displacement of the state vector has to be considered,
for example in the direction = 1 , . . . , M . The Operators pi generate such a
displacement in the i-direction. Let us examine it for in the first-order of :
i
hx|i hx|ei pi |i = (1 + i i + ii i )(x)

= [(x) + i i (x)](1 + ii i )
i
= (x + )ei i .
(10.17)
In order to see, why it is a good thing to investigate infinitesimal displacements

of we have to keep in mind that the system we describe consists essentially of
free fermions. The wave function describes an electron at the Fermi surface,
where the dispersion relation can be linearised within a good approximation (Figure 10.3):
p = pF + ~k,
H=
pF ~k
p2
+ O(k 2 ) EF vF ~k.
2m
m
The system is conservative (i.e. the Hamiltonian is time independent), hence the
time evolution of is
|(t)i = eiHt |(0)i ei~kvF t |(0)i.
38
See [57], p. 23
278
Topology in Physics
Figure 10.3: Dispersion relation of free electrons

The Fermi-velocity vF times an infinitesimal t is equivalent to the i in equation
(10.17).
Because the pj commute among themselves (10.17) can be generalised to large
displacements. For example let us displace along a smooth curve
C : [0, T ] M .
Under C the state vector transforms as follows:
hx|i hx|ei
R
C
dxi pi
|i = (x + x)ei
R
C
(10.18)
Here is x = C(T ) C(0) and := i dxi is the one-form, which appears in the
position representation of the momentum operator. We assumed, that the initial
and the end point of the curve C lies in a single patch of the manifold M , since
otherwise the quantity x would not be well defined. If we take C to be closed
(C(T ) = C(0)), a path-dependent phase-factor will remain:
H
hx|i hx|ei C dx pi |i = (x)ei C .

(10.19)
H
This phase-factor exp(i C ) will vanish, if M is a star-shaped manifold and
a closed one-form. One can show this easily by using Pointcar
e ( = df ) and by
the fact that an integral over the exterior derivative df of a function f is equal to
the difference f (C(T )) f (C(0)), which for a closed path C is always equal to
zero. Alternatively using the Stokes theorem:
I
Z
= d = 0,
(10.20)
C
whereas S is a surface bounded by the curve C(:= S ). Stokes theorem is valid on

any manifold and this gives the possibility to generalize the phase-factor derived
above to non-trivial manifolds. Let us for example examine a quantum system
with an infinitely long solenoid in three dimensional space. Since this system has
279
Figure 10.4: Closed path encircling a disk

an translation symmetry along the axis of the solenoid, it is enough to look at a
configuration space M consisting of a two-dimensional plane R2 with a disk D of
radius removed. Consider now a path C encircling the disk D and define S as
the surface bounded by the loop C and the boundary D of the disk D (Figure
10.4). Stokes leads to the following expression:
I
I
Z
=
d = 0,
(10.21)
C
S0
since is a closed one-form. From this equation one can see, that the phase does
not depend on the particular closed curve used for the displacement, as long as it
encircles D once 39 . The phase factor depends only on how many times the closed
curve C winds around D and it is independent of continuous deformations of the
curve C. One calls such a phase, since it is totally defined by the topological
property of the curve, a topological phase.
Let us now treat a system of an electron traversing a closed loop C in the
configuration space R3 with an external magnetic field
B(el) = A(el) ,
(10.22)
whereas A(el) = (A1 , A2 , A3 ) is the electromagnetic vector potential. It can be

expressed in terms of differential forms according to A(el) = Aa dxa , and one gets

(el)
(el)
(el)
F
:= dA
=
A
dxb dxc
xb c

a
1
(el)
dxb dxc .
= abc B
(10.23)
2
The minus sign in the last expression comes from the Minkowski metric with
ab = ab whereas a, b = 1, 2, 3 are the space components. Then, the contravariant space coordinates of the electromagnetic space-time vector potential can be
shifted to covariant coordinates via: Aa := ab Ab = Ab .
39
See [57], p. 25
280
Topology in Physics
To get the interaction between the electron and the magnetic field into the
equations of motion, one has to replace the momentum operator p
by its covariant
generalisation
e
p
:= p
A(el) ,
(10.24)
c
which follows, for example, from the derivation of the Hamiltonian of a charged
particle in electromagnetic fields via Legendre-Transformation of the Lagrangian
L = 21 mq 2 e(q, t) + ec q A(q, t) or, alternatively, from field theory by assuming
a local U (1) gauge invariance. This means that the Lagrangian is required to
be invariant under (x) 7 ei(x) (x), whereas (x) is an arbitrary function
of space and time. The terms involving derivatives of (e.g. ) are not
invariant under such transformations. To remedy this one needs to introduce the
electromagnetic gauge potential a (x), with time and space components (a0 , a),40 .
The gauge covariant derivative of can then be defined by D := ia , if
we postulate that a transforms under the gauge transformation to a 7 a + .
Calculations show that D transforms the same way as does: D 7 ei D . The
space coordinates of the covariant derivative can be identified with the covariant
momentum
e (el)
i = p
i Ai = iDi = ii + ai .
(10.25)
c
We now incorporated the effect of the magnetic field in the definition of the
covariant momentum operator
and want to look whether the movement of
an electron along a path C may be achieved by using this momentum as the
generator of the accordant displacements. The momentum
in the coordinate
representation is

e (el)
hx|
a |i = i a Aa (x) +
a (x) (x),
(10.26)
x
c
on R3 . Since we are considering
whereas
a are components of a closed one-form
the configuration space R3 , this one-form must be exact. However, from the field
(el)
theory point of view we saw that Ai is defined up to addition of a gauge term
i (x), which is the i-th component of an exact one-form. Hence, we can absorb
in the definition of A(el) . This leads to the representation

e (el)
(10.27)
hx|
a |i = i a Aa (x) (x).
x
c
Displacing the electron along C gives
hx|i hx|ei
H
C
dxa
a
|i = (x)ei c
e
(el)
H
C
A(el)
(10.28)
We receive an electromagnetic phase ei c C A , which is in general not independent of the path C. If the magnetic field B(el) vanishes, then A(el) becomes a
40
See [58], p. 33
281
closed one-form and the electromagnetic phase is going to be a topological phase.

Hence, in the case of R3 the closedness of A(el) implies its exactness and the phase
angle vanishes. However we can view cases, in which the configuration space M
is just a portion of R3 in which the magnetic field vanishes. Hence, in general M
could have a non-trivial topology. In view of this observation, a comparison of
(10.11) and (10.27) suggests the identification of the closed one-form of (10.13)
with the electromagnetic one-form A(el) 41
e
= A(el) .
c
(10.29)
Let us now analyse the A-B effect. In the A-B system the magnetic field is
confined to an infinitely long solenoid, the electron is circulating around and not
allowed to pass through it. The magnetic field vanishes outside the solenoid and
the configuration space of the electron can be compared with the system described
directly after (10.20). Hence, the topological (A-B) phase associated with a
closed curve C is determined by the winding number of C and the fundamental
topological phase:
R
ie
(el)
e c DF ,
where we used (10.20). This phase can be computed by using the information
about the magnetic field, equation (10.23), and Stokes theorem
ie
e c
H
D
A(el)
ie
(el)
ie
(el)
= e c D dA = e c D F
R
ie
ie
(el)
= e c D B dS = e c B ,
(10.30)
whereas B is the magnetic flux through D. The phase derived in (10.30) is the
Aharonov-Bohm (A-B) phase and exactly the same result as we got in (10.7). Its
significance is that although the electron is classically confined to a region where
the magnetic field vanishes, it is influenced by the field quantum mechanically 42 .
Relation to the Berrys geometric phase
The A-B phase can be viewed as a special case of the Berrys geometric phase
n (t), which is defined as follows43
Z
R(t)
n (t) =
An ,
ihn; R(t)|d|n; R(t)i =

R(0)
(10.31)
whereas |n; R(t)i is the Eigenstate to the n-th Eigenvalue of the Hamiltonian h(t).
This Eigenstate depends on the parameter R(t), which is a point in the parameter
space describing the environment of the quantal system under consideration. We
41
See [57], p. 27
See [57], p. 28
43
See [57], p. 16
42
282
Topology in Physics
Figure 10.5: A-B system with the electron confined to a box.

want to show that under an adequate description of |n; R(t)i the Mead-Berry
connection one-form An is proportional to the electromagnetic vector potential
A(el) of the A-B system. To see this, let the electron be confined to a box placed
at a point R of space, with |R| , whereas is the radius of the solenoid
centered in the origin (Figure 10.5). The Hamiltonian of the system depends on
the space coordinate and the covariant momentum

h = h

, x R .
(10.32)
Here we have switched to a coordinate frame centered inside the box. The location
R of the box plays in this description the role of the parameter which parameterize
the Hamiltonian h = h(R). We assume that there is one single Hilbert space H
which is complete for any R of the parameter space. Then the Eigenstates |n; Ri
are defined by:

h
, x R |n; Ri = En (R)|n, Ri.
(10.33)
0 = h
0 (
If we consider n to be the Eigenvectors of the free Hamiltonian h
p, x R)
it can be readily shown that
ie
hx|n; Ri = e ~c
Rx
R
A(el)
n (x R),
(10.34)
283
since

, x R)|n; Ri =
h(
e (el)
A
c
2
|n; Ri
2m
~2 2
1 ie~
=
+
(A(el) ) + 2iA(el) ~
2m
2m c

e2 (el) 2 ie R x A(el)
n (x R)
+ 2A
e ~c R
c
R
ie x (el)
= = En e ~c R A n (x R)
= En |n; Ri,

where the term A(el) was set to zero using the Coulomb-gauge. From this equation one can also see, that the introduction of the magnetic field does not affect
the energy eigenvalues En . Furthermore the integration in (10.34) is independent
of the choice of the path along which the integral is evaluated as long as the path
does not intersect or encircle the solenoid, since the magnetic field outside the
tube vanishes.
Having an expression for |n; Ri, the Mead-Berry connection one-form can be
computed
An := ihn; R|d|n; Ri = ihn; R| i |n; RidRi

R

Z
ie
3
(el)
= i dx n (x R) A (R)n (x R) + dn (x R)
c
e (el)
=
A
(10.35)
c
The third equality holds, since n is assumed to be normalized (n can be choosen
to be real since the Hamiltonian is invariant under time-reversal). Equation
(10.35) together with the definition of the Berrys phase (10.31) gives us the
following identification:
H
ie
(el)
ein = e ~c C A .
(10.36)
The closed path C in M represents the curve along which the box, containing
the electron, is transported. The right hand side of (10.36) is independent of the
energy quantum number n. Hence (10.36) is valid for any superposition of the
energy eigenvectors and consequently for any wave packet. This point gives a
subsequent legitimation to the idea to start the analysis with an electron which
is confined to a finite box R.
This derivation shows, that the A-B phase is a particular example of Berrys
geometric phase. However, if we consider a regime in which the magnetic field
vanishes, the A-B phase is topological in nature unlike the usual geometric phase.
If we go to a regime where the magnetic field is nonzero, the A-B phase becomes
284
Topology in Physics
Figure 10.6: a: A-B System, b: A-C System

truly geometric, which means though independent of the details of the dynamics
it depends on the geometry of the closed path C.
Furthermore, it should be noted that in this specific example of a Berrys
phase, the adiabatic approximation is valid. The adiabatic approximation demands that the physical system which depends on the time evolution of its environment does not jump from the initial n-th eigenstate |n; Rihn; R| at t = 0
into any other eigenstate |m, R(t)ihm, R(t)|, m 6= n at any later time t. This
compare to the condition that the adiabatic approximation is valid if and only
if hm; R(t)| dtd |n; R(t)i can be neglected44 . Using (10.34) it can be shown that in
our case this expression vanishes for m 6= n. Hence, the adiabatic approximation
is satisfied for our system regardless of the speed with which the electron moves
around the solenoid.
The Aharonov-Casher-Effect
In their paper45 Aharonov and Casher thought of an experiment in which neutral

particles with a magnetic moment will exhibit the Aharonov-Bohm effect. The
general idea is to treat the solenoid in the A-B effect as a straight line of magnetic
moments and then in a second step to exchange the role of this moments and the
electrons (Figure 10.6). It is the subject of this section, to show how this can be
achieved.
The Dual of the A-B Effect
We start with the Lagrangian of an A-B system, which is an electron in a magnetic
vector potential (e is positive)
1
e~
L = mv 2 A
~v .
2
c
44
45
See [57], p. 14
See [59]
(10.37)
285
Evaluating the path integral of this system in the semi-classical approximation,

this Lagrangian leads to a phase factor in the propagator (and hence in the
wavefunction) of the form
I
I
i
ie
ie
~
~ d~x),
exp( S ) = exp(
A ~v dt) = exp(
A
(10.38)
~
c~
c~
Rt
whereas S = S [
x] := tab L(x, x,
t)dt is the action over the classical path46 x (in
the expression (10.38) the kinetic term was disregarded, since it does not produce
characteristic phenomena for the A-B effect).
Let us analyse a more detailed Lagrangian which describes not only the dynamics of the electric charged particle at point ~r, but also the dynamics of the
~ (in the plane). One may suggest that L is
solenoid located at the point R
1
1
e~
~ ~v ,
L = mv 2 + M V 2 A(~
r R)
2
2
c
(10.39)
since the solenoid is electrically neutral. However, evaluating the Euler-Lagrange

equations, the Lagrangian (10.39) leads to a non-vanishing force on the electron

d L
L
e X Aj dri Aj dRi Ai
= mv j
vj
dt rj
rj
c i
ri dt
Ri dt
rj
= mv j e/c vi (i Aj j Ai ) +e/c Vi i Aj = 0
{z
}
|
(v(A))j
The second term in the second line is the ordinary Lorentz force and vanishes
outside the solenoid because H = 0. The third therm generates a non-vanishing
force which is proportional to the velocity of the solenoid. To avoid this force,
Aharonov and Casher proposed an extra term in the Lagrangian (10.39):
1
1
e~
~ (~v V~ ).
r R)
L = mv 2 + M V 2 A(~
2
2
c
(10.40)
The extra term in (10.40) leads to an extra Term e/c Vi j Ai in the equations of
motion:
e
~ = M V,
~
m~v = (~v V~ ) B
(10.41)
c
~ = A(~
~ r R).
~ This new Lagrangian is gauge invariant
with the magnetic field B
~A
~ + . Moreover, it is translation invariant and
under the transformation A
hence the momentum is conserved.
In the following it will be proved that (10.40) follows from electrodynamics and
correctly describes the A-B system. As mentioned above a solenoid can be viewed
as a line of magnetic moments which interacts with the electron. Therefore, we
46
A classical path is a path for which the Variation S of the action S vanishes.
286
Topology in Physics
first want to look at a single magnetic moment ~ moving in an external static

electric field of an electron. This magnetic moment is generated by a current
distribution:
Z
~j = M~,
M~ = ~.
(10.42)
The velocity-Vierervector of the moment is U = (c, V~ ) and the generalized
current J = (c, ~j). In the inertial system, in which the velocity of the moment
vanishes, we have U0 = (c, ~0) and J0 = (0, ~j 0 ). The Minkovski scalar product of
two Vierervectors is invariant, hence
c2 V~ ~j = J U = J 0 U0 = 0
and we get
c2 = V~ ~j.
(10.43)
Equation (10.43) states that a moving magnetic moment generates a charge density . Thus, the Lagrangian of the magnetic moment is
Z
R
1
2
L = M V A0 = 12 M V 2 c12 A0 V~ M~
2
~ ~,
= 12 M V 2 c12 V~ E
(10.44)
~ = A0 is the electric field at the position of the magnetic moment.
whereas E
The last identity holds, because if we use Stokes, we get
Z
Z
~
0 = (A0 M ) = (A0 ) M~ + A0 ( M~)
~ The electric field
and the magnetisation is basically given by M~(~x) = ~(~x R).
~ is nothing else but the Coulomb field of the electron, therefore
E
~
~ ~ = e (R ~r) ~ = eA(~
~ r R),
~
E
4 |R r|3
(10.45)
~ is the vector potential of a magnetic dipole47 whose source is the

whereas A
~ Now we are ready to add the Lagrangian of the
magnetic moment ~ at R.
electron (10.37) and recognize immediately that this combination is equal to
equation (10.40).
Next we analyse a system of more than one electric charged particle interacting
with more than one magnetic moment. In the Lagrangian of this system we
include the interaction between charged particles and magnetic moments but
47
See also J. D. Jackson, Klassische Elektrodynamik, sect. 5.6.
287
interactions among charged particles or magnetic moments itself are excluded

because we are not interested in their dynamics. The Lagrangian is then:
L =
1X
1X
me ve2 +
Mm Vm2
2 e
2 m
X
~ re R
~ m)
+
(~ve V~m ) eA(~
(10.46)
e,m
Encouraged by this derivation of (10.40) for a many particle system, we may

~ the system
dare an interpretation: Because of the dependence of A on (~r R),
~ m are reversed 48 . This
possesses a kind of duality when the roles of ~re and R
statement leads in particular to the fact, that though a magnetic moment which
moves in the electric field of a straight homogeneously charged line feels no force,
it undergoes an A-B effect with the following phase shift
I
e
~ dR.
~
S =
A(~r R)
(10.47)
c
This phenomena is the Aharonov-Casher Effect (A-C Effect). We assume the
magnetic moment to be polarized along the z-axis and calculate the electrostatic
field of the charge line with Gauss law. Expression (10.45) for the vector potential
A leads to49
4
S =
,
(10.48)
~c
depending on whether the neutron has spin up or down. Thus the phase shift is
proportional to the lineal charge density and the magnetic moment .
The group of A. Cimmino et al. verified in 1989 the existence of the A-C
Effect50 . They used a neutron interferometer containing an electrode system
which was prism-shaped to amplify the expected phase shift. Since the A-C
phase is a topological phase, this generates the same effect as if they had used
a line charge. An experimental problem was the stability of the interferometer
(the measurements took several month). A polarized neutron beam would have
been disturbed so strong that the sensitivity of the experiment would have been
destroyed. Therefore, an unpolarised neutron beam was used. In order to get
a interference effect with an unpolarised beam, they introduced an additional
and spin-independent phase shift. This phase shift was maintained using the
gravitational potential. The interferometer was tilted about the incident beam
direction. Neutrons traveling along two opposing paths around the prism-shaped
center electrode experience different gravitational potentials. The result is a spinindependent phase shift. This experiment shows, that not only electromagnetic
48
See [59]
See [60]
50
See [60]
49
288
Topology in Physics
potentials can cause phase shifts but also for instance the gravitational potential.
The ratio of the experimental result compared to the theoretically predicted value
is
(AC )exp
= 1.46 0.35.
(AC )theor
Derivation of the A-C-Effect via Dirac Equation
In order to strengthen the a bit dirty foundation of (10.40), in which we used a
relativistic argument to explain the generation of a charge density of a moving
neutron, (10.40) is to be rederived from the Dirac equation. The Dirac Lagrangian
of a purely magnetic particle without electric charge though in an external electric
field is (we set ~ = c = 1)

1
L = i m F ,
2
(10.49)
where = 0 , m the mass of the particle, the magnetic moment, F the

electromagnetic field tensor and = 12 i[ , ],
0
Ex
Ey
Ez
Ex
0
Bz By
.
F =
Ey Bz
0
Bx
Ez By Bx
0
In equation (10.49) we used the following representation

1 0
0 k
0
k
==
,
=
,
0 1
k 0
(10.50)
with k {1, 2, 3} the space components and k = k .

Afterwards, we analyse the nonrelativistic limit of the Euler-Lagrange-Equations
of the Lagrangian given in (10.49). Therefore, we write the field , which
de
scribes the four-component vector-field of the particle, as a vector =
with u and v a two-component vector-field per each representing the particle

with positive energy and the antiparticle with negative energy. With the EulerLagrange equations
L
L
= 0,
( )
we get

1
0 i m iF [ , ] = 0
(10.51)
4
Since we consider a neutral particle, we can set the electric charge e = 0. The external magnetic field though we will take along to check whether the Lagrangian
289
(10.49) produces the correct interaction term between the magnetic moment
and the magnetic field B . With this assumptions equation (10.51) can be simplified to

k
~
~ ~
i0 ik + 0 m + 0 i(E
~ ) + 0 B
= 0.
(10.52)
This leads to the following equation of motion which will be the starting point
for subsequent non-relativistic approximations:

~
~
~
i0
= i~
+m
+ i~ E
+ B ~
(10.53)
In the case of the non-relativistic limit the rest energy of our particle is much
larger than the other terms and the field can be separated in a part carrying
the time evolution and a part which varies slowly in time:

imt
=e
.
(10.54)
This Ansatz will generate a solution with positive energy. Inserting it into (10.53)
leads to an equation for the slowly varying

0
~
~
~ ~
i~
= i~
+ i~ E
+ B
2m
.
t
(10.55)
~
In the second equation of (10.55) the terms and B ~ can be neglected

compared to 2m and we get approximately51
=

1
~ i~ E
~
i~
2m
(10.56)
~ can be identified with the momentum p~ m~v one can see, that in our
Since i
approximation is of the order v/c smaller than .
Inserting (10.56) into the
first equation of (10.55) we get

1
~
~ iE~
~ + i ~ E
~
~ iE~
~
~ (i)
~ (i)
i~
i0 =
2m
~ ~ )
+(B

1
2
2 ~2
~ + (B
~ ~ ),
=
(10.57)
p~ + E 2~ (~p E)
2m
whereas the following identity was used: (~ ~a)(~ ~b) = ~a ~b + i~ (~a ~b). The
second term in (10.57) describes the interaction of the magnetic moment with
51
See [61], p. 129
290
Topology in Physics
~ This term occurs also in the Pauli Equation for a

the external magnetic field B.
non-relativistic particle with Spin 1/2. Hence, the Lagrangian (10.49) seems to
~ ~ )2 = 2 E
~ 2 we have the
describe the system correctly. Finally, by using (E
nonrelativistic Hamiltonian (the magnetic field is now neglected)

2
1
2 ~2
~
HN R =
p~ (E ~ ) E .
(10.58)
2m
If we identify ~ = ~ with the magnetic moment (and this identification makes
~ the first term in (10.58) is precisely
sense, since ~ is proportional to the Spin S),
the Hamiltonian which corresponds to equation (10.44)52 . The second term in
~ is small compared to m~v . There(10.58) can be neglected if we assume that E
fore, the derivation which started from the Dirac Equation (10.51) leads to the
Lagrangian (10.46) and we are done.
Summary
The investigation on the Aharonov-Bohm effect, then, resulted that there are
several experimentally realisable quantum systems in which electrons or neutrons undertake interference effects even if they dont feel a force. For example
an electron, which moves around a solenoid feels no magnetic field. However, the
electron picks up a phase shift. This phase shift is proportional to the magnetic
flux through a surface which is bounded by the closed path of the electron. A
possible explanation for this phenomenon is that the electromagnetic potential,
which does not vanish outside the solenoid, evokes the phase shift. This means
that in quantum mechanics the potentials become physically relevant whereas
in classical mechanics they were not. The A-B phase is purely topological in
nature if the magnetic field is confined to an area which is not penetrated by the
electrons. This means that the phase shift depends only on how many times the
electron winds around the solenoid and it does not depend on the actual shape
of the electrons path. However, if the magnetic field extends into the area of the
electron, the A-B phase becomes a geometric phase and path-dependent. In the
experiments it is quite difficult to suppress this path dependence which destroys
the interference effects. It could be done further investigations on this topic. For
example, the group of Yang Ji et al.53 made an experiment with electrons which
are forced to move on a single path using the quantum hall effect. The result
was a very pure interference pattern. In the section about the relation between
A-B effect and Berrys phase the fact that the electrons in the A-B effect evolve
adiabatically showed that they can be well described by Berrys formalism. The
introduction of the Aharonov-Casher effect illustrated that the A-C effect is a
dual of the A-B effect since the roles of electrons and magnetic moments can be
52
53
See [59]
See [62]
291
reversed. The idea is to analyse the Lagrangian of the A-B system by including
the dynamics of the electron as well as the dynamics of the solenoid. After some
calculations we saw that this Lagrangian is invariant under the exchange of the
coordinates of electrons and magnetic moments. Hence, we can describe a neutron moving around an electric charged line with the A-B-Lagrangian. Though
electrically neutral, the magnetic moment of the neutron causes a phase shift.
The first derivation uses a classical Lagrangian but a relativistic argument to
explain the interaction of the magnetic moment with the electric field. This
derivation is inconsistent and has to be redone using a relativistic Lagrangian.
A neutron is a fermion and, therefore, it has to be described by the Dirac equation. With interest in the particle part of the Dirac equation, we looked at the
non-relativistic limit of positive energy and derived the same Lagrangian as in
the first place.
Finally, I would like to thank Dr. Sebastian Pilgram for many inspiring discussions and his competent support. Lars Steffens contribution on various irking Latex and computer problems were very helpful and many thanks go to my
younger brother Roman for patiently brushing up on my English.
292
Topology in Physics
11 Berry phase III: Berry phase

and Bloch states
Juan Carlos Andresen
The base for understanding the dynamics of electrons in a periodic crystal, is given by the Bloch theory for periodic systems.
In this text the dynamics of electrons in a crystal under perturbations is derived (up to the first order gradient correction). The
semi-classical equations are derived by describing the system with
the Bloch theory for perturbated crystals and then passing to the
semi-classical limit under some simple assumptions. In this equations new terms arise. For the specific case of electromagnetic
perturbations on crystals this new terms can be interpreted as
the anomalous velocity, which gives rise to the anomalous Hall
effect, and the magnetization energy. Applying the quantization rule of Einstein, Brillouin and Keller in the obtained semiclassical equations, new kins of Berry phases emerge.
Introduction
In this text, the dynamics of electrons in crystals under slowly varying perturbations are presented. In the development of the dynamics the Bloch theory for
periodic systems will play an important role, as the electron wave packet will be
expanded in a basis which is spanned by the Bloch wave eigenstates.
Figure 11.1: Perturbation of the periodicity of the crystal potential.

293
294
Topology in Physics
It will be then presented how, through some assumptions (spread of the wave
packet centered at xc and the distribution of the expectation value of the wave
vector q) the motion of the electron in a perturbated crystal can be described in
the semi-classical limit. For this purpose, a Lagrangian given by a time-dependent
variational principle is computed. Taking into account the assumptions that made
possible to see the electron dynamics in the semi-classical limit, the Lagrangian
function can be rearranged in a simple form. This rearranged Lagrangian can
be handled as in classical mechanics to derive the semi-classical equations of
motion of the electron in the perturbated crystal. By computing these equations
of motion new terms arises in a natural form. These new terms can be seen as a
correction to the known semi-classical equations of motion for a Bloch electron.
These equations can be re-quantized and through this re-quantization a
Berry phase emerges as a correction to the quantization condition.
Structure
These text is structured as following:
1. A brief review of the Bloch theory, the semi-classical dynamics and quantization theory.
2. A derivation of the semi-classical equations of electron for different Hamiltonians (a general one and one for an electromagnetic field applied on the
crystal).
3. A discussion for the extra terms that arise from the derivation of the semiclassical equations.
4. The quantization of the semi-classical equations and the Berry phase acquired by a wave packet (upon going round a closed orbit once) under a
uniform magnetic field and under a uniform electromagnetic field.
Review
Bloch Theory
In a perfect crystal, a free electron sees a periodic potential acting on it, i.e.
U(r)=U(r+R), where R is a Bravais lattice vector. With this assumption the
Schrodinger equation for a single electron is

~2 2
+ U (r) = E.
(11.1)
H =
2m
The electrons that obey this Schrodinger equation are called the Bloch electrons.
Berry phase III: Berry phase and Bloch states
295
Bloch Theorem The solution of the given Schrodinger equation with periodic
potential can be written as a modulated plane wave nk (r) = eikr unk (r) with
unk (r) = unk (r + R) for all R in the Bravais lattice.[63]
The strong periodicity postulation in the potential has other consequences that
follow immediately from the Bloch states. The Bloch waves, for which the wave
vector differ only by a reciprocal lattice are equal, this means
k+K (r) = k (r)
(11.2)
with K being a reciprocal lattice vector. This result combined with the eigenvalue
equation
Hk (r) = (k) k (r)
(11.3)
gives the periodicity of the energy as a function of the wave vector k, i.e.
(k) = (k + K) .
(11.4)
Semi-classical equations
The behavior of a free electron gas can be calculated using the classical equations,
provided that there is no need to localize an electron in a scale comparable to the
inter-electronic distance[63]. For a free electron with momentum ~k and energy
2 2
E (k) = ~2mk in a electromagnetic field, the equations of motion are (between
collisions)
~k
1 E (k)
v (k) =
=
~ k
m
1
~k = e E + v H
c
(11.5)
For the Bloch electrons the equations are very similar. In this case the explicit
function of the energy is not given, but the periodicity, i.e. E (k) = E (k + K).
It is assumed that the band index n is a constant of the motion and that there
are no inter-band transitions. The semi-classical equations of motion of the Bloch
electrons in a electromagnetic field are[63]:
1 En (k)
r = vn (k) =
~ k

1
~k = e E + vn H
c
(11.6)
296
Topology in Physics
Note: The momentum ~k is not the momentum of the Bloch electron, it is

the crystal momentum, i.e. only external forces apply on the electrons, not the
periodic ones.
Bohr - Sommerfeld quantization rule
Semi-classical Limit for Time-Independent Problems In stationary quantum mechanical states the wave function is separable in the form of (q, k) =
Et
(q) ei ~ , where E is the energy of the stationary state. With the ansatz
S(q)
S(q)
Et
(q) = ei ~ , the wave function is (q, t) = ei ~ i ~ substituting this expression to the Schroedinger equation gives
1
i~ 2
(S S) + V (q) E
S=0
2m
2m
(11.7)
Taking it to the classical limit ~ 0 gives the time independent Hamilton Jacobi equation
where S0 (q) =
tory.
Rq
q0
1
(S0 )2 + V (q) = E
2m
(11.8)
p dq with q0 being some initial point on a given trajec-
Wentzel-Krammers-Brillouin approximation Using the Wentzel-KrammersBrillouin (WKB) approximation, i.e. expansion

of the solution
1 P
for the above
n
differential equation in the form y (x) = exp ~ n=0 ~ Sn (x) , and solving the
Schroedinger equation with the ansatz
S(x)
(x) = ei ~
(11.9)
where S (x) = S0 + ~S1 + ~2 S2 + ... and solving (as in perturbation theory)

recursively form lower correction order to higher ones, leads us to the solution of
S0 and S1
Z x
Z xp
0
0
S0 (x) =
2m (E V (x0 ))dx0
(11.10)
p (x ) dx =
x0
x0
and
S1 (x) = ln
p (x) + ln c.
The general approximation to first order of ~ for the wave function is

Z
Z
i x
B
i x
A
0
0
0
0
p (x ) dx + exp
p (x ) dx .
(x) = exp +
p
~ x0
p
~ x0
(11.11)
(11.12)
297
For the classical allowed regions the solution must be real. The two exponentials
are chosen to give a real expression
Z x

A
1
0
0
(x) = sin
p (x ) dx +
(11.13)
p
~ x0
being a phase factor. Requiring that the solution has to join smoothly the form
to the left of a classical returning point x0 , the phase must be 4 .
Bohr-Sommerfeld quantization The quantization rule of Bohr-Sommerfeld

I
1
0
0
p (x ) dx = 2~ n +
(11.14)
2
comes assuming the situation where the motion is bounded between the two
classical turning points a x b and requiring to be single valued, therefore

Z x
A
1
0
0
(x) = sin
p (x ) dx +
(11.15)
p
~ a
4
must be the same as
B
(x) = sin
p
Z b

1
0
0
p (x ) dx +
~ x
4
and writing the first wave function as

Z b

Z
1
1 a
A
0
0
0
0
p (x ) dx + +
p (x ) dx
(x) = sin
p
~ x
4 ~ b
2
The generalization for many degrees of freedom is[64]
I

k
Ik =
p dq = 2~ nk +
4
Ck
(11.16)
(11.17)
(11.18)
where k is the number of caustics traversed.These is the EBK quantization

procedure (Einstein, Brillouin and Keller).
Semi-classical Dynamics and Berry Curvature
In this section the semi-classical equations of motion for an electron in a perturbed

crystal will be derived. The dynamics for the wave packet in a crystal can be
derived from the Schrodinger equation, we will not follow this line as it is not
convenient. The method used here will be a time -dependent variational principle
using the Lagrangian L given by[65]
L = h|i
H|i.
dt
(11.19)
298
Topology in Physics
It can be seen, that by varying this Lagrangian independently in respect to |i

and h| we get the Schrodinger equation.
Z
S = dtL (, ) i |i = H|i
(11.20)
t
using the convention ~ = 1.
The first step to get the motion equations is to compute these Lagrangian with
the Hamiltonian given by the system. The second is to consider the wave packet
expanded in a basis of Bloch eigenstates from a given band, i.e.
Z
|i = d3 ka (k, t) |k (xc , t)i.
(11.21)
Having this, we assume that the spread of the wave packet in the wave vector is
small compared with the dimensions of the Brillouin zone and we assume that
the spread of the wave packet centered at xc is small compared with the length
scale of the perturbations.
Under the above assumptions the Lagrangian can be written as a function
of the position xc , the wave vector kc and their time derivatives and t. That
means, that the system can be seen as a semi-classical one, and the motion of
the electrons will be described by the semi-classical equations derived from the
Lagrangian.
Mean position and mean wave vector
As mentioned above, the Lagrangian will become a function of the mean position,
the mean wave vector and their time derivatives and of t, so that it is convenient
to know the expression of these variables, (xc and kc ). For the mean wave vector
we have this expression, with the narrow distribution |a (k, t) |2 compared with
the Brillouin zone,
Z
kc = d3 kk|a (k, t) |2
(11.22)
and for the mean position we have this other expression
Z Z Z
xc = h|
x|i =
d3 x d3 k d3 k 0 ak k x ak0 k0
Z Z Z
1 ik0 x
e
=
d3 x d3 k d3 k 0 ak uk eikx ak0 uk0
i k0

Z Z Z
3
3
3 0
ik0 x
=i
d x d k d k ak k ak0 0 k0 + e
u
k
k0
(11.23)
299
with |uk i = |u (k.x, t)i = eikx |k (x, t)i being the periodic part of the Bloch
wave, normalizing the amplitude a (k, t) as
Z
d3 k|a (k, t) |2 = h|i = 1
and writing the amplitude as
a (k, t) = |a (k, t) | exp {i (q, t)}
integrate by parts
Z
xc =
d k |a|
u
+ hu|i i
k
k
(11.24)
finally being |a|2 a narrow packet xc is

xc =
c
u
+ hu|i
i.
kc
qc
(11.25)
The Lagrangian with a general Hamiltonian

Now having the expressions for xc and kc the Lagrangian can be calculated for a
The general Hamiltonian can be approximated up to first
given Hamiltonian H.
order perturbations
H
c + (
H
x xc ) gradxc i (xc , t)
(11.26)
c H (
where the H
x, p
; {i (xc , t)}) has the periodicity of the perfect crystal
and i (x, t) are the modulation functions characterizing the perturbations[66],
for example under an electromagnetic field, the Hamiltonian will have the form
= H0 (
H
x + 1 , p
+ 2 ) + 3
= H0 (k + A (
H
x, t)) e (
x)
(11.27)
with 1 (
x, t) = 0. The energy spectrum of the periodic part are the Bloch bands
and the wave functions are the Bloch states
c (xc , t) |k (xc , t)i = c (xc , k, t) |k (xc , t)i.
H
(11.28)
The band index is omitted as only one single band will be treated, i.e. the
transition probabilities to other bands are negligible.
300
Topology in Physics
To compute the Lagrangian L (, ) = h|i dtd H|i

we first take the first
term and compute it
Z
Z
d
d
d
3
| i = d ka (k, t) k (xc , t) + d3 k {a (k, t)} k (xc , t)
dt
dt
dt
Z
Z
d
d
= d3 ka (k, t) (uk (xc , t)) eikx + d3 ka (k, t) (uk (xc , t)) eikx +
dt
dt
Z
d
d3 ka (k, t) (i) {} k (xc , t)
dt
d
d
dc
= h|i i = hu|i ui k c xc +
=
dt
dt
dt
u
u
dc
u
k c xc + q c hu|i
i + x c hu|i
i + hu|i i
dt
kc
xc
t
(11.29)
The second part of the Lagrangian can be computed up to first order with
the linearized Hamiltonian given above.
c |i + h|H|i
= h|H|i
h|H
(11.30)
is = c + where c is the expecThe expectation value of the Hamiltonian H

c and after some calculation
tation value of the unperturbed Hamiltonian H
can be brought to this expression

u
u
c | i |k=kc
= = h
| c H
(11.31)
xc
k
Calculation of the perturbation energy For the calculation of the
expression of the perturbation of the Hamiltonian is needed, this is
(
)
H
=
H
(
x xc )
+
(
x xc )
2
xc xc
we begin with the matrix elements of the position operator

un0
i (k k0 )
hnk |
x|n0 k0 i = i nn0 + hun |i
k
k
(11.32)
and from the derivation of the eigenvalue equation in relation to the position xc
and the completeness relation
XZ
d3 k|nk ihnk | = 1
n
301
the gradient of the Hamiltonian-operator can be written as follows

c
H
c | n0 k0 i + n0 |n0 k0 i
|n0 k0 i = n0 H
xc
xc
xc

c
H
c + hnk | n
hnk |
=h
nk | n H
xc
xc
xc

Hc
c + |nk ihnk | n
= |nk ih
nk | n H
xc
xc
xc
(11.33)
with the former expression for the gradient of the operator, the matrix elements
of the gradient take this form
c
c
H
H
|n0 k0 i = (k k0 ) hun |
|un0 i =
xc
xc

unk
cn
0
(k k ) (cn cn0 ) h
|un0 k i +
nn0
xc
xc
hnk |
(11.34)
In the last equation the first term does not contribute to the expectation value as
the wave packet is expanded only in Bloch states of a given band n. Having this,
c
H
the next step is to calculate the matrix elements of x
x
c
hnk |

c
H
cn
unk
c | unk i (k k0 )
x
|nk0 i =
hnk |
x|nk0 i + ih
| cn H
xc
xc
xc
k
Now we can finally calculate the expectation value of the perturbation

c
H
unk
c | unk |k=kc
(
x xc ) |i = ih
| cn H
xc
xc
k

u
u
= h|H|i = Im h
| c Hc | i |k=kc .
xc
k
h|
(11.35)
Now with the expectation value of the Hamiltonian and neglecting the total
time derivatives, as it does not affect the equations of motion, the Lagrangian
can be written as
u
u
u
i + x c hu|i
i + hu|i i.
L = + kc x c + k c hu|i
kc
xc
t
(11.36)
302
Topology in Physics
The Lagrangian is invariant under gauge transformations and under the displacement of a reciprocal lattice vector. In the expression of the Lagrangian can be
seen, that under the taken assumptions (semi-classical limit) this function can
be seen as classical one, i.e. a Lagrangian that depends on the generalized
coordinates and velocities. The Lagrangian will be now be handled as a classical
one with the generalized coordinates xc and kc and the generalized velocities x c
and k c . This is in some sense unusual as now the crystal momentum stands as a
general coordinate next to the position, and the general velocities are their time
derivatives. This will not be a problem in the computations, one has only to put
the general coordinates asa vector
consisting of (xc , kc ) and the general velocities
as a vector consisting of x c , k c .
With this considerations from the Lagrangian that was computed, the semiclassical equations can be obtained by variation. This gives
x c =

kx x c +
kk k c + tk

kc

xx x c +
xk k c tx
k c =
+
xc
xx are defined by
where the components of the tensor

u
u
u
u
xx
x x i h
|
ih
|
i
kc kc
kc kc
(11.37)
(11.38)
(11.39)
and the vector tx by

(tx ) tx

u u
u u
i h |
ih
| i .
t xc
xc t
(11.40)
These tensors are known as the Berry curvatures.

These equations have an extra term in the velocity and in the energy. If the
crystal lies under electromagnetic perturbations the extra term in the velocity will
become the anomalous velocity, which gives rise to a spontaneous hall effect in
ferromagnetic materials and the extra term in the energy this will be the orbital
magnetization energy.
Berry curvatures in electromagnetic fields
Now the general semi-classical equations of motion for the Bloch electrons in a
crystal under weak perturbations are known. To illustrate what the Berry curvatures are under real perturbations we take an electromagnetic field acting on
303
the crystal. With this example the equations will be brought from a abstract form
to a more physical form. In this case, under electromagnetic field perturbation,
one of the two extra terms in the semi-classical equations will become the anomalous velocity, that predicts a spontaneous Hall effect in ferromagnetic materials
and the second will be the orbital magnetization energy. Under some symmetry
properties these two terms vanish and become the usual known semi-classical
equations of motion.
For a class of perturbations for which the Hamiltonian is of the special form
H0 {
x + 1 (
x, t) , p
+ 2 (
x, t)} + 3 (
x, t)
all the results can be expressed in terms of the unperturbed Bloch wave bases. In
this section, we shall consider electromagnetic perturbations for which 1 (
x, t) =
0[66]. An example for a Hamiltonian with 1 (
x, t) 6= 0 would be the deformation
of the crystal potential due to a deformation of the crystal.
0 (k) denote the Hamiltonian for the bare crystal, with the eigenstate
Let H
|u (k)i (the periodic part of the Bloch wave) and the band energy 0 (k) for a
particular band. The Hamiltonian gets modified by the slowly varying gauge
potentials [A (x, t) , (x, t)] of an electromagnetic field to
= H0 {k + eA (
H
x, t)} e (xc , t)
(11.41)
This has the above form, with the gauge potentials playing the role of the mod c must have the form
ulation functions, and hence the local Hamiltonian H
c = H0 {k + eA (xc , t)} e (xc , t) .
H
(11.42)
As eA (xc , t) is fly an additive constant to the crystal momentum k, the basis

states have the form |u (xc , k, t)i = |u (K), where K = k + eA (xc , t) is the gauge
invariant or mechanical crystal momentum[66].
From the crystal momentum K that is gauge invariant, some simplifications
c can be
can be achieved. The expectation value of the local Hamiltonian H
written as
c (xc , K, t) = 0 (K) + e (xc , t) .
(11.43)
Therefore the total energy can be written as
= c (xc , K, t) + = 0 (K) + e (xc , t) + .
(11.44)
At this point, we have to know the expression for the perturbation energy, i.e.
for , this becomes the orbital magnetization energy of the wave packet,
= M B
(11.45)
304
Topology in Physics
where B = curlxc A (xc , t) is the magnetic field, and M is the magnetic field

u
u
0 (K) | i |k=kc
M = e= h
| 0 H
(11.46)
K
k
is the orbital magnetic moment of Bloch electrons.[66] This expression comes
from the general one, derived above, and substituting the mechanical crystal
momentum in the equation.
Now, following the same schema as above, for the general derivation, the Lagrangian can be written explicitly and takes the final form of
c hu|i u i
L = M + e (xc , t) + x c Kc ex c A (xc , t) + K
Kc
(11.47)
where M = 0 (Kc ) MB.

To get the semi-classical equations of motion for the Bloch electrons in a crystal
under an electromagnetic field, one can use the general equations derived above
and only substitute with the new values, or with the special Lagrangian the
equations can be obtained variationally. This equations get the form of
M
c
K
Kc
c = eE ex c B
K
x c =
(11.48)
where E = gradxc (xc , t) A (xc , t) /t is the electric field and

1
KK
() =
(11.49)
2
KK given in
are the components of the vector form of the antisymmetric tensor
the general formulation of the semi-classical equations. Here repeated Cartesian
indices are taken to be summed[66].
Here we have the new equations of motion, if these are compared with the
ones given in the review (11.6), two terms clearly differ from these two equations.
The first one, as mentioned above, in the energy, is the orbital magnetization
energy, the other one, also mentioned above, is an extra term in the velocity,
which is called anomalous velocity. This term, has an analogous presentation as
the magnetic field dependent term in the Lorentz force. That means that the
vector of the antisymmetric tensor can be viewed as a magnetic field acting on
the reciprocal space. This vector will be called reciprocal magnetic field and has
some interesting properties. Some of this properties will be mentioned later, it
is more convenient to analyze the equations under some symmetries and see how
this new terms behave under these symmetries.

Symmetry
time reversal
spatial inversion
kc
-
k c
+
-
305
xc
+
-
x c
-
B
+
E
+
-
Table 11.1: Behavior of the reciprocal magnetic field and magnetic moment under
Spatial inversion and time reversal symmetries.
The reciprocal magnetic field and the magnetic moment in presence
of crystal symmetries If a crystal has a specific symmetry, then the derived
equations must have the same symmetries. These type of conditions imposed to
the equations restrict severely the possible form of the Berry curvature (reciprocal
magnetic field) and of the magnetic moment as function of K.
c
Under time reversal transformation, Kc , x c , and B change sign, while xc , K
and E are fixed. Under spacial inversion, E, xc , Kc , Kc and x c change sign,

while B is fixed (table 11.1). Under pure rotations, all of these quantities behave
as vectors. The presence of these symmetries has the following consequences.
If the unperturbed system has time-reversal symmetry, the symmetry condition
on the velocity formula requires that (K) = (K), (K) = (K) and
M (K) = M (K). If the unperturbed system is invariant under spatial inversion the symmetry condition on the velocity formula gives rise to (K) = (K),
(K) = (K) and M (K) = M (K). Therefore, with either time-reversal or
spatial-inversion symmetry, the band energy is an even function of K, whereas
the reciprocal magnetic field and the magnetic moment behave differently in the
two situations; odd with time-reversal symmetry and even with spatial-inversion
symmetry[67].
This analysis shows that if a crystal has simultaneously time-reversal symmetry and spatial inversion symmetry, both, the reciprocal magnetic field and the
magnetic moment vanish identically throughout the Brillouin zone. In this case
the velocity equation reduces to the usually known expression. In many cases
in solid state physics both symmetries are present, so that the usually known
equation is valid for a large number of situations. But there are many cases
where one of these symmetries are not present. For example, in ferromagnetic or
anti-ferromagnetic materials the crystal breaks the time-reversal symmetry.
Anomalous Velocity and Magnetic Monopoles in Momentum Space
In this section, some properties of the extra term in the velocity equation, namely
the anomalous velocity, are presented. This term does not vanish in a metallic
ferromagnet and in this case an anomalous Hall effect can be measured. This
effect is a phenomenon in which the transverse resistivity xy in ferro-magnets
306
Topology in Physics
Figure 11.2: The transverse conductivity xy is shown as a function of M0 , together with the calculated results[68].
contains a contribution from the magnetization M0 (this is not the same magnetization as the one given above, this magnetization is the one which leads to the
ferromagnetic properties of the crystal) in addition to the usual Hall effect. The
conventional expression for xy is
xy = R0 B + 4Rs M0
where R0 is the usual Hall coefficient and Rs the anomalous Hall coefficient[68].
This conventional expression says that the extra term in the Hall conductivity
is proportional to M0 . The conventional expression is not always supported by
experimental data, the experimental data shows, in contradiction to the conventional expression, for the conductivity a non-monotonous dependence of M0
(figure 11.2). Here we see clearly a limit of the conventional expression, it can
only be applied in a region where the extra term is proportional to M0 . But
if the extra term that gives rise to the AHE is seen as a consequence of the
new term that arose in the velocity equations (11.48) and calculate this with the
Berry curvature given above, then the samples measured follow the same rule
qualitatively[68], as seen in the figure 11.2
To continue the discussion of the anomalous Hall effect (AHE) it is convenient
to examine the Berry curvature, i.e. the reciprocal magnetic field more in detail.
As mentioned above, this extra term has the form of a magnetic field acting on
the momentum space, therefore we expect that this reciprocal magnetic field has
similar properties as a normal magnetic field. As known form electromagnetics,
the divergence of a magnetic field is equal zero, in the case of the reciprocal
magnetic field we have
X
(K) =
qni 3 (K Kni )
i
307
Figure 11.3: Calculated flux distribution in k-space, The sharp peak around
kx = ky = 0 and the ridges along kx = ky are due a to the near degeneracy[68].
with qni = 2, and we have restored the band index n[69].This means that it
is divergence free except for quantized monopoles sources with a charge quantum 2, which are associated with band degeneracies. These occur at isolated K points[69]. This result can be interpreted as magnetic monopoles corresponding to the source or sink of the curvature defined by the Berry phase
connection[68](figure11.3).
At this point it is useful to make a small summary of the AHE, to give the
last results for the AHE calculated with the Berry curvature. The reciprocal
magnetic field acting on the momentum space, can be seen as a field which
sources are magnetic monopoles in the momentum space, these monopoles arise
in the band degeneracies. Now to see that the AHE is given by the shape of
the Berry curvature, and to illustrate the dependence of M0 of the transverse
resistivity, we write the transverse conductivity xy as the sum of the anomalous
velocity over the occupied states[68]
X

KK
xy =
nF {n (K)}
xy
n,K
()

with the Fermi distribution nF () = 1/ e T 1 . Here a the transverse conductivity has a Temperature dependence, if the magnetization M0 is calculated
in dependence of Temperature, we can combine these two and get the magnetization dependence M0 for the transverse conductivity. As said before, this method
results to describe the experiments non-monotonous dependence of M0 qualitatively (figure 11.2).
308
Topology in Physics
Semi-classical quantization
Now we will proceed with the last part, namely, the re-quantization. Now the
equations of motion in the semi-classical limit are known, and the dynamics of
the electron are given. As we are in the limit of a classical system and a quantum
system we can proceed to quantize the equations derived by before. First the
general equations will be quantized, and then separately when an uniform electric
field is present and when an uniform magnetic field is present.
Semi-classical quantization for a general Hamiltonian
First of all we need to compute the canonical momenta conjugate to the generalized coordinates, these are given by the partial derivative of the Lagrangian in
respect to the generalized velocities
P1 =
L
u
= kc + hu|i
i
x c
xc
(11.50)
L
u
i
= hu|i
kc
k c
(11.51)
P2 =
and the semi-classical Hamiltonian is given by the Legendre transformation

u
i
(11.52)
t
The semi-classical Hamiltonian is independent of P1 and P2 , and the canonical
momenta do not depend on the generalized velocities. Therefore, one cannot
obtain the equations of motion from the Hamilton equations if the Berry-phase
terms are nonzero and the formal quantization procedure cannot be applied.
H = x c P1 + k c P2 L = (xc , kc , t) hu|i
If we make the restriction of static perturbations, the semi-classical quantization can be applied, and for a wave packet motion that is regular and is described
by closed orbit in the phase space (xc , kc ), semi-classical energy levels are obtained
using the quantization procedure due to Einstein, Brillouin and Keller[66]. This
procedure gives the quantized energy values as
I
I

(11.53)
P1 dxc +
P2 dkc = 2 m +
4
C
C
where C denotes an orbit of constant energy , m is an integer that labels the
eigenvalue and the number of caustics traversed. Inserting in this quantization
the canonical momenta the above quantization is

I
(C )
kc dxc = 2 m +
(11.54)
4
2
C

with
I
(C ) =
C
u
dxc hu|i
i+
xc
309
I
dkc hu|i
C
u
i
kc
(11.55)
being the Berry phase acquired by the wave packet upon going round the closed
orbit once.
Here we have seen, how in an easy way to of re-quantization the Berry phase
arises as a correction to the quantized energy levels.
Semi-classical quantization in presence of an uniform magnetic field
In presence of only an uniform magnetic field the equations of motion simplify as

Kc
Kc = ex c B
x c =
(11.56)
These two equations can be combined in a single one for the K-motion. It
follows from the equations that the component of K along the magnetic field and
the electronic energy (K) are both constants of the motion. With these two
conservations laws the orbit of the motion is completely determined in K-space:
Electrons move along curves given by the intersection of surfaces of constant
energy with planes perpendicular to the magnetic field[63]. Such a closed orbit
is known as a cyclotron orbit. As the motion over the K-space is closed the
quantization formula of EBK can be applied. For this we need the canonical
momenta which is given by
P1 = Kc eA (xc , t)
u
P2 = hu|i
i
Kc
(11.57)
and the quantization yields

I
e|B|
1
1 (C )
B
Kc dKc =
m+
2
~
2
2
C
where
I
(C ) =
dK hu|i
C
u
i
Kc
is the Berry phase acquired by the wave packet upon going around the orbit C .
This phase influences energy levels, and affects the density of states[66].
310
Topology in Physics
Semi-classical quantization in presence of an uniform electric field

In presence of an uniform electric field, the general solution of the equations are
K (t) = K (0)
eE
~
All electrons change their wave vector in a time t by the same amount. We note
that this is not a current-carrying configuration, which may not be intuitive in a
classical form. The current carried by an electron is proportional to its velocity,
which in this case is not proportional to the wave vector K

eE
v (K (t)) = v K (0)
~
Since v (K) is periodic in the reciprocal lattice, the velocity is a bounded function of time and, when the field E is parallel to a reciprocal lattice vector,
oscillatory[63]. This behavior is a consequence of the additional force exerted
by the periodic potential, which , though no longer explicit in the semi-classical
model, lies buried in it (through the functional form of (K))[63].
For simplification we will restrict the discussion to one dimension, this will
guaranties that the electric field will be parallel to a reciprocal lattice vector, so
that the motion in real space will be bounded and periodic. So in the reduced
zone scheme, the motion is closed in the phase space (xc , Kc ). The Hamiltonian
for the presence of a uniform electric field is
H = 0 (Kc ) + eExc
Following the semi-classical quantization of EBK the formula yields

Z
a
dKc xc (Kc ) = 2 m +
4 2
(11.58)
where a stands for the lattice constant, and

Z
a
u
=
i
dKhu|i
K
a
is known as the Zak phase, xc (Kc ) is the constant energy curve for the mth
energy level defined by Wm = 0 (Kc ) + eExc . Averaging this expression over the
orbit, we obtain

Wm = 0 + eEa m +
(11.59)
4 2
where 0 is the average of the band energy over the Brillouin zone. This spectrum
is known as the Wannier-Stark ladder.[66]
12 Superfluidity and winding

numbers
Barbara Theiler
Supervisor: Lode Pollet
After a brief introduction on the phenomenon of superfluidity
and on its macroscopic description known as the two-fluid model,
the quantum-classical isomorphism will be introduced. In the
path integral the thermodynamic properties of a Bose system
are mapped onto those of a classical system of interacting ring
polymers. An explicit formula for the superfluid density s can be
derived in terms of the winding number, a topological invariant
of the polymer system. As for most many-body problems it is
impossible to treat this problem analytically. By use of Monte
Carlo simulations however, numerical results for s are obtained
which are in good agreement with experimental measurements.
Introduction
Superfluidity is a phase of matter which is characterized by the total absence of

viscosity and is related to the behavior of quantum liquids at low temperatures.
The first observations of this phase were made in 1937 in liquid 4 He which is
the classic example of a superfluid. Below a characteristic temperature some
properties of the system change abruptly. For instance, the heat flow is no longer
proportional to the temperature gradient, and the liquid is able to flow without
apparent friction. This sudden change implies that a phase transition has taken
place which divides the liquid phase into He-I and He-II. While He-I behaves like
a normal fluid, He-II exhibits superfluid behavior.
This second-order transition to the superfluid state is known as the -transition
due to the -shaped anomaly in the specific heat curve. Along the vapor pressure
311
312
Topology in Physics
Figure 12.1: Phase diagram of 4 He near the transition temperature.

curve, the -transition occurs at the -point
T = 2.17K,
3
V = 46.2
A /atom.
Near T , the specific heat c is of the form
c(T ) A |T T | ,
where A is the amplitude for positive/negative |T T |, and the critical exponent. The theory of Ginzburg-Landau predicts = 0. Recent measurements
[70] however obtain = 0.0127 0.0003. This implies that the heat capacity
does not diverge at T = T but stays finite.
A remarkable property of helium is that it does not solidify under its own vapor pressure down to the absolute zero of temperature. Solidification occurs at
external pressure of at least 25bar. Qualitatively, this may be understood as follows [71]: As a noble gas, its only molecular interaction is a weak Van der Waals
attraction. In addition, the mass of He is smallest among the noble gases. These
circumstances result in a large zero-point motion which inhibits the localization
of the atoms at well-defined lattice sites.
The fact that helium is liquid down to zero temperature makes it a nearly perfect
system for studying macroscopic quantum effects. Since the atoms are delocalized, the indistinguishability of the particle becomes very important. Furthermore, as helium has two isotopes, 3 He (= 2p + 1n + 2e Fermion) and 4 He
(= 2p + 2n + 2e Boson), one can experimentally distinguish the effects of
correlation from those of statistics.
3
He and 4 He behave quite differently at low temperatures; 3 He does not undergo
a phase transition to a superfluid state near T = 2.17K. Apart form a difference in atomic masses, the only difference between these two isotopes is that 4 He
Superfluidity and winding numbers
313
Figure 12.2: Experimental specific heat of liquid 4 He along the vapor pressure curve. The low-temperature behavior implies a linear dispersion relation
(phonon-excitations).
atoms obey Bose statistics, whereas 3 He atoms follow the statistic of Fermi-Dirac.
The lower the temperature, the more impact do quantum statistics have on the
system. It is therefore natural to assume that the phenomenon of superfluidity
is a quantum statistical effect. But before we study the properties of an interacting many-Boson system such as liquid 4 He, we have a look at the macroscopic
description of superfluidity.
The Two-Fluid Model
Below a critical velocity, He-II can be observed to flow through a thin capillary
(superleak) with zero resistance, a phenomenon which implies zero viscosity.
On the other hand, if a cylinder is rotated in a He-II-bath, there is a momentum
transfer from the rotating body to the liquid, indicating that the viscosity is not
zero. This apparent contradiction gives rise to the idea of the two-fluid model.
Macroscopically, the anomalous flow behavior of the He-II phase can be understood in terms of this model; He-II is assumed to be made up of two components
called the normal and the superfluid component. Between these two components
there is no friction, i.e., no transfer of momentum, and they cannot actually be
separated. Experimentally, the two fluids are distinguished through their behavior in the presence of moving boundaries as the example of the rotating cylinder
shows; the normal component gets accelerated by the external motion, whereas
the superfluid component stays at rest.
A first quantitative theory of the flow properties of He-II is due to Landau (1941).
It provides an explicit construction of the two-fluid model near absolute zero
314
Topology in Physics
([71], [72]): The superfluid component is identified with the part of the liquid
that remains in its ground state, while the normal component corresponds to
low-temperature excitations. Assuming that there are no other excitations near
the ground state except elementary excitations (whose spectrum, illustrated in
Figure 12.3, is found experimentally by neutron scattering [73]), it is possible to
describe the normal component as an ideal gas of Boson-like quasiparticles with
energy
X
En = E0 +
~k nk ,
k
where ~k is the energy of the elementary excitation of wave number k, and {nk }
a set of occupation numbers. For T sufficiently low, one can assume the nk s to be
independent and their thermodynamic average to be given by hnk i = [e~k 1]1 .
The internal energy of the liquid of volume V is then
Z
X
V
k 2 ~k
U = E0 +
~k hnk i = E0 + 2
dk ~
.
(12.1)
k 1
2
e
0
k
Figure 12.3: Energy spectrum of elementary excitations in liquid 4 He.

According to [72], the excitation spectrum can be derived from the structure
factor S(k) by the formula
~2 k 2
(k) =
.
2mS(k)
A typical form of the structure factor of a compressible system is shown in Figure 5. For k 0, S(k) behaves approximately linear which implies (k) ~k,
whereas the peak of S(k) leads to the roton minimum near k = 2/d, where d is
the particle distance.
315
Figure 12.4: The structure factor of 4 He at 1.38K and saturated vapor pressure:
solid line, calculated by PIMC; , measured by neutron scattering (Sears et al.,
1979); , measured by x-ray scattering (Robkoff and Hallock, 1981).
At low temperatures, there are two types of quasiparticles which contribute significantly to the integral in (12.1): phonons with spectrum (k) = c~k and rotons, which are regarded as quantized rotational motion with spectrum (k) =
2
2
0)
. The parameters are experimentally found [71]:
+ ~ (kk
2
m
1
c = 239 , = 8.65 K, k0 = 1.9
A , = 0.16mHe .
s
The specific heat cV = U
can be approximated as the sum of the contribution
T
of the phonons and the one of the rotons
cV = cphonon + croton .
The phenomenon of frictionless flow for a velocity less than a critical velocity
vc = minp (p)
may be understood in terms of the gas of quasiparticles: Consider
p
an external object with mass m moving with velocity ~v through the stationary
liquid. The only way of transferring momentum and energy to the liquid is by
creating an excitation. Let p~ be the momentum of the new quasiparticle. Then,
by conservation of momentum, m~v = mv~0 + p~, where v~0 is the remaining velocity
of the object. Its kinetic energy is
p2
1 02 1 2
mv = mv ~v p~ +
.
2
2
2m
The creation of an excitation with energy (p) requires 12 mv 2 >
which for large m leads to the criterion
p2
(p) < ~v p~
~v p~ vp.
2m
1
mv 02
2
+ (p)
(12.2)
316
Topology in Physics
Therefore, in order to transfer energy to the liquid it must hold: v > vc . vc is determined by drawing the tangent with smallest slope from the origin to the curve
of the energy spectrum (Figure 12.3). The critical velocity for 4 He obtained in
this way is ~k0 60 ms . The experimental value however is only about 1 cm
. This
s
implies that there are other excitations at small velocity (Vortices, see [72]).
Nevertheless, from (12.2) one can conclude: Any spectrum in which sufficiently
small excitations are phonons may lead to superfluidity, as (p)
> 0 near p = 0.
p
The dispersion relation for a massive particle satisfies (p)
0 as p 0. Hence,
p
such a system cannot be superfluid. In particular, there is no superfluid phase in
an ideal Bose gas.
Besides the ground state property, the superfluid component is conceived as carrying zero entropy and flowing irrotationally; by contrast, the normal component
behaves like any other viscous fluid. From these apparently minimal postulates
Landau was able to derive a complete, quantitative theory of two-fluid dynamics.
While this two-fluid hydrodynamics provides a conceptual basis for superfluidity,
which still stands today, it is phenomenological in the sense that both properties
of the superfluid and the nature of the excitation spectrum are postulated in an
intuitive way rather than being explicitly demonstrated to be a consequence of
the Bose statistics obeyed by the atoms.
Bose-Einstein Condensation
An ideal Bose gas below a characteristic temperature, which depends on the
mass and density, exhibits Bose-Einstein condensation (BEC, see [74]); A finite
fraction of all the particles (and at zero temperature, all of them) occupies a
single one-particle state. For the usual case of periodic boundary conditions and
translational invariance, this is the zero momentum state.
In a noninteracting gas with the mass and density of 4 He, the phenomenon of
BEC would occur at 3.1K: The critical temperature for an ideal Bose gas with
specific volume V is given by
Tc =
~2
2
2
( 32 ) 3 kB mV
2
3
where kB is the Boltzmann constant and ( 23 ) = 2.612, where (x) for x > 1 is
3
the Riemannian Zeta-function. For 4 He with V = 46.2
A /atom, Tc = 3.1K.
In order to generalize this notion to interacting many-Boson systems, we introduce the one-body density matrix
(~ri )(~
rj )i,
(1) (~ri , ~rj ) = h
r) is the field operator that annihilates a single particle at ~r in the
where (~
system. The eigenvectors and eigenvalues of (1) (~ri , ~rj ) describe the so-called
317
natural orbitals and the corresponding occupation numbers. BEC is present if

one of the natural orbitals is macroscopically occupied. This means that the
condensate fraction fc = NNc remains finite in the thermodynamical limit. Here,
Nc is the number of condensate particles which is given by the occupation number
of the orbital providing the condensate wave function c .
The presence of BEC is equivalent to the existence of off-diagonal long-range
order (ODLRO), i.e., the matrix elements of the one-body density matrix do not
go to zero far off the diagonal but remain finite:
(1) (~ri , ~rj ) 6 0,
|~ri ~rj | .
For a translationally invariant system

1 X i~k(~ri ~rj )
e
hn~k i
~k
Z
hn~0 i
d3 k i~k(~ri ~rj )
=
+
e
hn~k i,
(2)3
(~ri )(~
rj )i =
(1) (~ri , ~rj ) = h
and in the limit |~ri ~rj | , the second term (all ~k 6= 0) vanishes by cancellation.
BEC and Superfluidity
Macroscopic occupation implies phase coherence which means that all properties
of the condensate wave function may be expressed in terms of the single-particle
wave function c (~r) = (~r)ei(~r) . The condensate density is independent of the
phase: c (~r) = |(~r)|2 . The current density of the condensate on the other hand
is

~jc (~r) = ~ c (~r)c (~r) c (~r)c (~r) = |(~r)|2 ~ (~r).
2mi
m
This suggests the definition of the condensate velocity as
~vc (~r)
~
(~r).
m
The flow described by this velocity field is irrotational, ~vc = 0. Moreover,

the state of the condensate is well-defined and thus the entropy must be carried
entirely by the particles occupying other single-particles states than c .
These two observations provide the basis for Landaus two-fluid hydro-dynamics.
One might leap to a false conclusion by identifying the phenomenon of superfluidity with the one of BEC. Note however, that the only assumption necessary in
order to derive the velocity field ~vc (~r) is the existence of phase coherence. Macroscopically, the superfluid density is defined through the response of the system
to an external potential while the condensate fraction is an equilibrium quantity.
The Mermin-Wagner theorem tells us that in a 2-dimensional system at nonzero temperature, spontaneous continuous symmetry breaking is not possible
318
Topology in Physics
and thus, ODLRO does not exist. The one-body density matrix however decays algebraically (BKT-transition). This quasi long-range-order is sufficient for
a superfluid state to be established showing that BEC and superfluidity can be
observed one without the other. In addition, in liquid 4 He the superfluid density
s tends to the total density = N
in the limit T 0, whereas the condensate
V
fraction remains only about 10%.
Imaginary-Time Path Integrals
In this section we consider a quantum system with fixed particle number N , temperature T , and volume (canonical ensemble). Using the path-integral method
we will see that the thermodynamic properties of Bose systems are equivalent
to those of classical systems of interacting ring polymers. This will give us a
classical picture which is not only simple but also exact for all thermodynamic
properties, including superfluidity.
We will first develop the formalism of imaginary-time path integrals for the general case of a many-body system which obeys Boltzmann statistics. Later on,
Bose statistics will be taken into account.
The Thermal Density Matrix
According to the theory of thermodynamics, all information about the statistical
properties of a many-body system is contained in the density matrix . At high
temperature, quantum systems reduce to classical systems, and one may obtain
by classical calculation. But as temperature decreases quantum effects must be
accounted for. This is where the path-integral method enters.
The density matrix can be defined in terms of the exact eigenvalues and eigen
functions of the Hamiltonian H
X
1
e H =
|i ieEi hi |,
.
kB T
i
is
The equilibrium value of an operator O
= Z 1 tr(O
) = Z 1
hOi
i ieEi ,
hi |O|
where Z is the partition function

Z tr(
) =
eEi .
In the following, we will work exclusively in position representation where the

particles are labeled. In configuration space a density matrix element is given by
X
(R, R0 ; ) hR|e H |R0 i =

i (R)eEi i (R0 ),
i
319
where R = (~r1 , ..., ~rN ), and ~rk represents the position of the k th particle. In space
dimension 3, is a function of 6N + 1 variables.
From the definition of () one derives the Bloch Equation

,
=H
with initial conditions (0) = Id, or (R, R0 ; 0) = (R R0 ) in position representation.

Basis of Path Integral
The basis for the path integral representation is the identity
e(1 +2 )H = e1 H e2 H .
It gives a way to express the density matrix at temperature T as a product of
density matrices at temperature M T with M N, i.e., for M
e H = (e H )M .
In position representation one finds
Z
Z
(R0 , RM ; ) = ... dR1 ...dRM 1 (R0 , R1 ; )(R1 , R2 ; )...(RM 1 , RM ; ),
(12.3)
where Rk = (~r1,k , ..., ~rN,k ) R is the configuration of the N particles at link k.
The succession of the points (R0 , R1 , ...RM ) is called a path. Expression (12.3)
is the sum over all discretized paths. Note that it is exact for any M > 0.
In the limit M , the path becomes continuous, but its derivative will be
discontinuous at almost all points of the path. Nevertheless, for small it provides
sufficiently accurate analytical approximations of the density matrix. This is due
to the fact that for sufficiently large M , each density matrix of the right-hand
side of equation (12.3) can be computed in the high-temperature limit for which
the so-called primitive approximation holds:
3N
e (T +V ) e T e V .
The error is of order 2 which may be seen by the Baker-Campbell-Hausdorff
formula
2
3

e T e V = e (T +V ) e C1 C2 +...
1
[T V , [T, V ]]. The Trotter formula (which holds
with C1 = 21 [T, V ] and C2 = 12
for operators that are bounded from below)
e(T +V ) = lim [e T e V ]M
M
320
Topology in Physics
guarantees a well-controlled process as the number of time steps increases.

Suppose the interaction of the many-body system is well-described by a nonrelativistic Hamiltonian of atoms interacting by a pair potential,
N
X
= T + V =
H
2i +
i=1
v(|~
ri r~j |),
i<j
~ .
with 2m
In the primitive approximation the discrete path-integral formula (12.3) for the
density matrix in configuration representation is
Z
Z
3N M
(R0 , RM ; ) =
... dR1 ...dRM 1 (4 ) 2
2
exp
M
X
(Rm1 Rm )2
m=1

+ V (Rm ) .
(12.4)
According to the primitive approximation in position space, the potential and

the kinetic part can be treated separately:
Z
(R0 , R1 ; ) dR < R0 |e T |R >< R|e V |R1 > .

From V |R >= V |R > it follows directly that
< R|e V |R1 >= e V (R1 ) < R|R1 >= e V (R1 ) (R1 R).
For the kinetic part consider N free, distinguishable particles in a cube of side
L with periodic boundary conditions. The eigenfunctions and eigenvalues are
3N
n
n (R) = L 2 eiK~n R and E~n = K~n2 with K~n = 2~
, ~n Z3N . Then
L
X
< R0 |e T |R > =
< R0 |n >< n |e T |m >< m |R >
~
n, m
~
1 X iK~n R0 K 2 iK~n R
~
ne
e
e
L3N
~
n
(R0 R)2
1 X ( 2 L~n +i (R0 R) )2
2
= e 4
e
L3N
~
n
3N
2
= (4 )
(R R)2
04
(12.5)
The last step follows by approximating the sum by an integral. This requires
that L2 which will always be assumed.
The density matrix can thus be calculated at any temperature from an integral
over the intermediate configurations R1 , ..., RM 1 . Such an integral looks like a
partition function of some classical system. In particular, the integrand is always
non-negative and can be interpreted as a probability; an essential property for
Monte Carlo simulations.
321
The Classical Isomorphism

i
In quantum mechanics, matrix elements of the time evolution operator e ~ H(tf ti )

in terms of path-integrals [75] yield the expression
Z
(xf ,tf )
U (xf tf , xi ti ) =
D[x(t)]e ~ S[x(t)] .
(xi ,ti )
Analogously
onePinterprets the exponent in formula (12.4) as an action S
PM
M
m=1 Sm
m=1 ln[(Rm1 , Rm ; )]. Since it has the form of a classical
potential-energy function multiplied by , plays the role of imaginary time,
t = it~ .
Now let us have a look at a single particle where V is some external potential.
The path consists of a list of points ~r0 , ..., ~rM . The kinetic action for the k th link
in equation (12.4) is
(~ri,k1 ~ri,k )2
3N
ln(4 ) +
.
2
4
The second term can be interpreted as a spring potential. A path corresponds
to a polymer where only next neighbors in the chain are connected with springs.
Each bead on the chain represents the particle at different times tk = k . Hence,
moving a quantum particle is equivalent to evolve the polymer.
The same interpretation can be applied to the many-body system since
M
X
N
N X
M
M X
X
X
2
(~rj,k1 ~rj,k )2 .
(~rj,k1 ~rj,k ) =
(Rk1 Rk ) =
k=1
k=1 j=1
j=1 k=1
Quantum particles may be represented by a classical system of polymers. By

adding the potential interaction, this picture only changes to a system of interacting polymers. But contrary to real-time polymers, these polymers interact
only at the same time and only between beads on different chains.
Thermodynamical properties, or static properties diagonal in position space, are
determined by the trace of the density matrix, i.e., the integral of (12.4) over
R0 with R0 = RM . This involves all possible paths that return to their starting
point after M steps. In the classical analog, these correspond to closed chains,
to ring polymers.
Bose Symmetry
Thus far only systems of distinguishable particles have been considered. Now we
want to introduce Bose statistics to our path-integral method.
For Bose systems only totally symmetric eigenfunctions i (R) contribute to the
density matrix - those such that i (P R) = i (R) for all permutations P with
322
Topology in Physics
Figure 12.5: Trace of the paths of six helium atoms in the extended cell view at
2K with 80 time slices.
P R = (~rP1 , ..., ~rPN ). The Bosonic density matrix is obtained by summing over all
possible symmetrized states in configuration space, i.e.,
1 X
(R0 , P R1 ; ),
B (R0 , R1 ; ) =
N! P
where is the Boltzmann density matrix element which we have been considering
up to this point.
A Bosonic simulation consists of a random walk through the path space and the
permutation space. The partition function for a Bose system has the form
Z
Z
3N M
1 X
ZB =
... dR0 ...dRM 1 (4 ) 2
N! P
exp
M
X
(Rm1 Rm )2
m=1

+ V (Rm ) ,
with new boundary conditions on path closure: P RM = R0 . Thus the partition

function includes contributions from N ! closures, and the allowed paths can close
on any permutation of their starting position. At high temperature the identity
permutation dominates, while at zero temperature all permutations have equal
contributions (see section 3).
As a consequence of the paths closing at any permutation, ring polymers can
cross-link. A two-atom system of M links can be in two possible permutation
states: either two separate ring polymers, each with M links, or one larger polymer with 2M links.
323
Any permutation can be broken into a product of cyclic permutations. Each

cycle corresponds to several polymers cross-linking and forming a larger ring
polymer. According to Feynmans 1953 theory [76], the superfluid transition is
represented in the classical system by the formation of macroscopic polymers, i.e.,
those stretching across an entire system and involving on the order of N atoms.
Paths of atoms involved in cyclic exchanges wrap around the periodic boundary
conditions. They are called winding paths and are a direct manifestation of superfluidity.
Figure 12.6: The extended trace of six helium atoms at a temperature of 0.75K.
Three of the atoms are involved in an exchange which winds around the periodic
cell in the x direction.
In thermodynamics one may assign to any system its thermal wavelength .
Let d be the interatomic distance. In the limit d quantum statistics can
be neglected and the system treated classically. For d ' quantum statistics
may have a great impact on the system and must be taken into account.
Analogously we want to find a criterion for the polymer system to decide whether
quantum statistics are important. In the absence of interaction, the size of a path
(or polymer) is of the order of its thermal wavelength ; The path-length hsi of
the ith particle (using equation (12.5)) is given by hsi2 hs2 i = h(~ri, ~ri,0 )2 i =
2
P
PM
ri,k ~ri,k1 )2 i = M h(~ri,1 ~ri,0 )2 i = M 2 =
h
ri,k ~ri,k1 ) i = h M
k=1 (~
k=1 (~
1
2 = 2 . When equals the interpolymer spacing, roughly 3 where is the
particle density, it is at least possible for the polymers to link up by exchanging
1
end points. This relationship = 3 , defines the degeneracy temperature
2
3 ~2
TD =
.
mkB
324
Topology in Physics
For temperatures higher than TD , quantum statistics are not very important.
In a liquid state, TD gives a surprisingly good estimate of the transition temperature: For ideal Bose condensation in three dimension, TTDc = 3.31, for liquid 4 He
at saturated-vapor-pressure conditions, TTD = 2.32.
The Superfluid Transition
A typical feature of the superfluid transition is the anomalous shape of the specific heat curve near the transition temperature. Feynman showed that it is the
formation of macroscopic polymers which leads to this peak. A qualitative
argument may be given as follows ([77],[76],[72]); The effect of interaction can be
taken into account by replacing the mass of the atoms by an effective mass, i.e.,
, and the partition function simplifies to
Z
Z
3N M
1 X
ZB
... dR0 ...dRM 1 (4 ) 2
N! P
exp
1 X
N! P
M
X
(Rm1 Rm )2
m=1
dR exp
4
(R P R)2
.
4
Not all particle configurations are equally probable; Configurations with wellseparated atoms are important while configurations where atoms overlap can be
neglected (due to the hard-core pair-potential of helium). This is taken into
account by introducing a normalized configuration distribution f (R) into the
partition function
Z
X
K
(R P R)2
ZB
dRf (R)
exp
.
N!
4
P
As the pair-correlation function of liquid helium does not change much in the
region of -transition [77],P
one may assume f (R) to be independent of tempera1
ture and write f (R) = N ! P (R P R0 ) where R0 is a typical configuration of
atoms.
In contrast to ideal Bosons which tend to attract each other to maximize exchange, the interatomic potential of liquid helium does not allow regions of high
density. Thus, we may choose R0 to be a perfect cubic lattice with spacing d
( 3.6
A). Permutations among particles can be visualized as polygons with arrows on the lattice. The sum over all permutation is therefore
equivalent to the
d2
sum over all possible polygon patterns. At 2.2K, exp 4 0.3, which shows
that a side of a polygon longer than d is not important, and that the sides of all
325
polygons that contribute have length of order d, i.e.,

X
P
exp
X
(R0 P R0 )2
s(P )d2
g
exp
,
P
4
4
P
where fP is the sum over all polygons made up of lines joining near neighbors,
and s(P ) is the total number of sides. With these assumptions, an r-sided polygon
2
contributes to Z roughly with the weight y r where y = exp 4d ), and an rcycle contributes with nr y r where nr is the number of polygons with r sides
which increases with r. y depends essentially on the temperature. While at
high temperature long exchanges may be neglected as y 1, they become more
important as the critical temperature is approached, y 1. This yields the phase
transition.
According to thermodynamics, the internal energy U of our system is
U
T 2 hxi
,
N
as the relation between the internal energy and the partition function is given by
U =
ln(Z). Therefore, one obtains for the specific heat
cV 2
T hxi T 2 h(x hxi)2 i

+
.
N
N
The specific heat depends crucially on the mean-squared fluctuation of the exchange distance. Its anomaly at the -point is due to the fact that in this region
both short and long exchanges contribute significantly to Z.
Path Integral and Superfluidity
In section 2 we have seen that certain properties of liquid helium can be related
to path integrals. We will now explicitly do so for the momentum distribution
and for the superfluid density.
The Momentum Distribution
The probability density of finding a particle with momentum ~~k1 is defined as
Z
Z
2

3N
n~k1 = (2)
d~k2 ...d~kN dR(R)eiKR ,
where (R) is the many-body wave function, R = (~r1 , ..., ~rN ), and K = (~k1 , ..., ~kN ).
By expressing the one-body density matrix in configuration space
Z
(1)
0
(~r1 , ~r1 ) =
d~r2 ...d~rN (~r1 , ~r2 , ..., ~rN , ~r10 , ~r2 , ..., ~rN ; ),
Z
326
the momentum distribution simplifies to
Z
1
0
~
n~k =
d~rd~r0 eik(~r~r ) (1) (~r, ~r0 ).
3
(2)
Topology in Physics
(12.6)
R
The normalization factors in the last two expressions are such that d~r(1) (~r, ~r) =
R
and d~kn~k = 1, where is the volume of the periodic cell. In terms of n~k
the condensate fraction which is the probability of finding a particle with zero
3
momentum is given by fc = (2)
n~0 .
From equation (12.6) follows that n~k is the Fourier transform of (1) (~r, ~r0 ), which
itself is the contraction of an off-diagonal element of the density matrix. In the
classical analog, such a density matrix element is represented by a possibly crosslinked configuration of (N 1) ring polymers and 1 linear polymer (worm). A
cross-link of the linear with a ring polymer results in the destruction of the ring
polymer and the worm gets longer, i.e., the worm eats up the ring polymer.
Figure 12.7: Extended trace of five 4 He atoms at T = 0.75K. The dotted path is
that of the cut polymer, the one that is not periodic in time.
In the high-temperature limit, n~k results in the Maxwellian momentum distri
bution exp ~k 2 ; As quantum effects are not important (no macroscopic
exchange), the end-to-end distribution of the linear polymer is almost free-particle

~
r0 )2
like, i.e., (1) (~r, ~r0 ) exp (~r4
.
Note that in the presence of macroscopic exchanges, the two ends of the linear
polymer can become much more separated than a thermal wave length as they
may be involved in a permutation cycle as well. This separation however depends
not only on the statistical mechanics of the system, but also on its dimension.
The one-body density matrix of a 3D bulk liquid in the superfluid state tends to
327
a constant as r gets large [77], and the momentum distribution shows a peak at
the origin which implies the existence of a condensate. For a homogeneous liquid,
we get the condensate fraction
Z
(2)3
1
fc =
n~ =
d~rd~r0 (1) (~r, ~r0 )
0
2
Z
1
=
d~r(1) (r).
=const.), one sees that the conIn the thermodynamical limit (, N , N
densate fraction is the large-distance limit of the one-body density matrix

fc = lim (1) (r).
r
Thus the condensate fraction can be interpreted as the probability of finding a

configuration where the worm-ends have attached themselves to a macroscopic
exchange and the worm stretches across the whole system.
The Superfluid Density
As mentioned in section 1, the normal and the superfluid components can be
determined experimentally by their different response to boundary motion. In
particular, one finds an explicit expression for the superfluid density s in a
periodic system in terms of a nonclassical free-energy change due to this motion
[78]. Note that in this section the energy is measured with respect to the center
of mass.
Consider a system enclosed between two cylinders of radii R and R + d which
rotate with angular frequency . For Rd 1 centrifugal effects can be neglected
and this set-up is equivalent to a system between two planes moving with velocity
v = R which has periodic boundary conditions in one dimension (L 2R).
The density-matrix operator in the moving frame is
v = e Hv ,
where
v =
H
X (~pj m~v )2
+ V.
2m
j
p~j denotes the momentum of the j th particle in the lab frame.

Since the superfluid component is viscous-free, the part of the liquid which responds to the boundary motion is exactly the normal component,
tr(P~ v )
n
N m~v = hP~ i =
,
tr
v
328
Topology in Physics
where P~ is the total momentum of the liquid in the lab frame and = n + s
the total particle density. This equation may also be written in terms of the free
energy Fv = 1 ln(Z) = 1 ln(tr
v ) of the system with moving walls:
n
N m~v =
ln(tr
v ) + N m~v
~v
Fv
=
+ N m~v .
~v
The free-energy change due to the boundary motion is for small velocities
Fv
Fv Fv=0
1
s
=
= mv 2 + O(v 4 ).
N
N
2
The last step follows by the Taylor series expansion of

FNv
Fv
N
(12.7)
in 12 mv 2 around v = 0
v
using ~v = m~v ( 1Fmv
2 ) . This free-energy change Fv depends on the superfluid
2
component and does not occur for a classical fluid (s = 0).
In the following, we generalize our considerations to a system which is periodic
in all spatial directions. For the computation of Fv and hence s by the pathintegral algorithm we switch to the system of stationary walls. v satisfies the
Bloch equation

v (R, R0 ; ) 1 X
=
(i~j m~v )2 + V v (R, R0 ; ),
2m j
with periodic boundary conditions

~ ...~r0 ; ) = v (~r1 , ..., ~rN , ~r0 ..., ~r0 , ...~r0 ; ).
v (~r1 , ..., ~rN , ~r10 , ..., ~rj0 + L,
N
1
j
N
m
By introducing the gauge transformation (R) ei ~ ~v

equivalent system
~
rj
(R), we get the

(R, R0 ; ) 1 X
=
(i~j )2 + V (R, R0 ; ),
2m j
with boundary conditions
m
~ ...~r0 ; ) = ei ~ ~vL (~r1 , ..., ~rN , ~r0 ..., ~r0 , ...~r0 ; ).

(~r1 , ..., ~rN , ~r10 , ..., ~rj0 + L,
1
j
N
N
Hence satisfies the usual Bloch equation in
the system with stationary walls.
m P
0
Note, in addition, that v (R, R0 ; ) = ei ~ ~v j (~rj ~rj ) (R, R0 ; ). The calculation
of the density matrix ~v can therefore be done by use of the usual density matrix
if an additional phase factor is included as a weight. In particular, in a pathintegral calculation the contribution to the density matrix from a path ending
329
~
on a periodic image of its initial point must include the factor ei ~ ~vL . The fact
that the density matrix element depends now on the actual trajectories of the
~:
particles suggests the definition of the winding number W
X
~L
W
(~rP (j) ~rj ), P SN .
m
The winding number is obtained by tracing the path of each single particle from
its initial position ~ri to its final position ~rP (i) a time later and counting the
number of times periodic boundary conditions have been invoked. Usually there
are periodic boundary conditions in all spatial direction and the winding number
is a vector which is given in units of the box length. The winding number is a
topological invariant of a given path; It can be determined by counting the flux
of paths across any plane, no matter where it is inserted. As we will see, paths
with a nonzero winding are the signal for superfluidity.
The free-energy change is related to the winding number by
m
Fv
m
tr
v
tr(ei ~ ~v j (~rj ~rj ) )
~
=
=
= hei ~ ~vW L i.
tr
v=0
tr
Thus Fv is the Fourier transform of the winding number distribution and is

h
~ vanishes, and for small
-periodic in v. By symmetry, the average value of W
mL
velocities
m2 v 2 hW 2 iL2
Fv =
+ O(v 4 ),
2
2~
3
1
where the factor 3 comes from symmetrizing over the three equivalent axis assuming a cubic periodic cell. By comparison with equation (12.7) one obtains
the winding number formula for the superfluid density
s
m hW 2 iL2
= 2
.
~ 3N
The analogous derivation in d spatial dimensions results in the formula
s
m hW 2 iL2d
= 2
.
~
d
The superfluid density is essentially given by the mean-squared winding number,
i.e., a nonvanishing probability for winding paths implies that the system is in
the superfluid state. This formula provides a very elegant way to evaluate s .
The tricky part about it is the computation of hW 2 i which cannot be done in an
analytical way but must be solved numerically, by Monte Carlo simulations.
330
Topology in Physics
Path-Integral Monte Carlo Simulations
Monte Carlo simulations play a very important role in the theoretical investigation of complex systems such as quantum many-body systems. In addition to
accurate results, they may also improve our conceptual understanding of certain
phenomena.
For Bose systems, the most powerful method is the path-integral Monte Carlo
simulation (PIMC) which is a computational technique for simulating quantum
systems at T 6= 0K, and which is exact in the sense that all approximations are
controllable. PIMC is based on the quantum-classical isomorphism introduced in
section 2. The explicit expression for the density matrix (12.3) requires altogether
3N (M 1) integrations. Without introducing very severe approximations, the
only way of doing the integrals as N gets large is stochastically, i.e., by sampling
the integrand. In a few words, certain values of the input random variables in a
simulation have more impact on the parameter being estimated than others.
The Monte Carlo evaluation for the given problem consists of generating a large
s
set of independent paths-configurations X s [R0s , ..., RM
], s = 1, ...S statistically
sampled from a probability density proportional to (R0 , RM ; ). The thermal
can then be estimated as a statistical average over the
average of any operator O
s
set of values {O(X )},
S
1X
O(X s ).
hOi =
S s=1
Importance sampling is mostly based on the Metropolis algorithm, which is a
Markov chain Monte Carlo sampling algorithm for generating an arbitrary given
distribution (s). A Markov chain is characterized by a transition rule P (s
s0 ) = T (s s0 )A(s s0 ) and generates a reversible random walk {s0 , s1 , ...},
where s = [P, R0 , ..., RM ] with RM = P R0 is an element of the total configuration
space. At each point on the walk a random trial move from the current position in
configuration space is selected. This trial move is then either accepted or rejected
according to a simple probabilistic rule.
If the transition rule is ergodic and fulfills the condition of detailed balance
(s)P (s s0 ) = (s0 )P (s0 s),
(12.8)
then the probability distribution converges to an equilibrium state satisfying

X
(s)P (s s0 ) = (s0 ),
s
and the Markov process samples (s) (see [75]). Note that detailed balance is not
a necessary, but for practical use a very convenient condition. Ergodicity means
that one can move from any state to any other state in a finite number of steps
with a nonzero probability.
331
In the Metropolis method the transition probability is split into an a priori

sampling distribution T (s s0 ) and an acceptance probability A(s s0 ),
P (s s0 ) = T (s s0 )A(s s0 ),
where
T (s0 s)(s0 )
A(s s0 ) min 1,
.
T (s s0 )(s)
If A(s s0 ) > , where is a randomly drawn number between zero and one then
the move is accepted, otherwise rejected. This P (s s0 ) satisfies by definition
(12.8) which ensures asymptotic convergence.
Note that the acceptance rate is not unique, there are many possible choices.
Another choice that also satisfies the detailed balance equation is for example
A(s s0 ) =
(s0 )
(s)
0)
+ (s
(s)
This is known as the heat-bath algorithm and will be used in the next subsection.
Monte Carlo Simulation on a Lattice in One Dimension
In order to understand the basic concepts of PIMC, we consider a one-dimensional
system of N hard-core Bosons which occupy arbitrary sites of a finite spatial
lattice. We work at finite and impose periodic boundary conditions. The term
hard-core refers to the fact that the particles are not allowed to overlap and
therefore cannot occupy the same lattice site at equal times.
Suppose the Hamiltonian of our system can be written in the form
=H
1 + H
2,
H
1 and H
2 each being diagonalizable. In the primitive approximation we
with H
find for the partition function
X
Z =
hi1 |e H |iL ihiM |e H |iM 1 i...hi2 |e H |i1 i
i1 ...iM
hi1 |U1 |i2M ihi2M |U2 |i2M 1 ihi2M 1 |U1 |i2M 2 i...hi2 |U2 |i1 i, (12.9)
i1 ...i2M
with Uj = e Hj and = M
. By choosing the intermediate state |ik i in a
1 for k odd, and as an eigenstate of H
2
convenient way, e.g. as an eigenstate of H
for k even, one obtains a quite simple expression for the partition function Z.
In the following, we study a system with nearest-neighbor interactions described
by the Hamiltonian
N
X
i,i+1 ,
H=
H
i=1
332
Topology in Physics
N,N +1 = H
N,1 . By defining
with periodic boundary conditions H
X
1 =
i,i+1 ,
H
H
i odd
2 =
H
i,i+1 ,
H
i even
we get operators given by the sum of N2 mutually commutating operators (as they
act between different pairs of particles) and hence,
Y
U1(2) = e H1(2) =
e Hi,i+1 .
i odd(even)
Let the Hamiltonian be given by

=
H
N
X
i=1
1
1
t(Ci+1
Ci + Ci Ci+1 ) + V (
ni )(
ni+1 ),
2
2
where Ci and Ci are the creation and annihilation operators and n

i = Ci Ci
the particle number operator. We denote the state in which ni particles are
on the ith lattice site and ni+1 on the (i + 1)th by |ni , ni+1 i, where ni , ni+1
{0, 1}. By applying the commutation relations [Ci , Cj ] = ij and [Ci , Cj ] =
[Ci , Cj ] = 0, and taking into account that no site can be occupied by more
i,i+1 commute. Thus
than one particle, one may verify that the two terms of H
e Hi,i+1 = e t(Ci+1 Ci +Ci Ci+1 ) e V (ni 2 )(ni+1 2 ) and
e Hi,i+1 |0, 0i = |0, 0ie 4 ,

e Hi,i+1 |1, 1i = |1, 1ie 4 ,
e Hi,i+1 |1, 0i =
e Hi,i+1 |0, 1i =
V
cosh( t)|1, 0i + sinh( t)|0, 1i e 4 ,
V
cosh( t)|0, 1i + sinh( t)|1, 0i e 4 .
(12.10)
These equations follow by expanding e t(Ci+1 Ci +Ci Ci+1 ) into its Taylor series and
noting that (Ci+1

Ci + Ci Ci+1 )|1, 0i = |0, 1i and (Ci+1
Ci + Ci Ci+1 )2 |1, 0i = |1, 0i.
The world-line perspective is very suitable for depicting this process. The x-axis
represents the periodic spatial lattice of sites, the y-axis shows the imaginarytime evolution. Each time step is divided further into two steps. In the first
step U1 acts on the system, in the second step U2 . This results in a checkerboard pattern where the shaded boxes correspond to the areas of space and time
in which particle exchange can take place, that is, on which world-lines may be
drawn across.
We want to develop an algorithm for generating all allowed world-line configurations. This is achieved by successive local changes of the world-line. As we work
333
Figure 12.8: Checkerboard pattern for the one-dimensional system. The world
lines represent the particles which may interact in the shaded squares.
in the canonical ensemble where particle number N is fixed, these local changes
must conserve N in each intermediate state, i.e., within each shaded box. Otherwise, the move must be rejected. After creating new configurations at random,
the algorithm accepts or rejects the configurations according to a sample probability which is proportional to the product of amplitudes in equation (12.9).
To derive an explicit expression for the acceptance probability, we designate each
lattice site by (i, j), where i denotes the spatial and j the temporal position,
i = 1, ..., N and j = 1, ..., 2M . As particle number must be conserved within
each shaded box, a permitted change always involves a move across an unshaded
box (Figure 12.9). In addition, the occupation number n(i, j) at each site can
only be 0 or 1. Thus, a move across a box, whose lower left-hand corner is at
(i, j), is possible if and only if |s| = 2, where
s n(i, j) + n(i, j + 1) n(i + 1, j) n(i + 1, j + 1).
s = +2 corresponds to a move from left to right, whereas s = 2 to one from
right to left.
By using the heat bath algorithm, the probability of acceptance for a proposed
new configuration is given by
R
,
P =
1+R
where R is defined as the ratio of the configurations before and after the move.
This ratio depends on the actual form of the world-line as well as on the direct
neighborhood of the local change. The occupation numbers n(i + 1, j 1) and
n(i+1, j +2) determine whether the world-line to be moved is vertical or diagonal,
and n(i1, j) and n(i+2, j) provide information about the existence of additional
world-lines in the boxes to the left and right of our unshaded one. One finds

su
sv
sv
R = tanh( t)
cosh( t) e V 2 ,
334
Topology in Physics
where
u 1 n(i + 1, j 1) n(i + 1, j + 2),
v n(i 1, j) n(i + 2, j).
Figure 12.9: Example of a local change in the world line where the solid line show
the world line before, the dashed line after the move.
Note that with these local updates, the winding number never changes during the
whole process. This means in particular that increasing the number of time steps
(i.e., cooling down our system) will not result in an increase of the macroscopic
exchanges. Thus, this algorithm is not very appropriate in order to study systems
which manifest superfluidity.
The Worm Algorithm

A rather novel computational model which is also based on PIMC is the worm algorithm (WA). It allows efficient calculations of winding numbers and one-body
density matrices for systems up to various thousand particles. The main difference to the conventional MC algorithms is that it operates on the extended
configuration space, containing both closed world-line configurations which contribute to the partition function Z, and configurations containing one open line
(worm). The worm-configurations contribute to the one-body density matrix and
are referred to as off-diagonal configurations. All topologically non-trivial modifications of the paths occur in the off-diagonal configurational space; When the
system arrives at a diagonal configuration (all world-lines closed), the number of
particles and the winding number are updated.
Consider the discrete paths X (R0 , R2 , ..., RM ), periodic in the imaginary time
interval = M . The probability density is given by
(X) = eU (X)
M
Y
0 (Rj , Rj+1 ; ),
(12.11)
j=0
QN
where 0 (Rj , Rj+1 ; ) = i=1 0 (~ri,j , ~ri,j+1 ; ) is the product of N free-particle

propagators. According to equation (12.5), 0 (~ri,j , ~ri,j+1 ; ) are Gaussian for all
335
i, j and can be used as probability weights. U describes the spatial and imaginarytime interactions between the particles and is chosen such that (X) (R, R; )
as 0.
With the WA it is possible to perform simulations in the grand-canonical ensemble where the number of particle N is not fixed; An off-diagonal configuration
contains a worm, i.e., a world line with two ends which we denote by I and M .
These two points are localized in the space-time at (~rI , tI ) and (~rM , tM ). As
usual, the paths are sampled by a Metropolis random walk but the local updates
result exclusively in stochastic movements in space-time of I and M . At times tI
and tM the conservation of the particle number N is violated and therefore, N is
both configuration- and time-dependent.
The set of local updates consists of complementary pairs which switch between
diagonal and off-diagonal configurations, as well as local changes which are selfcomplementary as they preserve the number of variables. Possible worm-movements
are:
1a) Open. In diagonal configuration X, three numbers i, j, m are selected at random. i denotes the world line to be altered and j defines the position (~ri,j , tj ) of
I. M is placed on the same line a time mt later, (~ri,j+m , tj+m ), and the beads in
between are removed. This process thus splits an arbitrarily chosen world line i
into two independent world lines i and i0 ; The first ending at I, the second starting at M . Let us denote the configuration after the move by X 0 . The acceptance
probability is
NX

eU m
Aopen = min 1,
,
V 0 (~ri,j , ~ri,j+m , m )
where U = U (X) U (X 0 ), V is the volume of the system, NX is the number
of particles before the move, and the input parameter represents the chemical
potential.
The expression for Aopen can be derived as follows:
k=j+m1
U (X)+m
(X) = e
0 (~ri,k , ~ri,k+1 ; ),
k=j
0
(X 0 ) = CeU (X )
1
T (X X 0 ) = popen
,
NX M m
Qk=j+m1
T (X 0 X) = pclose
0 (~ri,k , ~ri,k+1 ; )
.
0 (~ri,j , ~ri,j+m , m )
k=j
popen and pclose denote the probabilities that the algorithm implements the move
Open or Close. C is a constant controlling the relative statistics of the diagonal
and off-diagonal configurations and is here set to C = V M1 m , where m
< M
1
is a fixed number defining the maximal time interval. The factor NX M m is the
probability for selecting an arbitrary triplet i, j, m. In order to connect world line
336
Topology in Physics
i with world line i0 , one must generate a path with fixed initial and end points,
~ri,j and ~ri0 ,j+m . But not each of these paths is equally probable. This leads to
the quotient in the expression of T (X 0 X).
1b) Close. Consider the off-diagonal configuration with I being the j th bead on
world line i and M the (j + m)th bead on i0 . The move consists in generating a
world line of m 1 beads connecting the two worm ends. It eliminates the world
line i0 along with the worm and creates a diagonal configuration. If m > m,
the
update is rejected. For m m
the acceptance probability is
V U +m

Aclose = min 1,
e
0 (~ri,j , ~ri,j+m , m ) .
NX 0
2a) Insert. A worm of length m is created out of the vacuum in diagonal configuration. Its starting position M as well as the number of beads m m
are
picked at random and it is accepted with probability

Ainsert = min 1, eU +m .
2b) Remove. If the existing worm is of length m m,
it is proposed to remove
the whole world line connecting M and I and accepted with

Aremove = min 1, eU m .
3a) Advance. An simple modification of the configuration is done by advancing
I forward in time by a random number m of slices which has acceptance rate

Aadvance = min 1, eU +m .
In this move, I may advance past M .
3b) Recede. By erasing m consecutive beads, I may also move backwards in time
with probability of acceptance

Arecede = min 1, eU m .
The number of beads to be eliminated should not exceed the number of beads of
the worm. Otherwise, the move is a priori rejected.
4) Swap. This move operates on the off-diagonal configuration. Let (ri,j , tj ) be
(~
r ,~
r
)
m
,m
the position P
of I and select a world line k with probability Tk = 0 i,j k,j+
i
where i l 0 (~ri,j , ~rl,j+m , m
) is the normalization factor. If k contains M
at a time slice in [j, j + m]
the move is rejected. The beads ~rk,j+1 , ..., ~rk,j+m1
are removed and the position of I is connected to the bead ~rk,j+m by a randomly
generated piece of world line ~ri,j+1 , ..., ~ri,j+m1
. The new end point I is now
localized at rk,j . The final configuration, which is again off-diagonal, may possess
a winding number that does not agree with the one of the initial configuration.
337
Figure 12.10: Illustration of the Swap move: left before, right after the move.
Thus, with repeated application of the Swap move it is possible to generate all
permutations of the particles. This move is accepted with probability

i
Aswap = min 1, eU
.
k
Note that there is a high acceptance probability for the moves 3a, b and 4 as they
do not require the particles to be close together (within the region of repulsive
potential) and therefore, macroscopic exchanges occur rather frequently in this
algorithm.
Conclusion
Experimentally it is rather easy to find out whether a system is in the superfluid

state, e.g., by examine its response to moving boundaries. With the winding
number formula from section 4 we have found a way to theoretically predict the
behavior of our system.
The results for s obtained by PIMC simulations agree with the experimental
values within the statistical uncertainties (see Figure 12.11, taken from [78]). The
discrepancy above the experimental transition temperature is due to finite-size
effects. The worm algorithm from section 5 overcomes this limitation of previous
PIMC technology allowing simulations for systems of several thousand particles
and affords efficient large-scale computations of thermodynamic properties for
many-body systems.
338
Topology in Physics
Figure 12.11: Normal-component fraction along the vapor pressure curve for bulk
He. Solid line is the experimental data, the open circle are obtained by PIMC
simulations for 64 atoms.
4
13 Introduction to quantum
computing
Lukas Gamper
Supervisor: Lode Pollet
This chapter starts with the basic concepts and operations of
quantum computing. Afterwards the most important algorithm
known in quantum computing, the factorization algorithm, is derived. In the last part the basic concepts of cryptography are
introduced and the most widely used quantum key distribution
protocols are presented.
Introduction
Classical Computers
Classical computers work with bits. A bit is like a coin: there is either head or
tail, so one bit can store one binary decision. Operations are performed with so
called gates. One of the most important gates is the XOR (exclusive or) gate.
The XOR gate implements the operation x + y (mod 2) x y. If the inputs
have more than one bit, the operation is performed bitwise. The XOR operation
is its own inverse: xxy y. An other important gate is the NAND gate (not
and gate). The NAND gate is 0 if both input bits are set to 1, 1 otherwise. The
NAND gate is universal, which means, that every other gate can be implemented
as a combination of NAND. E.g. x y (x NAND (y NAND y)) NAND ((x
NAND x) NAND y).
What is a Qubit?
A qubit is the analogous concept for quantum computation. A qubit has two
eigenstates |0i and |1i, which correspond to the states 0 and 1 for a classical bit.
339
340
Topology in Physics
In contrast to classical bits, qubits can store a superposition of |0i and |1i:
|i = |0i + |1i,
(13.1)
where and are complex numbers normalized to 1 (||2 + ||2 = 1). The
special states |0i and |1i are known as computational basis states, and form an
orthogonal basis for vector spaces.
A possible realization of qubits are two orthogonal polarisations of a photon.
We can express |i as a point(, ) on a unit sphere:
|i = (cos |0i + ei , sin |1i)
(13.2)
where and are real numbers. This representation is called the Bloch sphere
representation.
If we do a binary extension of
|i =
ai 2i ,
(13.3)
k=0
we see, that we can store an infinite number of classical bits in one qubit.
This conclusion turns out to be misleading, because in a measurement of a qubit
we will only get one of the eigenvalues |0i or |1i. It turns out that only if infinitely
many identically prepared qubits were measured, we would be able to determine
and for such a qubit.
Suppose we have two qubits. Then the computational basis states are |00i,
|01i, |10i and |11i. Thus the state vector describing the two qubits is
|i = a00 |00i + a01 |01i + a10 |10i + a11 |11i,
with
|aij |2 = 1.
(13.4)
(13.5)
i,j{0,1}
Quantum gates
A quantum gate must fulfill the normalization condition. If U (|0i + |1i) =
0 |0i + 0 |1i, then 02 + 02 = 1 must hold!
Introduction to quantum computing

Single qubit gates
Pauli-X
Pauli-Y
Pauli-Z
Hadamard gate
Phase gate
/8 gate
341

0 1
1 0
0 i
i 0
1 0
0 1
1 1
1
2
1 1
1 0
0 i

1
0
i/4
0 e
Pauli-X gate The Pauli-X gate is the quantum mechanical analogy to the
classical NOT gate:

X
=
.
(13.6)
Pauli-Z gate The Pauli-Z gate leaves |0i unchanged and flips the sign of |1i
to |1i.
Hadamard gate Note that H = (X + Z)/ 2 and H 2 = I. On the Bloch

sphere, the Hadamard gate is a rotation on the sphere about the x axis by 90o ,
followed by a reflection at the x y plane.
/8 gate The /8 gate is called /8 for historical reasons. It can be written
as
i/8

e
0
i/8
T =e
.
(13.7)
0
ei/8
and the global phase is unimportant, so both entries of the Matrix have a phase
of /8.
Multi qubit gates
The standard base of an n qubits system is given by |{0, 1}n i, where the basis
states are mapped to a vector space of dimension 2n .
Example 2. A system of two qubits is usually represented as follows:

1
0
0
0
0
1
0
0

|00i =
(13.8)
0 , |01i = 0 , |10i = 1 , |11i = 0 .
0
0
0
1
342
Topology in Physics
Controlled gate Out of any gate described above, a controlled gate can be
constructed. It consists of two qubits labeled as control qubit, |ci, and target
qubit, |ti. The action of a single qubit gate is only performed if the control qubit
is set to |1i. The quantum scheme of the gate is:
The matrix representation of the controlled gate in the basis described above
is:
1 0
0 0
0 1
0 0
.
0 0
U
0 0
CNOT The controlled-NOT-gate is a quantum two qubits gate. The action
of the CNOT is given by |ci|tito|ci|t ci, i.e., if the control qubit is set to |1i,
then the target qubit is flipped, otherwise the target qubit is left unchanged. The
quantum scheme of the gate is
The matrix representation is
1
0
0
0
0
1
0
0
0
0
0
1
0
0
.
1
0
Swap gate The swap gate swaps the states of two qubits. The sequence of
gates has the following sequence of effects on a computational basis state |a, bi
|a, bi |a, a bi
|a (a b), a bi = |b, a bi
|b, (a b) bi = |b, ai.
where the is the XOR operator. The quantum scheme reads
(13.9)
(13.10)
(13.11)
343
Copying qubits
Is it possible to make a copy of an unknown quantum state? Suppose we have a
quantum machine with two slots, the data slot and the target slot:
The data slot starts in an unknown but pure quantum state, |i. This is the
state which is to be copied into the target slot. We assume that the target slot
starts in some standard pure state, |si. Thus the initial state of the copying
machine is
|i |si.
(13.12)
Some unitary evolution U now affects the copying procedure. Ideally,
|i |si U U(|i) |si) = |i |i.
(13.13)
Suppose this copying procedure works for two particular pure states, |i and |i.
Then we have
U(|i) |si) = |i |i
U(|i) |si) = |i |i.
(13.14)
(13.15)
Taking the inner product of these two equations gives

h|i = (h|i)2 .
(13.16)
But x = x2 has only two solutions, x = 0 and x = 1, so either |i = |i or

|i and |i are orthogonal. Thus a cloning device can only clone states which
are orthogonal to one another, and therefore a general quantum cloning device
is impossible. A potential quantum cloner
cannot, for example, clone the qubit
states |i = |0i and |i = (|0i + |1i)/ 2, since these states are not orthogonal.
EPR, Bell states
The name EPR pairs or Bell states comes from the people who first pointed out
the strange properties of this state: Einstein, Podolsky, Rosen and Bell. The
states are produced by the following circuit:
In
|00i
|01i
|10i
|11i
Out
(|00i + |11i)/2 |00i
(|01i + |10i)/2 |01i
(|00i |11i)/2 |10i
(|01i |10i)/ 2 |11i
344
Topology in Physics
The circuit has the following internal states:

|0 i = |xi|yi
ax (|0i + |1i) + bx (|0i |1i)
|1 i =
|yi
2
|0, yi + (1)x |1, yi
.
|2 i = |xy i
2
(13.17)
(13.18)
(13.19)
The states are called 00 , 01 , 10 and 11 because of the notation

|xy i
|0, yi + (1)x |1, yi
(13.20)
.
Quantum teleportation
Suppose Alice and Bob share an EPR pair, each having one qubit of the EPR
pair. Alice can only communicate over a classical wire with Bob. How can Alice
deliver a qubit |i to Bob?
Alice can modulate the qubit |i = |0i + |1i with her half of the EPR pair
with the following circuit:
.
It has the following states:
|0 i = |i|00 i
1
= [|0i(|00i + |11i) + |1i(|00i + |11i)]
2
1
|1 i = [|0i(|00i + |11i) + |1i(|10i + |01i)]
2
1
|2 i =
[(|0i + |1i)(|00i + |11i) + (|0i |1i)(|10i + |01i)]
2
1
=
[|00i(|0i + |1i) + |01i(|1i + |0i)
2
+ |10i(|0i |1i) + |11i(|1i |0i)] .
(13.21)
(13.22)
(13.23)
(13.24)
(13.25)
(13.26)
Then Alice performs a measurement and sends the result, one out of |00i, |01i,
|10i and |11i, to Bob. Depending on Alices measurement, Bob has one of the
345
following states:
00
01
10
11
|0i + |1i
|1i + |0i
|0i |1i
|1i |0i.
(13.27)
(13.28)
(13.29)
(13.30)
This fact prevents Alice and Bob from communicating faster than the speed of
light. If Bob does not know the result of Alices measurement, his qubit does not
contain any information.
Algorithms
In this chapter we will derive an algorithm to factorize with a quantum computer.

To achieve this, we will extend the quantum Fourier transformation algorithm to
a phase estimation algorithm. This will lead us to a solution of the order finding
problem. The order finding algorithm can be used as the core of a factorizing
algorithm.
In this chapter the following notation is used:
|0 0i, . . . , |1 1i |0i, . . . , |2n 1i
(13.31)
Quantum Fourier transformation

The classical discrete Fourier transformation takes as input a vector of complex
numbers, x0 , . . . , xN 1 , with length of the vector N . The output is a vector of
complex numbers y0 , . . . , yN 1 , defined by
N 1
1 X
yk
xj e2ijk/N .
N j=0
(13.32)
The quantum Fourier transformation is exactly the same. It is defined on an

orthonormal basis |0i, . . . , |N 1i as a linear operator acting as follows:
N 1
1 X 2ijk/N
|ji
e
|ki.
N k=0
(13.33)
The action on an arbitrary state may be written as a base transformation,

N
1
X
j=0
xj |ji
N
1
X
yk |ki,
(13.34)
k=0
where the amplitudes yk are the discrete Fourier transforms of the amplitudes xj .
346
Topology in Physics
In the following we take N = 2n , where n is some integer, and the basis

|0i, . . . , |2n 1i is the computational basis for an n qubit quantum computer.
It is helpful to write the state |ji using the binary representation j = j1 2n1 +
j2 2n2 + . . . + jn 20 . It is also convenient to adopt the notation 0.jl jl+1 . . . jm to
represent the binary fraction jl 21 + jl+1 22 + . . . + jm 2m+l1 . This leads to the
product representation
(|0i + e2i0.jn |1i)(|0i + e2i0.jn1 jn |1i) (|0i + e2i0.j1 j2 ...jn |1i)
.
2n/2
(13.35)
The product representation makes it easy to derive an efficient circuit for the
quantum Fourier transform. We need the Hadamard gate and the gate Rk

0
1
(13.36)
Rk
.
k
1 e2i/2
|j1 , . . . , jn i
The resulting circuit is shown below:
The swap operations, used to reverse the order of the circuit, are omitted from
the circuit for clarity.
On the first wire we use n gates, on the second we use n1 gates, up to 1 gate
in the last wire. In this way we get n+(n1)+. . .+1 = n(n+1)/2 required gates
plus the gates involved in the swap. At most n/2 swaps are required, and each
swap can be accomplished using three controlled NOT gates, as described in the
introduction. Therefore, the circuit provides a (n2 ) algorithm for performing
the quantum Fourier transformation.
In contrary to the quantum Fourier transformation, a classical algorithm for
computing the discrete Fourier transformation on 2n elements such as the Fast
Fourier transformation, needs (n2n ) gates to compute the discrete Fourier transformation. This is an exponential speedup! But there is no way of determining
the Fourier transform amplitudes of the original state by measurements.
Example: Three-qubit quantum Fourier transform. The circuit of the three qubit
quantum Fourier transformation:
347
Here S is the phase gate and T the /8 gate. As a matrix the quantum Fourier
transformation
in this instance may be written out explicitly, using = e2i/8 =
i, as
1 1 1 1 1 1 1 1
1 2 3 4 5 6 7
1 2 4 6 1 2 4 6
3
6
4
7
2
5
1
1
.

(13.37)
4
4
4
1 4
8
1 5 12 7 14
1 6 3
1 6 4 2 1 6 4 2
1 7 6 5 4 3 2
Phase estimation
Suppose |ui is an eigenvector for the unitary operator U with the eigenvalue e2i ,
where is unknown. The goal of the phase estimation algorithm is to estimate
.
The quantum phase estimation procedure uses two registers. The first contains t qubits initially in the state |0i. How to choose t depends on two things:
the accuracy we want to achieve for our estimate for and the probability we
want the phase estimation procedure to be successful. The second register starts
with the state |ui, and contains as many qubits as necessary to store |ui.
The overall scheme of the algorithm is
The phase estimation is performed in three stages. First, we apply the Hadamard
transformation to the first register, followed by an application of controlled-U operation on the second register, with U raised successively to higher powers of two.
The final state of the first register is
1
2t/2
2i2t1
(|0i + e
2i2t2
|1i)(|0i + e
2i20
|1i) (|0i + e
|1i) =
N 1
1 X
2t/2
k=0
e2ik |ki.
(13.38)
The second state of the phase estimation is to apply the inverse quantum Fourier
transformation on the first register. This can be done in (t2 ) steps.
The third and final stage of the phase estimation is to read out the state of the
first register by doing a measurement in the computational basis.
The heart of this procedure is the ability of the inverse Fourier transformation to
348
Topology in Physics
perform the transformation

t
2 1
1 X
2t/2
e2ij |ji|ui |i|ui,
(13.39)
j=0
where |i
denotes a state which is a good estimate for when measured.
Example 3. Suppose may be expressed exactly in t bits, as = 0.1 . . . t . The
first register after the first step may be rewritten as
1
2t/2
(|0i + e2i0.t |1i)(|0i + e2i0.t1 t |1i) (|0i + e2i0.1 1 t |1i).
(13.40)
Comparing this equation with the product representation of the Fourier transform, we see that the output state of the second stage is the product state
|1 . . . t i. A measurement in the computational basis therefore gives us exactly.
Order - finding
For positive integers x and N , x < N , with no common factor, the order of x
modulo N is defined to be the least positive integer r, such that xr = 1 (mod N ).
Example 4. Let x = 5 and N = 21. Then 56 = 15625 = 744 21 + 1, and thus
the order of x modulo N is 6.
The quantum algorithm for order finding is just the phase estimation algorithm applied to the unitary operator
U|yi |xy mod N i,
(13.41)
r1

1 X 2isk/r k
|us i
e
x mod N ,
r k=0
(13.42)
with y {0, 1}L . The states
for an integer 0 s r 1 are eigenstates of U, since

r1

1 X 2isk/r k+1
U|us i
x
mod N = e2is/r |us i.
e
r k=0
(13.43)
The phase estimation procedure allows us to obtain the corresponding eigenvalue

exp(2is/r).
349
There are two important requirements for us to be able to use the phase estimation procedure: we must have efficient procedures to implement a controlledj
u2 operation for any integer j, and we must be able to efficiently prepare an
eigenstate |us i with a non-trivial eigenvalue, or at least a superposition of such
eigenstates.
The first requirement is satisfied by using a procedure known as modular
exponentiation. We want to compute the transformation
t1
|zi|yi |ziU zt 2 . . . U z1 2 |yi

E
zt 2t1
z1 20
x y mod N
= |zix
z
= |zi|x y mod N i.
(13.44)
(13.45)
(13.46)
The second requirement is a little trickier: preparing |us i requires that we know
r, so this is out of question. Fortunately, there is a clever observation which
allows us to circumvent the problem of preparing |us i, which is that
r1
1 X
|us i = |1i,
r s=0
(13.47)
so we get to |us i for free.

The continued fraction expansion The reduction of order-finding to phase
estimation is completed by describing how to obtain the desired answer, r, from
the result of the phase estimation algorithm, s/r. We only know to a finite
number of bits, but we know a priori that it is a rational number.
The idea of the continued fraction algorithm is to describe real numbers in
terms of integers alone, using expressions of the form
[a0 , . . . , aM ] a0 +
1
a1 +
a2 +
(13.48)
1
...+ a 1
M
where a0 , . . . , aM are positive integers. It is clear that the continued fraction

algorithm terminates after a finite number of split and invert steps for any
rational number, since the numerators are strictly decreasing (e.g 31, 5, 3, 2, 1
in the example below). The decomposition can be done in O(L3 ) operations.
Example 5. Suppose we are trying to decompose 31/12 as a continued fraction.
The first step of the continued fraction algorithms is to split 31/13 into its integer
and fractional part
5
31
=2+ .
(13.49)
13
13
350
Topology in Physics
Next, we invert the fractional part, obtaining

1
31
1
= 2 + 13 = 2 +
.
13
2 + 35
5
(13.50)
Continuing in this way, we come to the final continued fraction representation of

31/13:
31
1
.
(13.51)
=2+
13
2 + 1+ 1 1
1+ 1
2
Performance There are two possibilities for the order-finding algorithm to

fail: First, the phase estimation procedure might produce a bad estimate for s/r,
or, s and t might have a common factor, in which case the number r0 returned
by a continued fraction algorithm is a factor of r, and not r itself. This can be
solved by repeating the phase estimation continuous fraction procedure several
times.
Factorizing
The factorizing problem tuns out to be equivalent to the order-fining problem.
The reduction of factorizing to order finding proceeds in two steps: First, we
show that we can compute a factor of N , if we can find a non-trivial solution to
x2 6= 1 (mod N ) to the equation x2 = 1 (mod N ). The second step is to show
that a randomly chosen y co-prime to N is quite likely to have an order r which
is even, and such that y r/2 6= 1 (mod N ), and thus x y r/2 is a non-trivial
solution to x2 = 1 (mod N ). These two steps are embodied in the following
theorems (for proof see [81]).
Theorem 6. Suppose N is an L bit composite number, and x is a non-trivial
solution for the equation x2 = 1 (mod N ) in the range 1 x N , that is,
neither x = 1 (mod N ) nor x = N 1 = 1 (mod N ). Then at least one of
gcd(x 1, N ) and gcd(x + 1, N ) is a non-trivial factor of N that can be computed
using O(L3 ) operations.
Theorem 7. Suppose N = p1 1 . . . pmm is the prime factorization of an odd composite positive integer. Let x be an integer chosen from a uniform distribution,
with the restriction that 1 x N 1 and x is co-prime to N . Let r be the
order of x modulo N . Then
p(r is even and xr/2 6= 1
(mod N )) 1
1
.
2m
(13.52)
Now we can give an algorithm which returns a non-trivial factor of any composite N with high probability. All steps in the algorithm can be performed
efficiently on a classical computer except an order-finding subroutine. By repeating the procedure we may find a complete factorization of N .
351
The factorizing algorithm The algorithm takes a composite number N as

input and returns in O((log N )3 ) a non trivial factor of N with probability O(1).
1. If N is even, return the factor 2
2. Determine whether N = ab for integers a 1 and b 2, and if so return
the factor a
3. Randomly choose x in the range 1 to N 1. If gcd(x, N ) > 1 then return
the factor gcd(x, N )
4. Use the order-finding subroutine to find the order r of x modulo N
5. If r is even and xr/2 6= 1 (mod N ) then compute gcd(xr/2 1, N ) and
gcd(xr/2 + 1, N ), and test if one of these is a nontrivial factor, returning
the factor if so. Otherwise the algorithm fails.
Example: Factorizing 15. The use of order-finding, phase estimation, and continued fraction expansion in the quantum factorizing algorithm is illustrated by
applying it to factorizing N = 15. First we choose a random number which has
no common factors with N ; suppose we choose x = 7. Next we compute the order
r of x with respect to N , using the quantum order-finding algorithm. Begin with
the state |0i|0i and create the state
t
2 1

1 X
1
|ki|0i =
|0i + |1i + |2i + . . . + 2t 1 |0i,
2t k=0
2t
(13.53)
by applying t = 11 Hadamard transformations on the first register. Next, compute f (k) = xk (mod N ). Leaving the result in the second register, we get
t
2 1

1 X k
1
|ki x mod N = (|0i|1i + |1i|7i + |2i|4i + |3i|13i + . . .) . (13.54)

2t k=0
2t
We now apply the inverse Fourier transformation F T to the first register and
measure it. Since no further operation is applied to the second register, we can
assume that the second register is measured, obtaining a random result from 1,
7, 4 or
q 13. Suppose we get 4; this means the state input to F T would have
been
1
2t
(|2i + |4i + |10i + |14i + . . .) |4i (from the equation above we take the
states wherePthe second bit is in the state |4i). After applying F T , we obtain
some state
l l |li, with the possible measurements 0, 512, 1024, 1536, each
352
Topology in Physics
with probability almost exactly 1/4. Suppose we obtain l = 1536 from the
measurement; Computing the continued fraction expansion thus gives
(2t
1536
1
=
,
2048)
1 + 13
(13.55)
so that 3/4 occurs as a convergent in the expansion giving r = 4 as the order of

x = 7. By chance r is even, and moreover
xr/2 = 72 6= 1
(mod 15),
(13.56)
so the algorithm works; computing the greatest common divisor gcd(x2 1, 15) =
3 and gcd(x2 + 1, 15) = 5 tells us that 15 = 3 5.
Cryptography
In cryptography a fixed terminology is used: The communication parties are

always called Alice and Bob, where Alice sends a message to Bob and Eve tries
to eavesdrop the communication between.
Cryptography is the art of enabling two parties to communicate in private.
In our days nearly everybody is using cryptographic methods by using ssh, managing a bank account by the internet or doing e-commerce. First, the classical
cryptographic methods are described:
One Time Pad
Alice and Bob have an n-bit secret key string. Alice encodes her n-bit message
by adding message and key together, and Bob decodes by subtracting the key
to invert the encoding. Normally the bitwise XOR is used to add and subtract,
because XOR is its own inverse. The One Time Pad can be demonstrated in
the following scheme:
Alice
y =x+e
Bob
y
x = y e
Example 6. Encode the word QUANTUM with the key GQYRWAD. We use
only upper case letters, so we can encode our letters in numbers from 1 to 26.
The letter A has the code 1, the letter Z has the code 26. All operations are done
in a Cyclic Group with 26 elements (that means that X + D = 24 + 4 = 2 = B).
So the encoding can be done as:
Original message
Encryption key
Encrypted message
Received message
Decryption key
Decrypted message
Q U
+ +
G Q

W L
W
G
353
A N T
+ + +
Y R W

Y F Q
U M
+ +
A D

U P
Public Channel
L
Q
Y
Y
F
R
Q
W
U
A
P
D
Even if Eve has an infinite amount of computation power she cannot decrypt
the message. She is able to get all messages with 7 characters, but she is not able
to get any informations out of it.
The major problem of the One Time Pad is the distribution of the key bits.
In particular, the One Time Pad is only secure if the number of key bits is at
least as large as the size of the message being encoded, and the key can not be
reused, else there will be a correlation between the messages!
Suppose we have a community of n people, which want to communicate pairwise.
Then everybody needs n 1 keys!
This method is used if security is very important, for example in the Cold
War to encrypt the telephone hot line between Moscow and Washington. But
there where whole airplanes full of hard disks to provide the keys.
Public Key Cryptography
The main idea of the public key cryptosystem is to have two keys, a public key
and a secret key.
Example of Public Key Cryptography. A very easy example of a public key crpytosystem is the postbox in front of your house. Everybody can leave you a
message, but only you (or anybody, who owns a key to the box) can get the messages. This shows that the concept of a public key cryptosystem is not restricted
to the computer technology.
The most used cryptosystem, the RSA, is now introduced.
354
Topology in Physics
RSA
RSA is the most widely used cryptosystem, named by the initials of its creators:
Rivest, Shamir and Adelman. The security of RSA is based on the difficulty of
factorizing on a classical computer. The scheme of RSA cryptography:
1. Select two large prime numbers, p and q
2. Compute the product n pq
3. Select randomly a small odd integer, e, that is co-prime to (n) = (p
1)(q 1)
4. Compute d, the multiplicative inverse of e, modulo (n)
5. The RSA public key is the pair P = (e, n). The RSA secret key is the pair
S = (d, n)
. This leads to the following scheme:
Alice
Bob
Generate primes p and q

npq
(n) (p 1)(q 1)
Select e co-prime to (n)
d e1 (mod (n))
D(E(M )) E(M )d (mod n)
n, e
E(M )
Plaintext
M {1, . . . , m 1}
E(M ) M e (mod n)
Suppose we have two parties Alice and Bob. Bob wants to send the message
M to Alice using the RSA scheme. Assume that M has only blog nc bits, as
longer messages may be encrypted by breaking M up into blocks of most blog nc
bits and then encrypt the blocks separately. Bob encrypts the message with
M E(M ) = M e
(mod n).
(13.57)
Now Bob can transmit E(M ) to Alice. Alice can quickly decrypt the message
with
E(M ) M = D(E(M )) = E(M )d (mod n).
(13.58)
To verify this, note that ed = 1 (mod n) and thus ed = 1 + k(n) with k N.
To proof RSA, we need the following theorems (for proof see [82]):
355
Eulers generalization of Fermats little theorem.

if a is co-prime to n a(n) = 1
(mod n).
(13.59)
proof of RSA. If M is co-prime to n, then from Eulers generalization of Fermats

little theorem follows that M k(n) = 1 (mod n) and thus
D(E(M )) =
=
=
=
=
E(M )d (mod n)
M ed (mod n)
M 1+k(n) (mod n)
M M k(n) (mod n)
M (mod n).
(13.60)
(13.61)
(13.62)
(13.63)
(13.64)
If M is not co-prime to n, assume p divides M and q does not divide M (the other
cases are analogous). Because p divides M , we have M = 0 (mod n) and thus
M ed = 0 = M (mod n). Because q does not divide M we have M q1 = 1 (mod n)
by the theorem above and thus M (n) = 1 (mod n), since (n) = (p 1)(q 1).
Using ed = 1 + k(n), we see that M ed = M (mod q) and it follows that we must
have M ed = M (mod n).
For further informations look also at [83]
Example 7. Encode the letter Q using p = 3 and q = 11, thus n = pq = 33
and (n) = (p 1)(q 1) = 20. We choose a random odd integer e 7 that is
co-prime to 20. The multiplicative inverse of 7 is 3 (mod 20). So the public key
is P = (7, 33) and the secret key is S = (3, 33).
If we encode our letters in numbers from 1 to 26, the letter Q has the
representations 17. So
E(17) = 177 = 8 (mod 33)
D(20) = 203 = 17 (mod 33).
(13.65)
(13.66)
So if we use the quantum factorizing algorithm to factorize n we can now

break RSA:
Suppose Eve could factorize n = pq, extracting p and q, and thus is given a
means for efficiently computing (n) = (p1)(q1). It is then an easy matter for
Eve to compute d, the inverse of e modulo (n), and thus completely determine
the secret key (d, n).
Quantum cryptorgraphy
Quantum cryptography is only used to distribute the keys for a classical cryptographic protocol. Quantum key distribution is provably secure, and therefore
private key bits can be created between two parties over a public channel. The
key bits can then be used to implement a classical private key cryptosystem.
356
Topology in Physics
The basic idea of quantum key distribution is the fundamental observation

that Eve cannot gain any information from the qubits transmitted from Alice to
Bob without disturbing their state and because of the following
Proposition 18. Information gain implies disturbance in any attempt to distinguish between two non-orthogonal quantum states, information gain is only
possible at the expense of introducing disturbance to the signal.
Proof. Let |i and |i be the non-orthogonal quantum states Eve is trying to
obtain information about. She uses unitary intersections of the states |i and
|i with a prepared state |ui. Assuming that this process does not disturb the
states, in the two cases one obtains
|i|ui |i|vi
|i|ui |i|v 0 i.
(13.67)
(13.68)
Eve would like |vi and |v 0 i to be different so that she can acquire information
about the identity of the state. However, since inner products are preserved under
unitary transformations, it must be that
hv|v 0 i h|i = hu|ui h|i .
(13.69)
and since u and v are unitary

hv|v 0 i = hu|ui = 1,
(13.70)
which implies that |vi and |v 0 i must be identical up to a global phase. Thus,
distinguishing between |i and |i must inevitably disturb at least one of these
states.
This is a weaker statement of the fact that copying of qubits is not possible!
The proposition leads to the following three quantum key distribution protocols:
The protocol from Bennett and Brassard (BB84) Alice generates two
strings a and b, each of (4 + )n random classical bits. She then encodes these
strings as a block of (4 + )n qubits,
(4+)n
|i =
|ak bk i,
(13.71)
k=1
where ak is the k th bit of a (and similarly for b), and each qubit is one of the four
states
|00 i
|10 i
|01 i
|11 i
=
=
=
=
|0i
|1i
|+i = (|0i + |1i)/ 2
|i = (|0i |1i)/ 2.
(13.72)
(13.73)
(13.74)
(13.75)
357
The protocol is described in the following scheme:

Alice
Bob
a, b = rnd 2n-bit strings

N
|i = 2n
k=1 |ak bk i
(ak = k th bit of a)
Publish b
|i
ACK
Generate 2n-bit string b0

Measure each qubit in
X or Z bases determined by b0
a0 measurement {0, 1}
Keep only those {a, a0 } for which b = b0

Select n-bits at random
Check if too many are wrong,

if so, retry
Use the One Time Pad with the remaining n-bits
The effect of this resource is to encode a in the basis X or Z, as determined

by b. Alice then sends |i to Bob, over their public quantum communication
channel. Bob receives E(|ih|), where E describes the quantum operation due
to the combined effect of the channel and Eves actions. He then acknowledges
that he has received the message. At this point, Eve has no knowledge of what
basis she should have measured in to eavesdrop on the communication; she can
only guess, and if her guess was wrong, then she disturbed the state received by
Bob.
For Bob E(|ih|) contains also no information at this point, because he does
not know anything about b either. Nevertheless, he creates a random (4 + )n bit
string b0 and measures each qubit in the X or Z basis determined my b0 . He saves
the measurement in the classical bit string a0 . After that, Alice publishes b, and
by discussion over a public channel, Bob and Alice discard all bits in {a0 , a} except
those for which corresponding bits of b0 and b are equal. Their remaining bits
satisfy a0 = a, since for these bits Bob measured in the same basis Alice prepared
in. It is important, that Alice does not publish b until after Bob has confirmed
the reception of Alices qubits. For simplicity in the following explanation, let
Alice and Bob keep just 2 n bits of their result.
Now Alice selects n bits (of their 2n bits) randomly and publicly announces
the selection. Bob and Alice then publish and compare the values of these check
bits. If more than t disagree, then they abort and retry the protocol from start.
358
Topology in Physics
t is selected such that if the test passes, then they can apply information reconciliation on privacy amplification algorithms to obtain m acceptably secret key
bits from the remaining n bits.
If the bits are accepted, the n bits can be used as a key in a classical decrypting protocol like the One Time Pad.
The protocol from Bennett (B92) The BB84 protocol can be generalized
by using other states and bases. In fact, a particular simple protocol exists in
which only two states are used. The scheme reads:
Alice
Bob
a = rnd 2n-bit string

N
|i = 2n
k=1 |ak i
(ak = k th bit of a)
|i
Generate 2n-bit string a0

Measure each qubit in
X or Z bases determined by a0
b measurement {0, 1}
Keep only those {a, a0 } for which b = 1

Alices key is a
Bobs key is 1 a0
if so, retry
Use the One Time Pad with the generated key

To understand the differences to the protocol before, it is sufficient to consider
what happens to a single bit at a time.
Suppose Alice prepares one random classical bit a, and, depending on the
result, sends Bob
(
|0i
if a = 0
|i =
(13.76)
|0i+|1i
if a = 1
2
Depending on a random classical bit a0 generated by Bob, hemeasures in the Z
basis (|0i, |1i) if a0 = 0, or in the X basis (|i = (|0i |1i/ 2) if a0 = 1. From
this measurement, he obtains the result b0 , which he publicly announces. Alice
and Bob then conduct a public discussion and keep only those pairs {a, a0 } for
which b = 1. Note that when a = a0 , then b=0 always. Only if a0 = a 1 Bob
will obtain b = 1, which occurs with probability 1/2. The final key for Alice is a
and 1 a0 for Bob.
One possible physical implementation of this two protocols are polarized pho-
359
tons in two orthogonal directions.

The EPR protocol (Eckert 93) Suppose Alice and Bob share a set of n
entangled pairs of EPR pairs in the state
|00i + |11i
(13.77)
The scheme reads:

Alice
Bob
Alice and Bob share a set of 2n EPR pairs in the state |00 i
b = rnd 2n-bit string
Measure each half of
the EPR pairs in X or Z
bases determined by b
b, b0
a measurement {0, 1}
b0 = rnd 2n-bit string

Measure each half of
the EPR pairs in X or Z
bases determined by b0
a0 measurement {0, 1}
Keep only those {a, a0 } for which b = b0


if so, retry
Use the One Time Pad with the remaining a = a0 as key

Alice and Bob then select a random subset of the EPR pairs, and test to see
if they are real EPR states.
Since the EPR protocol is symmetric, Alice and Bob perform identical tasks on
their qubits. Suppose that Alice prepares a random classical bit b, and according
to it, measures her
half of the EPR pair in either the |0i, |1i basis, or in the basis
|i = (|0i |1i)/ 2, and obtains a. Let Bob identically measure in his randomly
chosen basis b0 , where he obtains a0 . Now they exchange b and b0 over a public
classical channel, and keep as their key only those {a, a0 } for which b = b0 .
Conclusion
We have seen that quantum computers work with quantum bits (qubits), an analogous concept to the classical bit. A qubit can not be copied, but as long as you
do not disturb the qubit, it is possible to save a tremendous amount of information in one qubit. Furthermore we saw an algorithm to factorize a composed
360
Topology in Physics
number, with an exponential speedup with respect to the classical algorithm.

In the last part, cryptographic protocols and the algorithms for quantum key
distribution were introduced.
14 Topologically protected
quantum computing
Bojan Skerlak
Supervisor: Helmut Katzgraber
The two main problems arising in quantum computation are
preparation of a coherent state and accurate interaction with
a qubit. These problems appear due to decoherence and the
fragile nature of any quantum mechanical system. Instead of
fighting those problems with clever circuit design and error correction software, one can theoretically build hardware that exploits topology in order to protect information. Systems with
Abelian anyons allow us to encode robust qubits and braiding
non-Abelian anyons allows us to process quantum information
safely. Theory is far ahead in this field and experiments have
not been realised yet. Still, it is a very interesting and promising
field as fractional statistics have been observed in the context of
strongly correlated systems such as fractional quantum hall liquids and proposed implementations promise very high fidelity.
Introduction
Richard Feynman stated in 1982 that quantum systems cannot be efficiently simulated on classical computers, i.e, without exponential slowdown [84]. Three
years later, David Deutsch described the universal quantum computer as a totally new kind of computer with higher computational power [85]. In 1994, Peter
Shor developed his famous factoring algorithm [86] which theoretically allows to
break most cryptographic systems easily, because the factoring problem scales
algebraically instead of exponentially with the number of digits. As an example, a 1024-bit RSA code can be cracked in reasonable time using 2051 qubits
(generally: an L bit code needs 2L + 3 qubits and O[L3 ln(L)] gates using the
361
362
Topology in Physics
circuit described in Refs. [87, 88]). This attracted considerable attention to the
relatively new field of quantum computing. From then on, it was certain that
quantum computer process certain algorithms far faster than any classical computer and that they can be of great use. New algorithms that profit from the
special features of quantum computers were found [89] and first experiments were
successfully made [90]. On the theoretical side, error correction codes were developed that allow to improve the reliability of given hardware. Those codes were
found to demand an accuracy of the hardware that is not reachable even today.
In the search for fundamentally better hardware which is intrinsically protected
against errors, topology plays an important role. Therefore, we have to find systems that show topological behaviour such as ones with fractionalized excitations.
Those excitations, which are only possible in two dimensions, are called anyons
and can be mathematically described using the braid group. We then also find
a way to not only protect our qubits against errors, which is done with Abelian
anyons, but also protect them during the computation. For this, we need nonAbelian anyons that have been shown to exist theoretically but lack experimental
verification.
Quantum computers: the standard model

The fundamental reason why quantum computers are more powerful than classical computers is that they do not use the classical bits that are restricted to
being in state 0 or 1 but quantum bits (qubits), which can also be in a quantum
superposition of the two states: any superposition |0i + |1i with , C and
||2 + ||2 = 1 is allowed. A quantum computer thus consists of n qubits that
correspond to a 2n -dimensional Hilbert space. Before the calculation, the qubits
are prepared in some well-known state |0in and afterwards, they are measured in
the computational basis |0i, |1i. Any computation consists of a sequence of gates
acting on the n qubits. All possible gates can be approximated with arbitrary
accuracy using a dense subgroup of SU (2n ). Such a set of gates is called universal
for quantum computation. A popular choice for an universal set consists of the
two one-qubit gates N OT SU (2), S GL(2, C) [91] and of a special two-qubit
gate, the controlled-not gate CN OT SU (4) [92]. The latter flips the target
qubit if and only if the control qubit is in state |1i.
1 0 0 0

0 1 0 0
0 1
1 0
1 0
N OT :=
, S :=
, Z :=
, CN OT :=
0 0 0 1
1 0
0 i
0 1
0 0 1 0
Topologically protected quantum computing
363
Experimental approaches to non-topological quantum computation

Up to date, no quantum computer with more than ten qubits has been operated.
The three most promising and successful approaches to traditional quantum computation as well as their main problems are shortly presented here.
NMR quantum computing
Figure 14.1: The molecule used by IBM research [93]: qubits are encoded in the
spin states of nuclei.
The nuclear magnetic resonance (NMR) approach uses the spin states of nuclei
within a molecule in an external magnetic field as qubits. As the atoms can
be distinguished by their resonance frequencies due to the different chemical
environment, every spin-1/2 nucleus acts as a qubit. One-qubit operations are
easily implemented with radio frequency fields. The coupling between qubits is
achieved via the naturally occurring spin-spin coupling. The main advantage
of this system is that NMR technology is relatively mature so the precision of
those interactions is very high (error rate . 103 ). In 2001, IBM researchers
implemented Shors algorithm in a 7-qubit NMR quantum computer using 1018
molecules [93]. After preparing the molecules, they factored 15 = 3 5 in four
hours of computation. The big problem is that one has to create a pure state of
many molecules as the signal coming from one molecule is too small. The system
is operated at room temperature and therefore, thermal fluctuations cause a very
small signal to noise ratio. Preparing large numbers of molecules in a desired
configuration is hard to achieve using appropriate pulse sequences. Distillation
processes [94] are used to deal with that problem but are believed not to work
with large molecules [95]. The number of qubits is limited by the size of the
molecule as each nucleus represents a qubit. The spin-spin interaction is shortranged and thus very distant spins cannot be entangled. As a consequence of
those problems, NMR implementations will not scale beyond 100 qubits [96].
364
Topology in Physics
Trapped ion quantum computers

Ions or charged atomic particles are confined using electromagnetic fields and
cooled down so that their motion is governed by quantum mechanics. Different
states of the ions are used as qubits. Lasers are used for both single-qubit operations and two-qubit logical gates. As the ions interact via the Coulomb force,
collective motion states are available. These can be excited and de-excited to
entangle different qubits. Creating arrays of ion traps, one could in principle
scale the system to large number of qubits. Trapped ion quantum computing
already has had some successes: the first CNOT gate has been realised this way
in 1995 [90] and in 2000, four particles have been entangled [97]. Controlling
the motion of ion traps is very difficult, they are very susceptible to decoherence
errors. These systems are also not likely to work with more than 100 qubits
[96], which at the moment still is out of reach.
Solid state quantum computing: Quantum Dots
By creating quantum wells with semiconductor hetero-structures [98], electrons
can be confined to a very small region ( 2 10 nm). Their spin degrees of
freedom, which have very long decoherence times ( ns [99]), are used as qubits.
Interactions between qubits can either be achieved by changing the tunneling amplitude between neighbouring electrons or using lasers. The latter version needs
the structure to be built inside a cavity whose modes can in principle be used to
entangle even distant qubits [100]. Those interactions can be performed quickly
( THz) [101]. Although fully controlling a qubit and performing accurate gates
still is a formidable task, solid-state implementations are thought to be scalable
to more than 100 qubits [96].
What is the problem?

The last section shows that the scalability is a crucial feature in the search for
useful approaches [95]. But not only the number of qubits is important, we also
have to control them well. Controlling qubits includes preparation of a coherent
state, performing accurate operations and reliable read out of the result. Each
step is susceptible to errors because of the very nature of quantum mechanics. A
qubit is, as all quantum mechanical systems are, very sensitive to disturbance by
the environment. We cannot hope to completely isolate the qubits which means
that they interact with the environment and thus suffer from decoherence. On
the other hand, we have to be able to perform accurate interactions without introducing additional errors during calculation. Implementation of non-destructive
measurement brings several additional problems such as the need for a supply
of qubits in a well-known state. All these effects can accumulate and end up in
making the calculation intractable.
365
On classical computers instead, we are able to both store and process information with a remarkable reliability. The mean time to failure for common
hard-disks is 106 h and processors operate easily with GHz frequencies. This
is the result of decades of development of both soft- and hardware but also due
to material properties that allow us to build such reliable hardware. Therefore,
we assume that we are able to store classical information perfectly and are able
to perform classical gates with arbitrary accuracy and sufficiently high speed.
Moores law supports this assumption [102] but it will not hold forever: a limit
on classical computer speed is visible since the hardware cannot be scaled to arbitrary small size.
The aforementioned implementations show much higher error rates ( 103 )
which means that after 1000 operations, the information is most certainly lost.
We therefore have to find a way to operate a system albeit its vulnerability to
errors.
Fault-tolerant quantum computation

A device that works effectively even when its elementary components are imperfect is said to be fault-tolerant [103].
Building such a system can in principle be done on either a software or hardware level. Classical hardware is very reliable and error correction software that
further improves the accuracy can be implemented with reasonable overhead. Using qubits, the situation is far more complicated: creating quantum mechanical
hardware that works in a noisy environment is a subtle task. The types of errors
occurring is not only restricted to flipping bits but can be an entanglement with
any state of the environment. Thus, correcting such errors is far more difficult.
We have to know what kind of error occurred without actually measuring the
encoded information which would lead to a collapse of the state. Albeit those
problems, quantum error correction codes (QECC) that protect the qubits and
the operations on them exist [104].
Quantum error correction
Suppose we have n qubits, described by a Hilbert space of dimension dimH = 2n .
The Pauli group is the group of operators

0 1
0 i
1 0
X :=
, Y :=
, Z :=
.
1 0
i 0
0 1
The n-qubit Pauli group Gn is then defined as the Pauli group acting on
the tensor product: Gn := {I, X, Y, Z}n . Let {M1 , M2 , ..., Mnk } Gn be inde-
366
Topology in Physics
pendent mutually commuting Pauli operators ([Mi , Mj ] = 0 i, j). They satisfy

Mi2 = I and define an Abelian subgroup S Gn called the stabilizer group.
We then define a subspace of k < n logical qubits with dimHcode = 2k as the
simultaneous eigenspace of the n k operators in S: Mi |i = |i i. These
logical qubits are no more located at one physical qubit but are entangled states
of multiple physical qubits. The stabilizer operators are chosen in a way that
every error operator that we want to be able to correct anticommutes with at
least one Mi . Thus such an error E Gn takes Hcode unambiguously to mutually
orthogonal subspaces, namely the eigenspaces corresponding to the eigenvalue
1 of the corresponding Mi s. We then measure all stabilizer operators, which is
called syndrome measurement, and obtain all information needed about the error
E. By performing E 1 we can correct our data.
We have to keep in mind that measurement of an operator acting on the qubits
destroys our computational state so we want to measure the syndrome without
measuring the data. This is possible and is accomplished by cleverly entangling the state with ancillary qubits. If we measure them, the code subspace is
preserved and we get all information needed about E without acquiring any information about the quantum information encoded in our device [105]. A similar
procedure is used to protect operations on encoded qubits. If we can construct
a set of fault-tolerant universal gates, we can overcome the deficiencies of the
hardware [106].
Example: 2 qubit code
Suppose we want to protect one logical state using two qubits. Define two stabilizer operators M1 := X X and M2 := Z Z. The protected space is the
simultaneous eigenspace consisting of |i+ := 12 (|00i + |11i). Thus, the protected subspace has dimension 2n2 = 1. We see that this is not really a code in
the sense described above since we cannot encode a qubit but it shows the main
ideas. The kind of errors we can correct is restricted: only one qubit is allowed to
MX
MZ
I I
+
+
X I
Y I
Z I
+
Table 14.1: Syndrome measurement detects the error unambiguously.

be damaged. Assume we damage the first qubit with an error E {I, X, Y, Z}I.
Measuring the operators of the stabilizer group, M1 and M2 , via ancillary states,
we obtain the syndrome (see Table 14.1), can distinguish between different errors
and are therefore able to repair the state by performing E 1 .
367
Limits of quantum error correction

We are only able to correct certain errors. In most codes, the stabilizer operators
are constructed in a way that lets the code correct errors up to a weight t. We
define the weight t of an error E as the number of qubits on which the error acts
non-trivially. If all errors up to weight t are distinguishable and thus correctable,
we say the code corrects t errors. All errors with weight greater than t will destroy
our data. As errors arise from local physics, we expect them to be not strongly
correlated, meaning that errors are not likely to hit multiple qubits at a time. It
is therefore exponentially less probable that an uncorrectable error occurs if we
can correct large t.
The effect of QECC on decoherence errors can be described using the density
matrix of a given state. Its fidelity is defined as F := tr(2 ) and describes how
coherent a state is. A perfectly coherent state has fidelity 1. If we want to prepare
a certain state but due to imperfections we just reach fidelity F = 1 , using
Steanes 7-qubit code [107], which uses 7 qubits to protect one against errors up
to weight 1, we can enhance the fidelity to F = 1 O(2 ). Better quantum error
correction codes [108] which correct t errors so that an uncorrectable error only
occurs if at least t + 1 independent errors occur in a single block, result in a
fidelity of F = 1 O(t+1 ).
This sounds very promising but we have to face another problem: if we raise
t, our code uses more qubits and gets more complicated, therefore it takes more
time to perform error correction as we have to perform more gates. This lack
of speed and the big number of gates result in a rising probability of another
error occurring during correction of the first one, making the code useless. This
leads to a limit on the accuracy needed when performing gates. The estimates for
the threshold differ but the order of magnitude is 105 (most recent value:
3.61 105 [109]), which of course is a very restrictive condition on the hardware.
Once we build hardware that is sufficiently accurate, we can theoretically improve
the fidelity as much as we want and thus perform arbitrarily reliable quantum
computation for an arbitrary duration using concatenated codes [105]. Technical
problems such as the system size growing by a factor 102 103 [110] (assuming
= 106 ) and the need for a support of each qubit by more than ten ancillary
qubits in a low-entropy state [95] remain unsolved.
Topologically protected quantum computation
Another approach to the problem is to build far better hardware. Using todays
techniques, we cannot reach . 105 . The key to better hardware is the locality
of errors. We are therefore looking for systems with states that are not changed
by local interaction which protects them from local errors. This principle is a
368
Topology in Physics
particular feature of topology: topological parameters are not measurable locally

and can thus not be changed by local errors. So the idea behind topologically
protected quantum computation is to encode information in topological properties of the hardware. Such hardware protects encoded qubits passively from
errors and is thus intrinsically fault-tolerant.
Figure 14.2: The states |lefti and |righti are topologically different.
As an example, consider a plane with a hole in it. Draw a path passing the
hole on the left site and another one passing it on the right with fixed endpoints:
see Fig. 14.2. Away from the hole, one cannot decide on which side the path
observed passes it. Errors that deform the path smoothly cannot change the
information |0i := |lefti into |1i := |righti or vice versa as seen in Fig. 14.3. If
we allow our path to be in a superposition of |0i and |1i, we have a topologically
protected qubit.
Figure 14.3: Smooth deformation of the path does not change the state.
Kitaevs toric code

A possible way to build such intrinsically fault-tolerant hardware is described by
a spin model introduced by Kitaev in 1997 [111].
Setup
Consider a k k square lattice with spin-1/2 particles placed on the links of the
lattice (see Fig. 14.4). This gives us n = 2k 2 spins. Here, we choose |1i := (1, 0)T
369
Figure 14.4: There is one spin on each link of the lattice.
and |0i := (0, 1)T , the eigenvectors of z , as the states of each spin. Denote by
s a site, i.e, a point connected by four links and by p a plaquette, i.e, a square
surrounded by four links.
We define the following operators:
Y
Y
jx .
jz
Bp :=
As :=
jStar(s)
jp
As counts the parity of |0is around a site and Bp flips the spins located on the
border of the plaquette. These operators are Hermitian As = As , Bp = Bp with
eigenvalues +1, 1 and satisfy the following commutation rules:
[As , As0 ] = [As , Bp ] = [Bp , Bp0 ] = 0.
Thus we can define a protected subspace Hcode H as the simultaneous eigenspace
of all operators As and Bp to the eigenvalue 1:
Hcode := {|i H : As |i = |i, Bp |i = |i s, p lattice}.
We see that As and Bp have the function of stabilizer operators as discussed in
QECCs. In order to have the lattice live on a topologically non-trivial manifold
(for reasons that are discussed in the section on topological order), we choose a
torus by imposing periodic
boundary
conditions (see Fig. 14.5). This yields two
Q
Q
additional relations, s As = p Bp = 1, which reduce the number of independent stabilizer operators to m = 2k 2 2. From the general theory on QECCs
[105] we therefore know that dim(Hcode ) = 2nm = 22 = 4. Each state |i Hcode
corresponds to a state of 2 qubits: we have encoded 2 qubits using m stabilizer
operators in a n dimensional Hilbert space.
370
Topology in Physics
Figure 14.5: Toroidal geometry is chosen by imposing periodic boundary conditions.
Ground state
The model Hamiltonian is
H0 :=
X
s
As
Bp
so the protected subspace Hcode is the ground state of the system. We chose
classical spin configurations as a basis from which our quantum ground state
configurations will be constructed. The former can be represented by pictures as
plotted in Fig. 14.6. Connections between lattice sites correspond to qubits in
Figure 14.6: Pictorial representation of configurations: the spin states are mapped
to connections.
|1i whereas qubits in |0i are represented by no connection. The operators As and
Bp also have a pictorial counterpart: see Figs. 14.7 and 14.8, respectively.
The ground state constraint As |i = |i demands an even number of |0is and
thus connections at every site. This requires our connections to form closed loops
as they cannot terminate at any site. Those closed loops can move around the
torus by the action of the Bp operator.
371
Figure 14.7: The operator As counts the parity of |0is, which correspond to links
without a connection. This is the same as the parity of |1is, which correspond
to links with a connection.
Figure 14.8: The operator Bp flips all spins around a plaquette resulting in exchanging connection and no connection.
Loop gas
Possible operations on the loops by flipping plaquettes are:
deform loops smoothly (i.e, without cutting them open)
create/annihilate contractible loops
surgery operation (connect two loops or cut a loop into two).
The motion of the loops is given by the time evolution operator U (t) = eiH0 t/~
which includes the operators Bp . Our ground state can thus be visualized as a
gas consisting of loops that transform according to the operations listed above.
This is what we refer to as the quantum loop gas.
All loops that can be obtained from each other by flipping plaquettes form an
equivalence class. Flipping any plaquette must not change the ground state:
Bp |i = |i. This constrains ground states to equal amplitude superpositions of
all loops in the same class. Although this loop gas does not look like a sorted system, there are conserved quantities that allow us to distinguish different ground
states.
372
Topology in Physics
Topological quantum numbers

As our model is defined on a topologically nontrivial manifold, a genus g = 1
torus, loops can wind along two non-contractible paths (along x and along y). The
parity (even or odd) of the winding numbers along those two paths is conserved
by the loop gas operations as can be seen in Figs. 14.9, 14.10 and 14.11.
Figure 14.9: Smooth deformation of loops does not change the winding number.
Figure 14.10: Creating/annihilating does not change the winding number.

This gives four different combinations which correspond to the equivalence
classes mentioned before. Our ground state thus splits in four topological sectors
with an unique ground state in each class (equal amplitude superposition):
|0, 0i, |0, 1i, |1, 0i, |1, 1i
where 0 := even, 1 := odd and |x , y i with x , y denoting the winding number
parity around x and y axis, respectively. We can use those four sectors to encode
two logical qubits:
|0i1 := |0, 0i, |1i1 := |0, 1i, |0i2 := |1, 0i, |1i2 := |1, 1i.
373
Figure 14.11: The surgery operation changes the winding number by two and
thus does not change its parity.
Two qubits are not much and scalability is an important issue. In the section
on topological order, the scalability of this model is discussed. The robustness of
those encoded qubits becomes visible when we investigate the action of errors.
Errors
Suppose an error E L(Hcode ) occurs. One can show [111] that each such error
E can be written as Pauli matrices acting along a closed loops:
Y
S x (t) :=
jx
jt
S z (t0 ) :=
jz .
jt0
S x (t) flips spin along a closed loop t on the lattice and S z (t0 ) gives the parity
of the spins along a closed loop t0 on the dual lattice (see Fig. 14.12). Since
we want the operators to preserve the ground state, they have to commute with
all stabilizer operators. This is the case if and only if the error paths are closed
loops. If the loop is contractible, it can be written as a product of the stabilizer
operators and thus is no error at all: see Fig. 14.13. An error that changes the
value of the logical qubits only occurs if the loop is non-contractible i.e, it has
to wind around the torus at least once: see Fig. 14.15. The crucial point is
that as the errors are assumed to be uncorrelated, the probability for such an
error is ek where k is the lattice size because at least k qubits have to be
affected. Thus, topology protects the logical qubit exponentially in system size
against decoherence errors.
374
Topology in Physics
Figure 14.12: A closed path t on the lattice and a closed path t0 on the dual
lattice. Those paths do not denote states of spins but the action of the error
operators.
Excitations
When implementing such qubits, we have to work at finite temperature which
means that thermal fluctuations introduce transitions to excited states. Leaving
the ground state is only possible if at
Qleast one
Q of the constraints As |i = |i or
Bp |i = |i is violated. Because of s As = p Bp = 1 we are not allowed to do
this at only one site or plaquette: we can only violate the ground state constraint
at an even number of sites or plaquettes. Using the known error operators, we
define the excited states:
|charge (t)i := S x (t)|i,
|flux (t0 )i := S z (t0 )|i.
Every pair of violated conditions can be connected via a string on the lattice or
the dual lattice. The excited states do not depend on the exact path taken but
on its homotopy class. The endpoints of that string can be though of as quasiparticle excitations. A charge-type error causes the loops of the ground state
to break up and form strings with the same endpoints as the path in the error
operator. We say charges live on those two sites (endpoints of t). A flux-type
error creates minus signs at certain configurations in the coherent superposition,
namely at those connected by a plaquette flip at one of the fluxes (endpoints
of t0 ): see Fig. 14.16.
Interestingly, these quasi-particle excitations interact topologically: we calculate what happens when we move a charge counterclockwise around a flux.
|initial i = S z (t0 )| x (q)i is our initial state where | x (q)i denotes the state with
a pair of charges at the endpoints of the path q. S z (t0 ) creates a pair of fluxes.
375
Figure 14.13: The left configuration is affected by the S x (t) error with t being
the path in Fig. 14.12. The result is plotted on the right side. The error thus
transforms one configuration of loops into another one in the same equivalence
class, i.e, it does not change the winding number parities and the encoded qubit
is not affected.
Figure 14.14: A non-contractible path t on the lattice.
We move one charge around one flux:

|final i = S x (c)S z (t0 )| z (q)i = S z (t0 )S x (c)| x (q)i = S z (t0 )| x (q)i = |initial i,
because x and z anticommute. Winding two excitations of different type around
each other results in the global wave function picking up a phase of 1. The familiar Aharonov-Bohm effect [55] explains the choice of the names for the quasiparticles.
These excitations can also be used to jump from one topological sector to
another one. Creating a pair of charges, winding one around the torus and then
re-annihilating the pair corresponds to a non-contractible error path t which
changes the winding number parities of the configuration. We can thus interact
376
Topology in Physics
Figure 14.15: The configuration on the left side is transformed into the one on
the right side by S x (t) acting along the non-contractible loop from Fig. 14.15.
The winding number parity along the x-direction is changed and this corresponds
to a non-trivial operation on the encoded qubit.
with our logical qubits using quasi-particle excitations. The energy cost for creating a pair of excitations is 2U where U is some energy scale in the Hamiltonian.
If U is big compared to the operating temperature, such processes will be exponentially suppressed e2U/T .
Kitaevs model describes thus how topology can help reliably storing quantum
information by exponentially eliminating the effects of both thermal and quantum
noise.
Quantum Dimer Model

In order to implement Kitaevs model for topologically protected qubits, one
has to find microscopic Hamiltonians that show the same effective behaviour. A
promising candidate is a more realistic model called the quantum dimer model
(QDM). It was first introduced in 1988 in the context of superconductivity of
cuprates [112] and has been thoroughly investigated [113, 114, 115, 116, 117]
on several lattices. The triangular lattice is the simplest one that supports a
resonating valence bond (RVB) phase [118] for a finite range of parameters with
deconfined, gapped excitations [114], see Fig. 14.18.
Setup of the QDM
Each site of the lattice, whose boundary conditions are chosen periodic, is connected to exactly one other site, thus forming dimers. The connections are
restricted to next neighbour bonds because, assuming a spin gap, such configurations have proved to be good low-energy variational states [112]. As this
condition, called the hard-core condition, has to hold for all sites, the lattice is
377
Figure 14.16: A charge-type error acting along a string on the lattice and a fluxtype error acting along a string on the dual lattice. Disks represent charges
and filled plaquettes fluxes.
Figure 14.17: Moving a charge along the path c counterclockwise around a

flux gives a phase 1.
completely covered with dimers that do not touch each other. Classical configurations are again taken as a basis and superimposed in order to find the quantum
states. The Hamiltonian of the model can be written in terms of dimers:
X
H = tT + vV =
t(| ih | + | ih |) + v(| ih | + | ih |).
The kinetic term T lets the configuration gain an energy t by flipping pairs of
parallel dimers while the potential term V acts as a repulsion between flippable
dimer pairs since it punishes such configurations with an energy v. For t v one
gets a highly frustrated and unordered phase called dimer liquid phase according
to its similarity with other liquid phases. This special phase, that has been shown
to be stable for 0.7 . v/t 1 [114, 119], corresponds to Andersons RVB phase
[118] and has interesting quasi-particle excitations.
378
Topology in Physics
Figure 14.18: The phase diagram for the QDM on a triangular lattice calculated
by A. Ralko et al. in 2005. [119] It shows, that the RVB phase persists for a finite
range of parameters and not just at one point as on the square lattice.
Transition graph
QDM states can be mapped onto the loop gas: choose one specific dimer configuration C0 as background. This can in principle be any configuration but usually
the columnar state is chosen. Superimposing a given configuration C with C0
yields a set of closed loops called the transition graph of C relative to C0 , see Fig.
14.19. The effect of the operators in the Hamiltonian of the QDM on those loops
Figure 14.19: Superimposing the hard-core dimer configuration on the left side
with the reference columnar configuration in the middle yields the transition
graph on the right side. It consists of closed loops that can be used to describe
the configuration.
correspond to the loop gas operations: see Figs. 14.20, 14.21 and 14.22. As
in the Kitaev model, there are conserved quantities that allow us to distinguish
different configurations. Instead of the winding number parity, the dimer count
parity along a non-contractible loop is chosen. It is left invariant by dimer flips
(see Fig. 14.23) so the Hilbert space splits into topological sectors corresponding
to different parities.
379
Figure 14.20: The isotopy move.
Figure 14.21: Create/annihilate contractible loops.
Implementation using Josephson junction arrays

This idea is used in an implementation of such qubits proposed by L. B. Ioffe et
al. in 2002 [110]. Josephson Junction arrays are used on a hexagonal lattice with
binding states of cooper pairs representing dimers between islands of hexagons of
such junctions. These dimers thus live on the dual lattice which is the triangular
one. By clever design, they meet the hardcore condition, periodic boundary
conditions and short ranged interaction between dimers so this system shows the
same behaviour as the QDM on a triangular lattice. Instead of a two-dimensional
torus in three dimensions, they use a annulus in the plane. The dimer count
parity along a line from the inner to the outer boundary acts as topological order
parameter and splits the Hilbert space into Heven Hodd . The two sectors are
not connected by local operators in the Hamiltonian and can be used to encode
a logical qubit. Operating on the qubit is achieved using one controllable link.
This implementation is treated in more detail in the next chapter.
Topological order and fractionalization

In both the Kitaev model and the QDM, non-local properties of the system are
used to hide quantum information. This is achieved by associating topological
quantum numbers such as winding numbers with states of qubits. The reason
for a system to have such non-locally encoded quantum numbers is a concept
380
Topology in Physics
Figure 14.22: Surgery operation.
Figure 14.23: The dimer count along the line is changed only by the surgery
operation which changes it by two thus preserving the dimer count parity.
called topological order. Conventional order is described by Landau symmetry

breaking and conservation laws come from symmetries in the Hamiltonian via the
Noether theorem. However, systems with topological order can have conserved
quantities without requiring a corresponding symmetry in the Hamiltonian. This
kind of order can not be understood using correlation functions and local order
parameters.
Quasi-particle excitations in topologically ordered systems interact topologically,
i.e, rotating them around each other has a non-trivial effect on the wave function
of the system. In the chapter about anyons we see that this is of great importance
for quantum computation.
Recently, it has been shown [120] that topological order is a necessary condition
for fractionalization which is defined as quasi-particle excitations carrying fractions of the quantum numbers in the system. Systems that show fractionalization
are thus interesting candidates for quantum computation.
A further hallmark of topological order is a ground-state degeneracy depending
on the topology of the system instead of being caused by spontaneous symmetry
breaking. This is the reason why the boundary conditions in the Kitaev model
are chosen periodic: the system then lives on a torus, which is a non-trivial genus
g = 1 manifold and allows us to encode two robust qubits. More general, the
ground state degeneracy in the Kitaev model is 4g and allows us thus to encode 2g
381
qubits. This permits to scale the number of encoded qubits exponentially: a torus
with 10 holes hosts 1024 robust qubits. Systems that show topological order can
be seen in reality, although it is hard to detect this kind of order experimentally.
The best-understood example is the fractional quantum hall effect.
Fractional quantum hall effect
We measure a voltage in the direction perpendicular to the current flowing when
putting a conductor into a magnetic field. This is the well-known classical Hall
effect:
IB
UH = RH
.
d
RH is the Hall constant, and for a single type of charge carriers, RH = 1/(ne). 54
The usual measurement setup is to fix the current and record the Hall resistance
H = UH /I. The latter is supposed to grow linearly in B as seen from the
formula above. But in a two-dimensional electron system such as a GaAs GaAlAs heterostructure, we see plateaus at low temperatures ( 10mK) and
strong magnetic fields 5T at H = 1/m h/e2 where m is an integer. This is
the integer quantum hall effect (IQHE) [121, 122, 123], which can be measured
incredibly precisely (109 ) independent of the material.
When thoroughly investigating such systems with even higher magnetic fields
and lower temperatures, extra plateaus appear in addition to the integer ones,
namely at
H = 1/ h/e2
with = p/q and where p and q are integers, q being odd. The parameter , which
is the filling factor of the Landau levels, describes the density of electrons with
reference to the present magnetic flux: = nelectrons hc/(eB). This phenomena is
called the fractional quantum hall effect (FQHE) [125]. It originates from many
electron correlations, and cannot be explained the same way IQHE is (using
Landau Levels). Quasi-particle excitations in fractional quantum hall states with
filling factor = q/(2q + 1) (main FQH sequence) carry a fractional charge [126]:
e =
e
.
2q + 1
(14.1)
The ground state of such strongly correlated electron systems has been found to
have a degeneracy depending on the genus g of the manifold it is defined on [127]:
qg ,
(14.2)
where q is the denominator of . The quasi-particles have fractionalized statistics

[128]: exchanging two excitations in a state counterclockwise (ccw) results in
54
I=current, B=magnetic flux density, q=charge of carrier, d=thickness of the conductor

parallel to the B-field and n=density of carrier particles
382
Topology in Physics
Figure 14.24: The fractional quantum hall effect: plateaus appear in the transversal Hall resistivity at fractional Landau level filling factors. Picture adopted from
Ref. [124].
the wave function picking up a phase

ei .
(14.3)
As the integer q resp. the fraction appear in Eqns. (1), (2) and (3), the
existence of a common background, which is topological order, is visible.
Besides the very important FQHE and the QDM [114, 129, 130, 119], there are
other examples for fractionalization and thus topological order:
Z2 gauge theory and high-Tc superconductors
The Ising gauge theory is a more general concept which has a limit that describes
the QDM [116]. The field has been extensively studied on various lattices in the
context of frustrated Heisenberg antiferromagnets and high-Tc superconductors
[115, 131]. The electron breaks into fractional particles and furthermore, there
are non-trivial, gapped topological excitations [132]. The electron splits up into
a chargeon that carries its charge and a spinon that carries its spin. The
flux of the Z2 gauge field is a gapped excitation dubbed the vison[133, 134].
So topological order in the form of a RVB state is also a possible explanation for
383
high-Tc superconductivity. An experiment in 2001 [135] has shown that if visons

exist, they are very unstable.
Rotating BECs
The important parameter is the filling fraction = N/Nv where N denotes the
number of Bosons and Nv the number of vortices. There is a zero-temperature
phase transition as a function of between a triangular vortex lattice phase and a
strongly correlated vortex liquid phase, which is quite similar to the incompressible liquid phase in FQHE, and is thought to be governed by topological order
[136].
Frustrated magnets
Experiments carried out in 2003 have shown that the quasi-2D spin 1/2 frustrated
Heisenberg antiferromagnet Cs2 CuCl4 is very likely to have spin 1 waves which
fractionalize into two spin-1/2 excitations (spinons) [137]. Those experiments
also showed that the quantum fluctuations dominate the mean field effect, which
would prefer order, indicating an RVB spin liquid state.
Quantum computation by anyons

Topological order can be used to hide quantum information: encoding qubits in
the global properties of the system protects them from local influences. Excitations that interact topologically such as Abelian anyons, can be used to jump from
one topological sector to another by moving them along topologically non-trivial
paths and thus perform one-qubit operations. Before we can further investigate
the idea of quasi-particles performing operations, we have to describe those excitations emerging in systems governed by topological order in a well-defined way.
Anyons
In three spatial dimensions, the statistics of identical particles is restricted to
Fermi-Dirac and Bose-Einstein statistics because a path that describes two counterclockwise (ccw) exchanges can be smoothly deformed into no exchange at all.
This is not possible in two dimensions therefore statistics change drastically. A
ccw exchange of two identical particles cannot only result in a phase of -1 or none
but also in any other phase ei . Anyons can also consist of more than one real
particle. As an example of an anyon, consider a flux-charge composite with flux
and charge q, sometimes called cyon. It acquires a phase eiq when counterclockwise rotated by 2 as an Aharonov-Bohm effect [55]. q = is also referred
to as topological spin of the object. The ccw exchange of two such composites
q
q
results in a phase: ei( 2 + 2 ) = eiq = ei . The topological spin is the same as the
384
Topology in Physics
exchange phase. Therefore we have a similar connection between spin and statistics as in 3D (exchange phase = e2is where s =spin). We can state the following
hand-waving argument: half-integer or integer spin correspond to Fermions or
Bosons and fractionalized spin thus corresponds to fractionalized statistics. Further details are presented in the last chapter.
The permutation group Sn does not suffice to describe anyons. Instead, we
draw the two-dimensional (2 d) world in which our anyons live in the x-y plane
and chose the time axis along z. The time evolution of the system lets the
anyon world lines form braids in 2+1 d. The braids are in distinct topological
Figure 14.25: Exchanging ccw and cw is not the same in 2d: the world lines of such
processes in 2+1 d are topologically different. Therefore we do not investigate
the permutation group to classify particles but the braid group, which takes this
difference into account.
classes. A ccw exchange is not the same as an clockwise (cw) exchange of two
particles because they cannot be smoothly deformed into each other. As a further
simplification, we usually do not draw 2+1 d but 1+1 d with a new relation over
and underneath. If the left particle goes over the right one, this corresponds
to a ccw exchange while the right going over the left is a cw exchange. So the
topological interactions between particles in 2+1 d are described in a well-defined
way by drawing braids in 1+1 d and the latter are mathematically described by
the braid group.
Unitary representations of the braid group Bn
The objects of the braid group act on lined up particles by moving them in
topologically different ways to another configuration of the same particles. We
call the n particles (1, ..., n) and let i , i = 1, ..., n 1 denote a ccw exchange
385
of the particles at positions i and i + 1. The i are the generators of the braid
group and they satisfy the following two rules:
j k = k j , |j k| 2
and the Yang-Baxter relation:
j j+1 j = j+1 j j+1 j = 1, ..., n 2.
The braid group is infinite and has thus an infinite number of unitary irreducible
representations. These representations are carried by the anyons in the sense
that braiding anyons of type with some braid b Bn results in (b) acting on
the Hilbert space. This leads us to a definition for anyons:
Abelian anyons := Indistinguishable particles that transform as a onedimensional representation of the braid group.
Each braid group generator thus corresponds to a phase (j ) := eij and
with the Yang-Baxter relation, we see that all the phases have to be the same:
eij eij+1 eij = eij+1 eij eij+1 eij = eij+1 =: ei .
All ccw exchanges are represented by the same phase, which tells us that a phase
is enough to fully characterize Abelian anyons. The special cases = 0 and =
correspond to Bosons and Fermions, respectively. The effect of braiding Abelian
anyons on the system is restricted to picking up phases. Those operations are
not sufficient for universal quantum computation. However, the mere existence
of Abelian anyons in a system implies topological sectors in the ground state that
can be used to encode robust qubits. To perform universal quantum computation
with those robust qubits, we need to be able to perform one- and two-qubit gates
which are matrices. The idea presented by M. Freedman et al. in 2000 [138, 139]
is to use non-Abelian representations to perform such gates.
Non-Abelian anyons := Indistinguishable particles that transform as higherdimensional representations of the braid group.
Braiding them results in performing gates on a topological Hilbert space because the representation has dimension greater than one and is thus a matrix:
|ifinal = (braid)|iinitial .
The gate performed by a braid only depends on the topological class of the braid
but not on the exact form of the world lines. The environment is allowed to
disturb the anyons during calculation as long as it does not affect the topological
properties of the braid.
386
Topology in Physics
Figure 14.26: Braiding particles that carry higher-dimensional representations

of the braid group results in matrices acting on the Hilbert space via |ifinal =
(braid)|iinitial and can thus be used for quantum computation [139]. Picture
adopted from Ref. [140]
If we keep the anyons far apart, it is exponentially improbable that such errors
occur. Another source of errors are virtually created anyon anti-anyon pairs that
interfere with our computational braid before re-annihilating. Such processes
can be suppressed by low operating temperatures. For certain anyon models, the
representation is dense in all reversible, unitary operations and therefore, braiding
such anyons can approximate any operation with arbitrary fidelity: they have the
capability of simulating universal topologically protected quantum computation
[139] .
Non-Abelian anyon models
Before we can give an example, we introduce some concepts. All details can be
found in Ref. [103].
A general non-Abelian anyon model describes particles in a two-dimensional
surface that carry locally conserved charges. It has three defining properties:
Labels: They specify what charges the different particles carry.
Fusion rules: They specify the possible values of the charge that can
be obtained when two particles of known charge are fused together and
analogous for splitting.
Braiding rules: They define the phase picked up when rotating a particle
by 2 and the result of an exchange of two particles.
When we combine two particles, the resulting compound also has a charge.
The possible charges obtained are defined by the fusion rules. As the fusion is
387
Figure 14.27: Non-Abelian anyons provide a topologically protected way to perform gates. Noise that alters the world lines does not change the gate executed
as long as the braid stays in the same homotopy class.
associative, there are different ways to fuse to a given total charge. These correspond to different bases which are related by an unitary transformation called the
F -matrix. An anyon model is non-Abelian if at least one pair of charges has more
than one possibility to fuse into another charge. The Hilbert space spanned by
the different possible fusion results is called fusion space of that pair of charges.
The Hilbert space of n anyons with given total charge c is a sum of fusion spaces
and one can define a standard basis for it [103]. Often, this Hilbert space is
referred to as the topological Hilbert space to emphasize that the information
is encoded non-locally: the state is defined by its fusion result but if the particles
are far apart, one cannot access this information. This explains that we have to
fuse the result of our computation in order to measure the calculation output.
Counterclockwise exchanging particles with charges a and b does not change
their total charge c. Therefore the exchange operator, which is called R, acts
on their fusion space. If the latter is multidimensional, the former is a matrix
dubbed the R-matrix. Furthermore, the eigenvalues of the monodromy operator
R2 are determined by the topological spins of the particles.
To fully specify the representation of any braid on the topological Hilbert space,
it suffices to investigate the effect of the generators of the braid group: using
the F -matrix, one can move from the standard basis to a block diagonal basis
for R, apply R and then transform back. Thus, the representation of the braid
group realised by n anyons is completely determined by the F -matrix and the
R-matrix, which have to fulfill consistency relations called the pentagon and the
hexagon relation [103].
Since R2 defines the topological spin, the constructed representation is defined
on a larger group, the mapping class group for the sphere with n punctures
388
Topology in Physics
which can be visualized by extending world lines to ribbons that can twist [111].
All those rules together define a mathematical structure called a unitary topological modular functor which is closely related to quantum field theories in 2+1
d and conformal field theories in 1+1 d [141, 91, 105, 139, 142].
Fibonacci anyons
To illustrate the ideas, the simplest non-Abelian anyon model that is universal
for quantum computation is briefly presented here: the Fibonacci anyon model
[103, 143].
Defining properties of Fibonacci anyons:
There is only one charge called the q-spin. It can take the values 0 and 1.
The fusing rules are: 0 0 = 0, 0 1 = 1, 1 1 = 0 + 1.
A q = 0 anyon is the same as no anyon at all. The two-dimensional fusion space
of two qubits thus consists of the state where two q = 1 anyons fuse to q = 0
and to q = 1, denoted by |(11)0 i and |(11)1 i, respectively. Two q = 1 qubits can
be in a superposition |i = |(11)0 i + |(11)1 i. Adding further anyons to the
system obeying the fusion rules enlarges the number of possible combinations for
a cluster of anyons to fuse to total q-spin 0 or 1. For n anyons, the dimension
of the topological Hilbert space is the (n + 1)th Fibonacci number, hence the
name. The consistency relations for braiding and fusing yield the R-matrix and
the F -matrix in an unambiguously way up to an unimportant global phase:
i4/5
e
0
0
ei2/5
0
R := 0
i2/5
0
0
e
51
F := 0 :=
.
2
0
0 1
Following [138], we encode a qubit in the three dimensional topological Hilbert
space of three Fibonacci anyons.
|0i := |(((11)0 )1)1 i, |1i := |(((11)1 )1)1 i, |N Ci := |(((11)1 )1)0 i
The q-spin of the two leftmost particles thus determines the state of the encoded
qubit. |N Ci is not used as computational state and transitions into this states
correspond to unwanted leakage errors (decoherence). Braiding within the qubit
does not change its total charge so as long as we perform only single-qubit operations, we do not face decoherence problems. One-qubit operations are here
389
described by the braid group of three particles, B3 . The representation of the

generators 1 and 2 can be calculated using the R-matrix and the F -matrix:
i4/5
e
0
0
(1 ) = R = 0
ei2/5
0
i2/5
0
0
e
ei/5 i ei/10
0
.
(2 ) = F 1 RF = i ei/10
0
i2/5
0
0
e
Each braid b B3 can be written as product of 1 and 2 and therefore we
can calculate the matrix corresponding to all possible braids by multiplying the
matching powers of (1 ) and (2 ). For example, the braid b = 2 11 2 11
results in |final i = (b)|initial i = (2 )(1 )1 (2 )(1 )1 |initial i
Finding the braid that simulates a given gate U SU (2) with desired accuracy
is the more interesting task. It can be done by brute force, which is very hard
and scales exponentially, but Solovay and Kitaev found an algorithm [144] that
allows systematic improvement of a braid. With this algorithm, the braid length
used to achieve error rates smaller than grows as |ln()|c , where c 4.
The real challenge is to implement the CNOT gate or another multi-qubit gate
that is universal for quantum computation. With more than one qubit involved,
we have to face the problem of leakage errors but clever weaving techniques
(injection weaving) deal with that. Bonesteel et al. have proposed a CNOT gate
using six Fibonacci anyons (three for each qubit) that reaches an error rate of
103 [143].
The effective operations executed by the braid are single qubit operations in
SU (2) that can be improved using the Solovay-Kitaev algorithm and therefore,
that CNOT can theoretically be carried out with arbitrary accuracy. Thus, we
can theoretically build a universal topological quantum computer using Fibonacci
anyons: see Ref. 14.30.
Anyons in real life?
Fractional quantum hall states are the most promising candidates for systems
with anyons, i.e., theoretically, quasi-particles emerging from a state have braiding statistics of = : see [128, 146] and references therein. But until 2005,
experimental proof was lacking. F. E. Camino et al. [146] have measured interference patterns when sending a e = 1/3 quasi-particle around an island containing
a = 2/5 FQH fluid which agree with the assumption that they are Abelian
anyons that interact topologically. Still, the results are not excessively explained
by this assumption which lets many scientists doubt that this experiment can be
seen as a proof for the existence of anyons.
Theory also predicts non-Abelian statistics for certain filling factors . The most
390
Topology in Physics
Figure 14.28: A CNOT gate is approximated up to an error of 103 using

six Fibonacci anyons. With the first part (Injection), identity is approximated
resulting in a permutation of two anyons of the control qubit with two anyons
of the target qubit. This ensures that the next part (Rotation) is carried out
within one qubit thus preventing leakage errors. The last step (Extraction) is the
inverse of the injection braid and permutes the anyons again. If the control qubit
is in state |0i, the effect of this braid is identity which makes it a controlled-NOT
gate. Picture adopted from Ref. [145]
important two are = 5/2 and = 12/5. Although the non-Abelian statistics
describing quasi-particles in the = 5/2 state are not sufficient for universal
computation, one can simulate a universal set of gates with a very modest error
threshold of > 0.14 [147]. M. Freedman et al. [148] have proposed another way
to extend the = 5/2 statistics to universal quantum computation and their
estimate for the error rate is . 1030 ! The = 12/5 state is believed to have
non-Abelian anyons that can be described using a SU (2)3 Chern-Simons gauge
theory [149] and thus are universal for quantum computation [139].
It is not yet clear whether quantum error correction or topologically protected
quantum computation will be the future standard of quantum computation but
the beautiful topology involved gives the latter an intrinsic coolness factor
[148].
391
Figure 14.29: Improved version of the CNOT with an error of 104 . The braid
length used to achieve a ten times better accuracy grows by a factor of 4.
Picture adopted from Ref. [145].
392
Topology in Physics
Figure 14.30: A possible quantum computing circuit approximated by braiding

non-Abelian Fibonacci anyons. Error rate & 103 . Picture adopted from Ref.
[145].
15 Topological quantum
computing implementations
Mario Berta
Supervisor: Fabricio Albuquerque
For any implementation of a quantum computer one can not
eliminate every source of decoherence. We show in this chapter how one can use the ideas of topological quantum computing
[150] to implement quantum bits that are topologically protected
and therefore stable against decoherence. We discuss a possible implementation, introduced by Ioffe and collaborators [110],
based upon quantum simulators built from Josephson junction
arrays for the triangular lattice quantum dimer model in its liquid phase.
Introduction
If we want to implement quantum bits (henceforth dubbed qubits) for a quantum

computer we have to overcome conflicting requirements. On one hand the qubits
should be manipulable through externals signals and on the other hand the qubit
should be screened as much as possible from its neighbourhood. As an example
we can think of an electron spin as a qubit (i.e. spin up and spin down are the
qubit states). We need to isolate this electron spin from everything except the
measurement and manipulation mechanism. Furthermore we need to be able to
turn off the coupling of the qubit to the measurement so as to not decohere the
qubit while performing an operation on it. Because of the very quantum mechanical nature of the qubits these requirements are difficult to achieve.
Hence any quantum computer has to incorporate some fault tolerance. It is
impossible to eliminate every source of decoherence. There exist quantum errorcorrection schemes [151] to deal with this problem and to scale down the number
of errors. These schemes require active interference with the system during runtime. Quantum error-correction schemes spread one qubit onto several qubits and
393
394
Topology in Physics
then the encoded quantum information is protected against errors of a limited

form. The problem is that the system increases by a factor of 102 to 103 when
such schemes are applied. Moreover, we require an accuracy about one error in
104 operations that it is useful to apply quantum error-correction schemes. This
implies that each gate must perform its task 10000 times faster than the decoherence time of the system. Up today it is not possible to achieve this.
Recapitulating we can say that decoherence is one of the biggest problem we have
to face when we want to build a quantum computer. Topological protection of
qubits offers a promising possibility to solve this difficulty. Kitaevs idea [150]
of topological quantum computing is to bring the stabilization against decoherence to the hardware level. This means that the desired quantum computer is
coherent because of its physical nature and the stabilisation against decoherence
is achieved passively.
For a physical implementation of a topologically protected qubit we need a many
body quantum system which fulfils the following constraints: (i) The Hilbert
space of the system splits into two orthogonal sectors. (ii) Each of these sectors
remains isolated under the action of local perturbations. (iii) The ground state
in both sector has the same energy (i.e. the ground state is twofold degenerate).
(iv) The excitation spectrum is gapped in sectors.
If we choose the two states of a qubit as the two ground states in the different
sectors, these states are stable under external disturbance and hence are stable
against decoherence. Furthermore the ground states within each sector are secured through a gapped excitation spectrum. To manipulate such qubits in order
to build a topologically protected quantum computer we have to find and implement global operators which can switch the qubits between the distinct sectors
of the Hilbert space.
A many body quantum system that displays all these desired properties is the
Quantum Dimer Model [112] [152] [153] [154]. Ioffe et al. [110] proposed to emulate such a Quantum Dimer Model in a Josephson junction array. It is the main
goal of this report to review and explain the ideas of this paper.
This report is organized as follows. In the next section we introduce the Quantum Dimer Model and discuss its properties on the square and triangular lattice.
Moreover, we present an exact solvable Quantum Dimer Model for the Kagome
lattice. Afterwards we describe the Josephson effect and arrays of Josephson
junctions. Finally we discuss the idea of implementing topologically protected
qubits in a Josephson junction array (Ioffe et al. [110]).
Topological quantum computing implementations
395
Quantum Dimer Model
Motivation of the Quantum Dimer Model

Two dimensional (2D) antiferromagnetic spin systems are well described by the
2D Heisenberg model whose Hamiltonian reads:
X
H=J
Si Sk
(15.1)
hi,ki
where Si is the spin operator of the electron on site i and the sum runs just over
nearest neighbours. J stands for a super exchange coupling between two spins
on nearest neighbour sites.
After extensive and intense research it is known today that the ground state of
Figure 15.1: For classical spins the ground state is the ordered Neel state. In this
configuration the spins are alternating pointing up and down.
the antiferromagnetic Heisenberg model in the square [155] and triangular lattice
[156] is Neel-ordered (see Fig. 15.1) at T = 0 and hence displays long-ranged
order with spontaneous symmetry breaking.
Because of the work of Anderson [118] people got interested in other possible
quantum phases without magnetic long-ranged order as well. Conditions that
favour unconventional states are low dimensional lattices with significant geometrical frustration (see Fig. 15.2) and strong quantum mechanical fluctuations
(they are maximal in the extreme quantum limit S = 21 ) [157]. Spins can lower
?
Figure 15.2: Geometrical frustration in a lattice with triangular symmetry: It is
not obvious how to orient the spins in a Neel-ordered way. The spin on the third
site can point up or down equal likely, because both of the states have the same
energy.
their energy if they build singlets and the most likely state in such a regime is a
state with all spins paired into singlets [118].
396
Topology in Physics
The state of two spins that are paired, sitting on sites i, j, is given by an antisymmetric combination of up and down spins:
1
|(ij)i = (| i j i | i j i)
2
(15.2)
where i denotes a spin up at site i and j a spin down at site j. These singlet
states are called valence bonds (see Fig. 15.3). We can get a state for the whole
Figure 15.3: A valence bond state for a 4 4 square lattice.

lattice if we assign valence bonds to all vertices and write the product state as a
tensor product of all the singlets:
Y
|(ik jk )i.
(15.3)
|V Bi =
pairs
Since there are a lot of ways to pair the spins, a possible ground state without
long-ranged magnetic order will be a superposition of such valence bond (VB)
states. VB states can be long-ranged, i.e. any two spins can pair (Fig. 15.3)
or short ranged, i.e. just spins that are nearest neighbours can pair (Fig. 15.4).
Note that in the last case the valence bonds are non-intersecting and we can
therefore think of hardcore dimer coverings instead of short ranged VB states.
For extremely frustrated spin systems we expect short ranged VB states. One
Figure 15.4: A short-range resonating valence bond state on a 44 square lattice.

The bold lines represent dimers.
can find that the short ranged VB states form a basis for the spin zero sector of
the Hilbert space of a given system. Furthermore they are linearly independent
for all lattices which are discussed in this text [158]. Quantum Dimer Models
397
(QDM) are defined in the Hilbert space of short ranged VB states and try to
describe phases without long-ranged magnetic order. So QDM do not operate in
the z spin basis anymore, but in a new basis: The dimer basis.
Given the basis we need to find an effective Hamiltonian from the Heisenberg
model that describes the dimer degrees of freedom. Because the short ranged
VB states are highly non-orthogonal we need to calculate the overlap-matrix
a,b = ha |b i for any two short ranged VB states |a i and |b i. To do this it is
necessary to establish the concept of transition graphs. The transition graph of
two dimer coverings C, C is constructed by drawing C and C on the same lattice.
The result is a covering with non-intersecting loops. Trivial loops are obtained
(i)
(ii)
(iii)
Figure 15.5: Two dimer coverings C, C and their corresponding transition graph
(trivial loops are skipped because we are not interested in them).
when the dimers are in the same positions in both coverings. For non-matching
bonds the loops get larger, although they are always closed.
Now lets first have a look at the square lattice. Since the square lattice is
bipartite55 every bond connects one sub lattice W with the other sub lattice
B and the transition graph is orientable. A dimer from covering C is oriented
from W to B and one from C from B to W. This results in oriented loops
W B W B etc. in the transition graph (see Fig. 15.5).
L
Sutherland [159] derived the result a,b = 21 2 where L is the length of the loops
of the transition graph between |a i and |b i. So an orthogonalized valence bond
P
1
state takes the form: |
ai = a ( 2 )a,a0 |ai and we can write the matrix elements
of the effective Hamiltonian that describes the situation in the dimer basis as:
X
1
1
eff
(15.4)
=
( 2 )a,a0 ha0 |H|b0 i( 2 )b,b0 .
Ha,b
a0 ,b0
In general this is a difficult expression, but Rokhsar and Kivelson [112] introduced
a simplification. They considered a formal overlap expansion parameter x and
replaced ha |b i by 2xL . Their idea is to expand the effective Hamiltonian in
the parameter x [112]. For physical SU (2) spin systems we have x = 12 and for
an expansion up to order n we can get an effective Hamiltonian which may lead
55
It is possible to break down the lattice in two sub lattices (white W and black B).
398
Topology in Physics
to a good approximate model.

There are many ways to choose such Hamiltonians but convenient choices are
much simpler than their original spin Hamiltonians and are often accessible to
numerical and analytical calculations [158]. In particular it is possible to write
down some simple QDM in a Kagome lattice which is solvable and all excited
state wave functions are known [130].
QDM can be good approaches to the phases of antiferromagnets which are dominated by short range valence bonds, as in an extreme quantum limit of a SU (2)
spin model [160]. Even if QDM are toy models in some way, they lead to many
nontrivial results and are interesting objects. As we will see in the next two sections, these models offer simple descriptions of unconventional orders as Valence
Bond Crystals (VBC) [112] as well as resonating valence bond (RVB) liquids [114]
[130].
Moreover, Ioffe et al. [110] realised that QDM can be used to implement topologically protected qubits.
Quantum Dimer Model on the square lattice
The first QDM was introduced by Rokhsar and Kivelson [112]. The model is
defined on the square lattice, and the Hamiltonian reads:
H = tT + v V =
Np
X
(t(| ih | + | ih |) + v(| ih | + | ih |))
(15.5)
i=1
where Np is the number of plaquettes of the lattice. The Hamiltonian contains

a potential term and a kinetic term (represented by V , T resp.). We can obtain
this Hamiltonian with a formal overlap expansion of the Heisenberg Model in the
parameter x, where we just keep the two most local terms (i.e. the lowest order
terms56 ). To analyze the Hamiltonian we first discuss the two terms T and V
separately.
The potential term stands for an attractive or repulsive (i.e. negative or positive
v) interaction between dimers. This term is diagonal in the dimer basis: V | i =
v| i and analogue for | i.
On the other hand the kinetic term represents a dimer move on the lattice.
Without lost of generality t can be taken greater or equal zero57 . The kinetic
term rotates parallel dimers and |i = | i + | i is an eigenstate of T on a single
plaquette system. Thus the kinetic energy maximizes the number of resonating
plaquettes. We can say the kinetic term resonates the dimers, what can naturally
be associated with kinetic energy. As we can see the two terms of the Hamiltonian
counteract each other and favour different phases.
At one end of the diagram, for vt > 1, the system minimizes the number
56
In fact t v x4 and v v x8 . Note that a three dimer kinetic term of order x6 is not considered
because the term is less local [112].
57
Because a change in the sign would just reverse the orientation of the states of the dimers.

(i)
(ii)
~0.2
399
(iii)
v/t
Figure 15.6: Phase diagram of the square lattice QDM: (i) columnar phase (ii)
plaquette phase made up resonating plaquettes (iii) staggered phase.
of flippable plaquettes (=plaquettes with two parallel dimers) due to the large
repulsive interaction between dimers. The resulting configuration is the staggered
one. This configuration is a zero energy eigenstate of the Hamiltonian in general
and a ground state for v t. We can see this by calculating the expectation
value for the energy on one plaquette. A non-flippable plaquette is annihilated
by H and a flippable plaquette has a potential energy of v and a kinetic of t or
t. This leads to the following constraint:
min(0, Np (v t)) E max(0, Np (v + t))
(15.6)
where Np is the number of plaquettes of the lattice. For vt > 1 this shows that
any zero energy eigenstate is the ground state. As we can see the staggered
state breaks several lattice symmetries. Such configurations with a long range
dimer-dimer order and broken lattice symmetries are called a Valence Bound
Crystals (VBC). Note that there is no SU (2)-symmetry breaking or long range
spin-spin order. We can rotate the state by 90 and obtain the same configuration.
Therefore the phase is four-fold degenerate. Furthermore no local dimer move
can take place in this regime. We can see this if we apply the Hamiltonian to the
staggered state; the state gets annihilated.
For v < 0 we realize the other extreme situation. Parallel dimers attract each
other and the system maximizes the number of flippable plaquettes. The state
that results is the columnar state. For t = 0 the columnar state is an exact
eigenstate, but not for t 6= 0. The excitations in this regime are gapped (E =
2 |v|) and correspond to a pair of dimers which are rotated by 90 compared to
the background configuration. The eigenstate for t = 0 is a VBC because we have
a broken lattice symmetry. Moreover the VBC will survive a small and finite t
term because of the gapped nature of the excitations [113].
In the middle of the phase diagram for t |v| (i.e. a large and dominating
kinetic energy) the system builds a resonating plaquettes crystal |i = | i + | i.
This phase is a VBC too. Leung et al. [113] showed using exact diagonalizations
that this plaquette phase is realized for 0.2 vt < 1.
All the phases we discussed up to here are VBC and therefore display conventional
order. But moreover there exists another type of phase with no broken lattice
400
Topology in Physics
symmetries. In fact, for the square lattice, we have this just for one position in
the phase diagram: The point where t = v. This is somehow intuitive because
the amplitudes of the kinetic and potential terms are exactly equal a this point
and the terms can antagonise maximally. A this point (called Rokhsar-Kivelson
(RK) point) the Hamiltonian can be written as:
X
H=
|p ihp |
(15.7)
p
|p i = | i | i.
(15.8)
Even if the system has no broken lattice symmetries at the RK point, this phase
can exhibit a kind of order, known as topological order. To analyze the situation
at the RK point we need to talk about this topological order.
Considering a lattice on a toroidal geometry (i.e. periodic boundary conditions in
x- and y-direction), we can define two winding numbers x and y . The winding
numbers are the net number of topologically non-trivial loops (clockwise minus
counter clockwise) in the transition graph encircling the torus in x and y directions, respectively (see Fig. 15.7). By defining a columnar state as the reference
y
X
x
(i)
Y
(ii)
Figure 15.7: (i) Transition graph with winding numbers x = 1 and y = 0.

(ii) Toroidal geometry with the two possible loops encircling the torus in x- and
y-direction, resp.
state it is possible to label any dimer configuration by its winding numbers due
to the reference state. Configurations with the same winding numbers can be
obtained by local dimer moves (the winding number of the transition graph associated to every local movement is zero). So the coverings can be grouped into
topological sectors, labeled by its winding numbers. These topological sectors
are disconnected sectors of the Hilbert space since two coverings with the same
winding numbers (due to the reference state) can by achieved by repeated application of the Hamiltonian and it is impossible to leave a given topological sector
under the action of the Hamiltonian. For the square lattice the Hamiltonian
is ergodic (meaning every configuration with the same winding number can be
reached under the action of the Hamiltonian). The number of x and y is of
order O(L2 ), where L is the length of the lattice.
401
Coming back to the RK point, the equal amplitude linear superposition of all
possible coverings |ci in a given topological sector is a ground state. We can
write this state as follows:
1 X
|0i =
|ci
(15.9)
Nc c
where Nc is the number of all possible dimer coverings in . |0i is called a RK
wave function. It is a ground state because the Hamiltonian annihilates it. We
can show this:
Consider one single plaquette in a configuration |ci. If |ci is a configuration with
no or just one dimer we have hp |ci = 0. For a two dimer configuration there
exists a configuration |c0 i in the same topological sector which differs from |ci
by a two dimer flip. |ci + |c0 i is then again orthonormal to |p i, which implies
H|0i = 0.
Rokhsar and Kivelson [112] calculated the excitation spectrum at the RK point
using a variational calculation and found a gapless spectrum.
At this special point they were even able to calculate the dimer-dimer correlations. It turns out that the correlation function is algebraically decaying with
distance ( r12 ) and therefore the dimer-dimer correlations are not truly short
ranged.
Recapitulating we notice that the QDM on the square lattice is believed to be
ordered (VBC) everywhere except at the RK point. Moreover the QDM exhibits
some interesting topological properties at this point. Our goal is to use these
properties for the implementation of topologically protected qubits. Unfortunately at least three problems occur. First the number of topological sectors is of
order L2 and unequal to two. Secondly the topological phase is just realized for v
exactly equal to t, what is rather difficult to implement. And thirdly the gapless
excitation spectrum makes the system susceptible to perturbation. However, on
the triangle lattice we can overcome these problems.
Quantum Dimer Model on the triangular lattice
The most local dimer Hamiltonian on the triangular lattice was first discussed
by Moessner and Sondhi [114] and reads:
H=
Np
3
X
X
{t(| ih | + | ih |) + v(| ih | + | ih |)}
(15.10)
i=1 =1
where the sum on i runs over all the rhombi of the lattice, the sum on stands
for the three possible orientations of the plaquettes (each rotated by 60 degrees)
and Np is the number of plaquettes. It is possible to give a motivation for this
Hamiltonian with an overlap expansion of the Heisenberg model.
As on the square lattice the Hamiltonian contains a kinetic and a potential term
402
Topology in Physics
(T and V ). These terms act exactly as on the square lattice.

The set {|ci|c = 1, ..., Nc } corresponds to all possible dimer coverings. Applying
V to a state |ci we get: V |ci nf l (c)|ci where nf l (c) measures the number of
flippable plaquettes of |ci.
Like on the square lattice we can have a look at the phase diagram. At the right
(i)
(ii)
(iii)
~0.7
(iv)
v/t
Figure
15.8: Phase diagram of the triangular lattice QDM: (i) columnar phase
(ii) 12 12 phase (iii) liquid phase (iv) staggered phase.
end of the diagram, for vt > 1, the system is in the staggered phase in which
the ground state manifold consists of all non-flippable configurations. It is easy
to check that equation (15.6) still holds and therefore the ground state is the
zero energy eigenstate. The non-flippable configurations actually are these zero
energy eigenstates. The staggered phase is a VBC and no local dimer move can
take place.
On the other end where v < 0 and t = 0 the ground states are the maximally
flippable states, i.e. those with maximal nf l . This phase is highly degenerate
because by shifting all dimers along a straight line in an ordered columnar state,
the number of flippable plaquettes is preserved (note that this is not true for the
square lattice). The number of such maximally flippable states is exponential in
L, the system size. For an infinitesimal t the degeneracy is expect to be lifted.
In this case the system gets ordered and favours the columnar state. This phase
is a VBC.
In
situation for v around zero the system favours a so called
a intermediate
12 12 crystalline phase. The dimers build a triangular super-lattice with a

12 site unit cell out of resonating diamond plaquettes. This phase is a VBC too.
One can calculate using Pfaffian Methods that the system is in a phase without
broken lattice symmetries at the RK point [114][161][162]). For a state as in
equation (15.9) we can get h0|H|0i = (v t)nf l . This is zero at the RK point
(v = t) and therefore the ground state is an equal amplitude superposition of all
states in a given topological sector.
On the square lattice the phase with no broken lattice symmetries is realized just
for one disconnected point. The new effect on the triangular lattice is that the
phase enlarges to a range 0.7 vc vt 1 and is even a liquid phase (exponentially decaying dimer-dimer correlation function). Moreover Moessner and Sondhi
403
[114] found from their Monte Carlo simulations (on clusters up to Lx Ly = 36)
that the spectrum is gapped and that the degeneracy of the ground state holds
in the thermodynamical limit for the whole sector vc vt 1. This picture is
consistent with the exact diagonalization studies of Ioffe and collaborators [110]
(done on clusters up to Lx,y = 6). This is an interesting situation because the
topological properties exist for a finite range for the parameters (t and v) and
the topological sectors remain isolated under perturbations because the system
is in a liquid phase.
But the topological properties are not the same anymore. Because the triangular
lattice is not bipartite it is not possible to orient the transitions graphs. Hence
the winding numbers x , y (in a toroidal geometry) are defined as the parity of
the number of non-trivial loops around the system in x- and y-direction in the
transition graph due to the reference columnar state.
The winding numbers are not conserved by local dimer moves; they are conserved
modulo two. Therefore they are equal to zero or one. This results in just four
topological sectors (x , y ): (0,1), (1,0), (0,0), (1,1). If we take instead of a torus
a cylinder (i.e. periodic boundary conditions in just one direction, lets say xdirection) there is only one winding number x present. Hence we only have two
topological sectors which we can use as the states of a qubit. x is the conserved
topological quantity and x = 0 corresponds to the state 0 of the qubit (analogue
x = 1 to the state 1 of the qubit).
But if we think of an actual physical implementation a problem occurs: It is
non-trivial to measure transition graphs and therefore difficult to determine the
state of a given qubit. We can solve this problem when we define the topological
conserved quantity in a different but equivalent way. We define it as the parity
of the number of dimers intersecting a given line . This line has the following
two properties: (i) ends at the boundaries of the cluster and (ii) does not
divide the cluster into two disconnected pieces. The first condition ensures that
the parity is really conserved when applying the Hamiltonian and the second one
guarantees the possibility of constructing configurations with both parities. On
a cylinder it is easy to see that the only choice is a straight line in y-direction,
going from one end of the system to the other.
Here we can see easily that the qubit states are protected against local disturbance. A local disturbance can just flip flippable plaquettes and this will not
change the parity and therefore not decohere the qubit states. The equivalence
of these two definitions can be shown as follows:
Consider the columnar reference state. If we shift one dimer to the left in xdirection all dimers along that line have to do so too. Because otherwise there
would be vertices with more than one dimer. So we can see that the transition
graph of this new dimer covering contains a non trivial loop that encircles the
cylinder in x-direction. Hence the two states have different winding numbers
x = 0 and x = 1. But the parity of counting along the line has changed
too. For the reference state the parity is even and for the shifted state it is odd.
404
Topology in Physics
(i)
(ii)
(iii)
Figure 15.9: Reference line : (i) count of 3 odd parity (ii) count of 5 odd
parity (iii) corresponding transition graph with x = 0.
Therefore the definitions for the conserved topological quantity are equivalent
in this situation. Now consider a state where the dimers are shifted along two
lines in x-direction in the same manner. For this covering the winding number is
zero (because two non trivial loops encircle the cylinder) and the parity is even.
Furthermore it is easy to see that this state and the columnar reference state can
be connected by local dimer moves.
As one can check these arguments generalize for any dimer covering of any size
triangular lattice. Thus the two topological sectors (x = 0, 1) can be identified
with the sectors of even and odd parity58 .
So on the triangular lattice we find promising properties to implement a topologically protected qubit, but we have to check if the properties concerning the
liquid phase and the topology still hold for a potential physical implementation. For that purpose it is necessary that the system is quite tolerant with
variations in the parameters (because in any physical system there are always
impurities and defects). As mentioned above expensive Monte Carlo Simulations
[114] and exact diagonalization [110] studies show that the liquid state is realized
for 0.7 vc vt 1. It was also found that there is a weak temperature (= T )
dependence in the interval 0.25t > T > 0.03t and that the gap in the excitation
spectrum is of order = 0.1t.
Furthermore the following articles were discussed by Ioffe et al. [110]: (i) The
mixing of the two ground states would require the creation of topological defects
and thats why the mixing is really weak. The effect is exponentially suppressed
in the system size Lx . (ii) The robustness of the degeneracy of the ground-state
can be checked by disturbing the system with a quenched disorder potential.
This is a mathematical method that varies t and v slightly over the lattice. It
was found that the degeneracy is robust to within a factor of 103 to 102 of the
disorder potential and the results are shown in Fig. 15.10. It is expected that this
58
On the square lattice it is possible to introduce the same idea. There the parity of counting
along in a given topological sector is preserved (i.e. the parity is conserved for local dimer
moves). But then it is not possible to label the topological sectors by the parity number. This
is evident because there are O(L) topological sectors (for a cylindrical geometry) and just two
different parity numbers.
405
robustness increases exponentially in the system size Ly . (iii) A possible decoher0.04
d=0.01t
d=0.02t
d=0.04t
cylinder
6*5
d /d
d /d
1.5
0.02
torus
1
0
v/t
6*5
1
0.01
0.5
0
0.8
0.9
v/t
Figure 15.10: Splitting d of the ground state energies under the action of a
disorder potential of strength d. Near v t the disorder splitting for the 6 5
torus is of order 103 of the disorder energy d.
ence in the relative phase of the two ground states has two different sources. The
creation of non-topological excitations in a sector or an adiabatic splitting by an
external low-frequency noise. One can eliminate the excitations by setting the
operation temperature T . The impact of noise is suppressed for the same
reason as the ground-state is robust. (iv) Because an implementation always has
finite size they checked if there are low-lying edge states, which would destroy
the energy gap in the spectrum. The calculation shows that there are none.
So for a large lattice (i.e. for Lx , Ly large) the system fulfils all requirements to
implement topologically protected qubits.
Note that we may also have to consider because of possible higher order dimer
flips. Such dimer flips include three or even more dimers on different plaquettes
and may destroy the topological order. For any potential implementation this
problem has to be analyzed carefully.
Because most of the results on the square and triangular lattice QDM are obtained by numerical calculations it is illustrative to insert here a section about
the Kagome lattice. For this model some of the results can be obtained relatively
easy.
Quantum Dimer Model on the Kagome lattice
An exactly solvable QDM on the Kagome lattice was introduced by G. Misguich,
D. Serban and V. Pasquier [130]. This model offers a very natural and simple
framework to illustrate the ideas discussed above. Especially, we will discuss the
dimer liquid phase, the topological order, the dimer-dimer correlations, the gap
406
Topology in Physics
in the excitation spectrum and the elementary excitations.

Let us first introduce the Kagome lattice. Consider a hexagonal lattice and
take the midpoints of the bonds. The Kagome lattice is the lattice of these
midpoints59 .
Now let us define the dimer model on the Kagome lattice. For any given hexagon
h we define the operator x (h) as the sum of all possible kinetic terms involving
h only:
32
X

x (h) =
|d (h)ihd (h)| + H.c. .
(15.11)
=1
The sum runs over the 32 loops that enclose a hexagon and around which dimers
can be moved. There are 8 inequivalent loops (see Fig. 15.11). The simplest
(i)
6
(ii)
3
10
12
Figure 15.11: Inequivalent loops on the Kagome lattice: (i) loop length (ii) number of dimers involved.
loop is the hexagon itself and it involves 3 dimers. Moves with 4, 5 and 6 dimers
are possible as well by including additional triangles to the loop. We find that
the loop length has to be even. The largest loop is the star. |d i(h) and |d i(h)
are defined as the two ways dimers can be placed along a loop on a hexagon
h. These operators are the kinetic terms, they act like the kinetic terms on the
square lattice. Note that for a given dimer covering |Di all kinetic operators but
one annihilate |Di.
The Hamiltonian for this dimer model introduced in [130] contains only these
kinetic terms:
X
H=
x (h),
(15.12)
h
where the sum runs over all the hexagons h of the lattice60 . This Hamiltonian is
not obviously solvable when written in this form.
59
In fact all the conclusions in this section can be generalised to any lattice made of cornersharing triangles.
60
It is possible to get a similar Hamiltonian with an overlap expansion of the Heisenberg
Hamiltonian on the Kagome lattice. But in this picture the kinetic operators x (h) have
different amplitudes or in a simplification different signs. Including this, the Hamiltonian is
407
One of the advantages of the Kagome lattice is that there is a correspondence

between dimer coverings and arrows sitting on the bonds of the hexagonal lattice.
The existence of this arrow representation is a central reason for which the QDM
on the Kagome simplifies. This representation was introduced by Elser and Zeng
[163] and is illustrated in Fig. 15.12. Each arrow sits on the bonds of a hexagon
Figure 15.12: Arrow representation on the Kagome lattice.

and has two possible directions: It points toward the interior of one of the two
neighbouring triangles. If site i belongs to a dimer (i, j) its arrow must point
toward the triangle the site j belongs to. For a triangle without any dimers this
rule implies that all arrows are outgoing.
For any given dimer covering it is possible to draw the arrows. Contrariwise a
dimer covering can be constructed from any arrow configuration provided that the
number of outgoing arrows is one or three (i.e. odd) on every triangle. Hence there
is a one-to-one correspondence between the dimer coverings and the arrow configurations that fulfill the constraint above. We can translate the x (h)-operators
into this representation: x (h) just flips the 6 arrows sitting on h (see Fig. 15.12).
It is easy to see that the x (h) conserve the constraint for all triangles. x (h)2
flips all arrows twice and thus x (h)2 = 1. Furthermore the x (h)-operators,
placed on different hexagons, commute with each other, i.e. [ x (h), x (h0 )] = 0.
Note that this is a non-trivial fact in dimer language.
In fact there is even another representation of dimer coverings. Zeng and Elser
[163] [164] realised that there is a close correspondence between Ising configurations of pseudospins sitting on hexagons and dimer configurations on the Kagome
lattice. The mapping works as follows:
The first step is to take an arbitrary reference dimer configuration |D0 i. We can
associate to any dimer configuration |Di, that lies in the same topological sector
as |D0 i, a pair of pseudospin configurations {hz = 1}hhex in the following way
(see Fig. 15.13). (i) Draw the loops of the transition graph of hD0 |Di. (ii) These
loops now divide the areas where the pseudospins are up and where they are
unfortunately not exactly solvable anymore. Note that the Hamiltonian from equation (15.12)
is therefore not a good approximation for an antiferromagnetic spin system on a Kagome lattice.
408
Topology in Physics
Figure 15.13: Pseudospin representation on the Kagome lattice.

down. Because |Di and |D0 i are in the same topological sector this can be done
in a consistent way. (iii) There are two possibilities to assign the pseudospin since
there is no natural way to define where is the exterior and interior of a closed loop
on a finite sample. Therefore two different pseudospin configurations (related by
a global pseudospin flip) represent the same dimer covering. This is a one-to-one
correspondence between a dimer covering in a given topological sector and a pair
of pseudospin states (with reversed sign). We can proof the uniqueness:
(1) If two dimer coverings |Di and |D0 i from the same sector have the same
pseudospin state (up to a global sign) they are identical. The transition graph
hD|D0 i is the difference between the transition graphs hD0 |Di and hD0 |D0 i. So
hD|D0 i contains the loops that separate the regions where the pseudospins are the
same in D and D0 and the regions where the pseudospins are different. Because
D and D0 have the same pseudospin on each hexagon hD|D0 i contains no loops
at all. Therefore D and D0 are identical. (2) For any pseudospin state we can
associate a corresponding dimer state. For a given reference dimerisation |D0 i
the transition graph between the dimer configuration we are looking for and |D0 i
will separate hexagons with z = 1 from hexagons with z = 1. This path will
of course depend on |D0 i but for a fixed |D0 i the path is unique. To see this,
consider a single hexagon. Whatever |D0 i is there is only one loop that encircles
this hexagon only. We can reconstruct the unique dimer state that corresponds
to the pseudospin state by leaving out the part of |D0 i in the transition graph.
The x (h)-operator just flips the pseudospin sitting on the hexagon h. We can
understand this easily:
Consider an arbitrary dimerization |Di and have a look at the hexagon h. All
kinetic operators but one annihilate |Di on this hexagon. Applying this operator
to |Di gives a state |D0 i that differs from |Di only by a single loop around h.
In pseudospin language this is exactly the unique state obtained from |Di by
flipping the pseudospin at h.
To determine the ground state in a given topological sector we can find that
| i = |d (h)i + |d (h)i and | i = |d (h)i |d (h)i are the eigenstates of the
Hamiltonian on a single plaquette. We can think of a pseudospin in x-direction
assigned to these two states. The corresponding energies are 1 for | i and 1
for | i, so the ground state is fully polarized in the pseudopin language in xdirection: x (h) = 1 h. If we think of that in the in the z representation, the
409
ground state the sum of all pseudospin configurations in the z basis. Translated
back into dimer language this is nothing but the sum of all dimer coverings in a
topological sector; the RK wave-function.
We can derive the dimer-dimer correlations in this state from the arrow representation. If two arrows on two bonds are not on a common triangle they
are independent. Therefore dimer-dimer correlations are strictly zero when their
corresponding triangles do not touch. So dimers on the Kagome lattice are independent above a finite distance. Because of this absence of long-ranged dimer
correlations, the RK state is a dimer liquid and breaks no lattice symmetry. To
draw a comparison, on the square lattice the dimer correlation function decays
algebraically and on the triangular exponentially. But the correlation always remains finite. Not so on the Kagome lattice, we can say the RK state is a close
as possible to a free dimer gas.
We can calculate the whole excitation spectrumQtoo. A said before the x (h)x
operators commute from hexagon to hexagon.
h (h) flips all arrows twice,
hence for a setup with periodic
Q boundary conditions and no edges the following
constraint has to be fulfilled: h x (h) = 1. So the first excited state is not just
a single but a pair of flipped hexagons with an energy cost of 4. A x (h) = 1
hexagon is called a vortex excitation or a vison and excited state consist of two
such hexagons a, b with x (a) = 1 and x (b) = 1. The excited states are
non-local because the flipped hexagons a and b can be arbitrarily far away from
each other. We can state the first excited wave-function explicitly. Consider a
string going from a hexagon a to a hexagon b (see Fig. 15.14). We define (a, b)
as the operator that measures the parity of the number of dimers crossing that
string. (a, b) commutes with all x (h), except for the hexagons a and b at the
Figure 15.14: Definition of (a, b).

ends of the string61 . There we have: x (a)(a, b) = (a, b) x (a) and analogue
for b. Any dimer move changes the sign of (a, b) if and only if the associated
loop crosses the string an odd number of times. So (a, b) flips x (h) in a and
b and leaves the rest in the same state62 . If we apply (a, b) to the RK wave
61
62
Note that we can choose any string going from a to b.

(a, b) = z (a) z (b) in the pseudospin language.
410
Topology in Physics
function |0i in a given topological sector we get (a, b)|0i which is still a linear
combination of all dimer configurations belonging to that sector. But the amplitudes at hexagons a and b have changed from 1 to 1. (a, b)|0i is the first
excited states of energy 4.
Josephson Junctions
The goal of this section is to describe a Josephson junction array. Lets first
consider a single junction.
A single Josephson Junction consists of two weakly coupled superconductors.
The weak link can be of different types. As an example a thin layer of normal
metal between the superconductors. Josephson [165] realised that there is a nonvanishing zero-voltage supercurrent of Cooper pairs through the link between
the superconductors. This phenomena is called Josephson effect or Josephson
tunneling.
The governing equations for this effect can be derived in many ways. We show
here a simply and heuristic derivation first proposed by Feynman [166].
We label the two superconductors with 1 and 2. Now consider a Cooper pair. It
can be either in 1 or in 2 and is under the influence of a Hamiltonian H = H1 +H0 ,
where H0 is the term that describes the uncoupled system and H1 is the coupling
term. For a weak coupling we can write the wave function of the Cooper pair
as a linear superposition of the two uncoupled states u1 and u2 . u1 is the
eigenstate of H0 and E1 the corresponding energy (analogue for u2 ). I.e.
H0 u1,2 = E1,2 u1,2
(15.13)
and the time dependent Schroedinger equation reads:

i~
= H = (H0 + H1 ).
t
(15.14)
We take the inner product of the Schroedinger equation with u2 and use =
a1 u1 + a2 u2 :
i~
a1
= E1 a1 + a1 hu1 |H1 |u1 i + a2 hu1 |H1 |u2 i.
t
(15.15)
We can do the same for u2 and get:

i~
a2
= E2 a2 + a2 hu2 |H1 |u2 i + a1 hu2 |H1 |u1 i.
t
(15.16)
The amplitudes a1 and a2 are complex numbers and we can therefore write a1 =
1 exp (i1 ) and a2 = 2 exp (i2 ), where 1 and 2 are the mass density
411
of Cooper pairs within the superconductors. 1 and 2 are the corresponding

phases. We get for the Cooper pair current I = 1 = 2 :
I=
2H12
1 2 sin (2 1 ) = Ic sin ,
~
(15.17)
where H12 = hu1 |H1 |u2 i, = 2 1 is the relative phase difference and Ic =
2H12
~ 1 2 is the critical current. The current of equation (15.17) is the Josephson
current that is responsible for the Josephson effect. From the imaginary part of
the equations we can get:
Z
1
(t) = 0
(E20 E10 )dt,
(15.18)
~
where 0 is an arbitrary constant and Ei0 = Ei + hui |H1 |ui i. For a macroscopic
system E20 E10 is equal to the chemical potential difference between the two
sides. Thus we can write:
d
=
.
(15.19)
dt
~
Out of the Josephson current we can calculate the coupling free energy F :
F = F0 E J cos ,
(15.20)
where F0 is an arbitrary constant and E J the Josephson coupling energy

EJ =
I0
.
2c
(15.21)
Here 0 = hc
is the flux quantum, h Plancks constant, c the speed of light
2e
and e the elementary charge. This energy is a characteristic energy scale for the
Josephson effect.
Furthermore, because the probability to find a charge at the edges of the junction
is higher than to find one in the middle, there is always a capacitance C associated
with the junction. The corresponding Coulomb energy reads:
EC =
(2e)2
.
2C
(15.22)
The next step is to put Josephson junctions together and build a superconducting
array. The simplest possible array is the square lattice. To analyze the dynamics
of such a system it is useful to calculate the Coulomb energy and its dependence
on the differently chosen capacitances. It is illustrative to keep in mind the simple
square lattice picture but the generalisation is straightforward.
We can number all the islands and the charge on the ith island is
X
(15.23)
Qi = Cg Vi +
Cik (Vi Vi,k ).
k
412
Topology in Physics
Cik
Cg
Figure 15.15: Josephson junction square array built from superconducting islands
that are coupled through Cik and have a ground capacitance of Cg .
Where Vi is the potential the ith island is on, Vi,k the potential one of the ith island
nearest neighbours is on, Cik the capacitance between two neighbouring islands,
Cg the ground capacity (which is assumed to be constant for all capacitances)
and the sum runs over the nearest neighbours of the ith island. As a next step
we can write out the Coulomb energy of the system
1
1X
1
E = CV 2 =
Vi Cij Vj = hV |C|V i
2
2 ij
2
(15.24)
and use this as an intrinsic definition for the capacity matrix Cij . So the capacitance matrix for our setup reads:
X
X
Cik j,i+k .
(15.25)
Cik )i,j
Cij = (Cig +
k
The capacitance matrix is regular for nonzero ground capacitance and therefore
the inverse of Cij exists. Moreover the matrix is even symmetric: C = C T .
Because we do not know the exact potentials all the islands are on we rewrite the
energy for the system. For V = C 1 Q we get
1
1X
E = hQ|C 1 |Qi =
Qi Cij1 Qj .
2
2 ij
(15.26)
With this equation we can calculate the Coulomb energy of the lattice in dependence of the charges on the islands. Of course it is in general difficult to find
the inverse of the capacitance matrix. For infinite lattices it is useful to transform this calculation into the Fourier space. There analytical calculations can be
made [167]. For finite lattices one usually depends on numerical approximations
to calculate the inverse.
413
Until now we are just able to calculate the Coulomb energy of static configurations. To include the tunneling of Cooper pairs from one island to another we
have to establish the Bose-Hubbard Model (BHM). A two dimensional BHM is
useful to study the effects of interactions between Bosons in a two dimensional
lattice. The Hamiltonian looks like:
X
X
(15.27)
H = HU + HT =
ni Uij nj
Tij (bi bj + bj bi )
i,j
hi,ji
where U contains the Coulomb interaction between the bosons, Tij is the hopping
integral, hi, ji runs just over the neighbours, bi is the creation operator for a
boson at site i, bi is the annihilation operator for a boson at site i and ni =
bi bi is the operator that counts the bosons at site i. The operators fulfill the
following commutation rules: [bi , bj ] = ij , [bi , bj ] = 0 and [bi , bj ] = 0. Due to
the electromagnetic interaction bosons can hop from one site to another. This
hopping process is just possible between nearest neighbours and the b-operators
exactly represent such dynamics.
For an array of capacitances the U-Term can naturally be identified with the
capacitance matrix. The bosons are Cooper pairs in our superconducting lattice
(with charge 2e) and we can insert the Coulomb interaction energy from equation
Qi
(15.26) into the Hamiltonian. If we understand the Q0i s as operators with ni = 2e
2
we can write:
1X 2
HU =
4e ni Cij1 nj .
(15.28)
2 i,j
The T-term is responsible for the hopping of Cooper pairs. Because of the Josephson effect there is a non-vanishing current over every junction and we can understand this as a hopping process. The T-term is therefore proportional to the
Josephson coupling energy E J . Thus our Hamiltonian can be written as:
H = HU + HT =
X
1X 2
4e ni Cij1 nj
EijJ (bi bj + bj bi ).
2 i,j
(15.29)
hi,ji
This Bose-Hubbard Hamiltonian is a good description for a Josephson junction

array. Generally one needs to make use of perturbation theory to get some useful
results.
Implementation
As said before we want to use the triangular lattice QDM for the implementation
of topologically protected qubits. We wish to exploit the following features of
this model: The topological structure of the Hilbert space and the presence of a
liquid ground state with a gap against excitations.
But Ioffe et al. [110] do not use spin systems for the implementation because
414
Topology in Physics
there exists no such system which is fully described by a QDM. And even if there
would be one, it would still be hard to manipulate the system in a controlled way
as it would be necessary for the implementation of qubit operations.
Ioffe and collaborators proposed two different Josephson junction arrays for emulating the quantum dimer liquid state on the triangular lattice; the Josephson
junction triangle (JJT) and the Josephson junction Kagome (JJK). The JJT ar-
Ih , C h
flip
CY
Il ,C l
flip
JJT
Figure 15.16: Josephson junction triangle (JJT) implementation.
ray is constructed with Y-shaped superconducting islands. One vertex of the
triangular lattice consists of six such islands with two ends forming a hexagon
and the third end linking to the neighbour hexagons (see Fig. 15.16). All islands
are coupled to their neighbours via a capacitance and a Josephson junction. CY
is the capacitance of the Y-island to the ground, Ch the linking capacity in the
hexagon and Cl the capacitance between two adjacent hexagons. Ih is the Josephson current within a hexagon and Il the current for the linking junction. In sector
2 we derived the corresponding Coulomb energy E C and the Josephson Coupling
energy E J (see equations (15.22) and (15.21)). We need to find CY , Ch and Il to
define the classical dimer states. First we choose Ch to be large for the purpose of
joining the hexagons electrically into one vertex. A small capacitance CY defines
EC
a large charging energy Ehex 6Y (for Ch large) of a hexagonal vertex. Ehex is
the basic energy scale of the array and can be calculated by thinking that one
vertex (hexagon) is basically nothing more than six capacitances CY connected
in parallel. We can understand Ehex as an approximation for the Coulomb interaction energy between two Cooper pairs sitting on the same hexagon.
The next step is to bias the whole array with a global electric gate to control the
number of Cooper pairs in the system. We choose the magnitude of the gate so
that there is on average just one Cooper pair for every second vertex. The large
charging energy Ehex lifts states with two or more Cooper pairs on one hexagon
to high energies.
Now, a Cooper pair on a given hexagon can lower its energy via tunneling through
the Josephson junction joining two adjacent vertices (involving the Il coupling).
415
The bonding state of such a Cooper pair defines the dimer state (valence bond
state). Because there is just half a Cooper pair per hexagon, every vertex builds
one and only one dimer.
The hopping of dimers comes in due to the vertex junction with small Josephson
current Ih Il . The hopping process involves the vertex junction first and proceeds via localisation of one dimer to a vertex (using the Il Josephson current).
Afterwards the located Cooper pair builds a new dimer with another vertex (via
the link junction between them). This vertex is now in a virtual state; occupied
by the new and the old dimer (i.e. two dimers on one vertex). The large charging
energy Ehex makes this situation improperly and the old dimer flips in a subsequent hop to another site. This results in the total hopping process predicted
from the quantum dimer model on the triangular lattice. The hopping amplitude
(E J )2
is of order t EhJ .
l
One can furthermore calculate the electrostatic interaction v between parallel
dimers by comparing the energy of the staggered and columnar configuration.
The second proposed idea for the implementation is the JJK array. The first
C
I h , Ch
flip
flip
CX
JJK
Figure 15.17: Josephson junction Kagome (JJK) implementation.

components are X-shaped islands arranged in a Kagome lattice. They are coupled through capacitances and Josephson junctions with energies EhC and EhJ and
their capacitance to the ground is CX (see Fig. 15.17). A second triangular lattice
is laid into the hexagons of the Kagome lattice and consists of star-shaped islands
(with ground capacitance C ). They are capacitively coupled to the six X-islands
surrounding each of these star-islands (via Ci ). The corresponding Coulomb energy is EiC . Note that even we use a Kagome lattice, the QDM implemented in
the JJK array is defined on the triangular lattice.
As on the JJT array we introduce a electric gate to bias the Kagome sites to
ensure that there is on average just half a Cooper pair for each hexagon. Dimers
are defined as Cooper pairs residing on X-islands. To the constraint that no
two dimers should touch each other (i.e. two Cooper pairs on the same hexagon
416
Topology in Physics
should be forbidden) we need to choose the capacities accordingly.

A small capacitance Ch isolates the hexagons electrically from each other and
a large capacitance Ci joins the six X-shaped islands into one island via their
strong electric coupling to the star-shaped island in the middle. Next we define
the charging energy Ehex as the energy required to put two Cooper pairs on the
same hexagon. This energy is the basic energy scale of the array and should
be large to ensure that there are never two Cooper pairs on the same hexagon
(EhJ Ehex ). From a perturbative calculation we get Ehex ( CCXi )2 EC which
shows that Ehex is determined by CX ,C and Ci . For a good screening on large
distances it is furthermore a good idea to choose Ci < C , CX . This guarantees
that there is no long range interaction between the Cooper pairs.
The dimer dynamics involve the Josephson junction between the X-shaped islands (with energy EhJ ). Due to this coupling one Cooper pair hops to another
hexagon which is then double occupied. Because of the large Ehex this is just
a virtual state and one Cooper pair hops to a further hexagon. This motion
of a Cooper pair is the hopping process as predicted from the QDM. Using a
perturbative theory we can calculate the the amplitude of this hopping process:
(EhJ )2
t Ehex
.
To realise the liquid phase of the QDM we need to satisfy the condition t v. It
is possible to choose values for the capacitances that this constraint is fulfilled.
We now construct a topologically protected qubit using the Quantum Dimer
Model on the triangular lattice with cylindrical boundary conditions. The QDM
is emulated in the Josephson junction array as discussed above. The qubit is
realised in a ring geometry (see in Fig. 15.18). The reference line joins the
inner and outer boundary of the array. The qubit states |ei and |oi are defined
as the states with odd and even dimer counting parity along the reference line.
These two states are the ground states in two distinct topological sectors and are
topologically protected. Hence the qubit states are protected too. In order to
qubit
F
amplitude shifter
phase
shifter
GSS
Figure 15.18: Implementation of one qubit and one qubit operations.

manipulate this qubit we need to implement a qubit Hamiltonian. We want to
417
be able to produce any state of the form:

|, i =
[|ei + exp i|oi]
,
1 + 2
(15.30)
i.e. we like to vary the amplitude and the relative phase of the qubit states. The
corresponding qubit Hamiltonian reads:
Hqubit = hx x + hz z ,
(15.31)
where x and z are the Pauli matrices and hx , hy are the parameters that produce the amplitude and the phase from equation (15.30). Thus we need to
implement hx and hy .
The implementation of hx requires a mixing of the protected qubit states what
results in a reduction of the ground states topological protection. In the JJT
array we place an amplitude shifter at the inner boundary of the ring geometry.
This amplitude shifter is a tunable Josephson junction. The tunable junction
has a variable Josephson coupling energy ElJ . With an energy cost of ElJ we can
break the dimer over this junction. This creates a virtual particle-hole excitation
where one Cooper pair is localised on one hexagon (the particle) and the other
hexagon is empty (the hole). While the particle remains at the weak junction the
hole is taken around the inner boundary through subsequent dimer flips. After
a whole circle the hole is recombined with the particle. Now the qubit state has
EJ
changed. The amplitude of this process is hx EhJ ( EhJ )M where M is the number
l
of links on the inner boundary of the ring. Hence we can change hx by changing ElJ . On the JJK array the implementation involves a virtual state with two
Cooper pairs on one hexagon.
To implement a phase shifter hy we need to construct a gated superconducting
strip (GSS) that lies on the reference line , i.e. the GSS joins the inner and
outer boundary of the array. The strip should be capacitively coupled to the
array and attract dimers onto the reference line . When we bias the strip, the
energies of the qubit states shifts with respect to another and so does the relative
phase between the qubit states. The energy u that acts on one dimer due to
this process and the duration of this operation have to two constraints: (i)
u t ElJ and (ii) > ~ . This guarantees that there are no excitations within
the dimer liquid. Moreover the fully connected strip represents a global operator
and would led to electric fluctuations that would decohere the system. Hence the
strip has to break up during idle time. It is possible to construct such a strip out
of superconducting Cooper pair transistors [168].
Note that both, the amplitude mixing and the phase shifting are strongly suppressed during idle time. We derived above that amplitude mixing is exponentially small in Lx and phase shifting exponentially small in Ly .
We are now able to implement a single qubit and to manipulate it accordingly.
418
Topology in Physics
In order to build a whole quantum computer we need many qubits. We can implement K qubits in a Josephson junction array with K holes (see Fig. 15.19).
Around every hole we can visualise the same ring structure from the single qubits.
Furthermore we have to implement the qubit Hamiltonian (as in equation (15.31))
for every hole. We can do this in the same way as stated above. The last step is
GSS
Figure 15.19: Implementation of many qubits and two qubit phase shifters.
the implementation of a two-qubit operation. We use again a GSS, now between
two holes that correspond to two qubits. Due to this GSS the phase between
the states |eoi, |oei and |eei, |ooi is shifted. With these two qubit phase shifters
and the general qubit Hamiltonian for one qubit operations we can construct
the controlled-NOT operation [169]. Thus we can perform any n qubit unitary
operation [170] and can therefore do all the operations we need for a quantum
computer.
Conclusions
We introduced the Quantum Dimer Model and showed why it can be used to
study disordered spin systems. For the square lattice we were able to describe
VBC and found a phase without broken lattice symmetry with an algebraically
decaying correlation function, a gapless spectrum and a Hilbert space that splits
up into several topological sectors. For the triangular lattice we found VBC and
a phase without broken lattice symmetry as well. This phase is realised for a
whole range of the parameters, has exponentially decaying correlations (hence is
a liquid phase) and a gap in the excitation spectrum. We found that the Hilbert
space splits up into four topological sectors on a torus and into two on a cylinder.
Moreover we discussed a solvable Quantum Dimer Model on the Kagome lattice
and found a liquid phase with finite correlation length and the same topological
properties as on the triangular lattice.
Thereafter we explained how to describe Josephson junctions and arrays of Josephson junctions with a generalised Bose-Hubbard Model.
In the next section we reviewed the idea of Ioffe et al. [110] how one can implement a qubit in the triangular lattice Quantum Dimer Model on a cylinder.
We showed, if one chooses the two ground states in the two topological sectors
419
as the two qubit states, that the qubit is topologically protected. In the last
point we described how one can build a quantum simulator for the triangular
lattice Quantum Dimer Model in its liquid phase in a Josephson junction array
and discussed its properties concerning an actual physical implementation.
420
Topology in Physics
16 Braid group statistics in

quantum mechanics
Klaus J
ager
Supervisor: Walid Abou Salem
We will discuss the configuration spaces of identical particles in
different dimensions n. In n 3 we will show that there are
bosons and fermions only. The situation is totally different in
two dimensions, where a continuum of states interpolating the
bosonic and the fermionic cases arises. These states are called
anyons. We will discuss braid groups as fundamental groups of
two-dimensional systems. In the end, we will discuss the cyon
system and will give a short introduction into the Chern-Simons
theory.
Introduction
In this chapter, we will give a short introduction to Abelian braid group statistics,
which arise in the context of identical particles in two dimensions.
At the beginning of section 2, we will find the correct configuration manifolds
of systems of many identical particles. To understand them better, we will discuss
two-particle systems in detail. We will find that the two-two-dimensional case is
different from higher dimensional cases [171].
In section 3, we will quantize identical two-particle systems. We will find,
that in two dimensions a continuous range of statistics can exist, connecting the
bosonic and fermionic cases. Furthermore we will derive in a canonical way that
in three and more dimensions only bosons and fermions can exist [171].
In section 4 we will introduce the braid group, the fundamental group of the
configuration spaces in two dimensions. We will investigate it and will derive its
connection with the statistics [172].
421
422
Topology in Physics
Finally in section 5, we will discuss anyons in a magnetic flux tube, called the
cyon system. Afterwards we will give a very short introduction into the ChernSimons formalism and show, how it is connected with fractional statistics. We
can only scratch on the surface of the beautiful issue of the Chern-Simons theory
[172].
Identical Particles and their configuration spaces
Configuration spaces of systems of many identical particles

We denote the regular coordinate space of a one-particle system by X and we
assume N identical particles to be moving in this space. Usually the configuration
space of this system is described by X N , the Cartesian product of N one-particle
spaces.
Since the particles are identical, no distinction can be made between points
in X N that differ only in the ordering of the particle coordinates. Thus the two
points

x = (x1 , ...xN )
(16.1)
x0 = p(x) = (xp1 (1) , ..., xp1 (N) ),
where p is a permutation of the coordinates, both describe the same physical configuration of the system and therefore must be identified. X N does not describe
our physical configuration. The true configuration manifold is X N /SN , where the
permutation group SN is divided out of X N . This space identifies points which
represent the same configuration of the physical system and therefore describes
our system well.
X N /SN is locally isomorphic to X N , except at its singular points, where two
or more particles coincide in position. The difference between the manifolds lies
in their global properties. X N has only regular points, whereas the singularities
give a topological structure to X N /SN .
In our discussion we suppose our particles to be hard core, i.e. two or more
particles cannot occupy the same position at the same time. Topologically, this
means, that we exclude the singular points from the configuration manifold. The
excluded points can be written as a generalized diagonal
= {(x1 , ..., xN ) X N |xi = xj , i 6= j}.
(16.2)
Hence, the configuration manifold for a system of N indistinguishable particles

in n dimensions is given by
MnN = (XN )/SN .
(16.3)
In the following, we suppose the one particle space X to be the n-dimensional

Euclidian manifold E n (E n Rn ). We can introduce center of mass (c.m.) coor-
Braid group statistics in quantum mechanics

dinates
N
1 X
X=
xi ,
N i=1
423
(16.4)
where xi E n are the coordinates of the ith particle. Now, the configuration
manifold of a classical system of N identical hard-core particles can be written
as a Cartesian product,
n
MnN = E n rN
(16.5)
n
is a relative space, describing the relative
where E n is the c. m. space and rN
motions of the particles. It has got nN n degrees of freedom.
Configuration spaces of two-particle systems

The relative space r2n is the result of identifying x = x1 x2 with x = x2 x1
in E n . It has one singular point at x = 0, corresponding to a coincidence of
the position of the two particles. As the singularity is excluded from MnN , the
relative space r2n can be written as the Cartesian product
r2n = (0, ) RP n1 .
(16.6)
It consists of the positive real line (0, ), giving the length |x| of a vector x in
E n , and the real projective space RP n1 . At this point we need to explain some
mathematical terms and definitions:
Definition 1. The Real Projective Space RP n1 is the set of lines p Rn
through 0. Its dimension is dim RP n1 = n 1 [14].
One can describe the projective space in the following way:
RP n1 = (Rn {0})/R ,
(16.7)
where R denotes the set R {0}. The elements of RP n1 are equivalence classes
of vectors v 6= 0 in R under the relation v v, R . One can visualize
RP n1 as the n 1 dimensional sphere S n1 with antipodal points identified.
Definition 2. Two paths q : [0, 1] M and q 0 in a manifold M are called
homotopic, if there exists a continuous function H : [0, 1] [0, 1] M, such
that H(0, [0, 1]) q([0, 1]) and H(1, [0, 1]) q 0 ([0, 1]).
If q(0) = q(1), the path is called closed path or loop. All loops in M that are
homotopic form a class.
Definition 3. The group generated by all equivalence classes of loops in M, is
called the first homotopy group or fundamental group 1 (M).
Example 1. In Rn , all loops can be deformed into each other, therefore the fundamental group contains only one element, i.e. it is trivial what we write as
424
Topology in Physics
Figure 16.1: The relative space r22 of two two-dimensional particles is the plane
with pairs of opposite points x and x identified. The identification may be
effected by cutting the plane along a line l from the origin 0 and then folding it
into a circular cone.
1 (Rn ) ' {1}. The same is true on the n-dimensional sphere S n , n > 1. All
loops can be continuously deformed into each other, 1 (S n ) ' {1}.
On S 1 this is not true: If a path encircles S 1 twice, it cannot be continuously
deformed into a path that encircles S 1 four times, for example. It is also important, into which direction one walks. So the fundamental group of S 1 is isomorphic to the set of integers with the addition as group action, 1 (S 1 ) ' {(Z, +)}.
Example 2. RP 0 is a point and has no more structure. RP 1 is a circle, its
fundamental group is 1 (RP 1 ) ' Z. It is countably infinite, as we have seen
above. For n > 2, 1 (RP n1 ) ' Z/2Z and has got two elements. It is often
written as Z2 . We will take a closer look at these properties later.
At this point, one can see the essential differences among one-, two- and
three- or higher-dimensional Euclidian spaces for the first time. Now we will take
a closer look at Mn2 in different dimensions:
Two identical particles in two dimensions
The configuration manifold of two identical particles moving in E 2 is
M22 = E 2 r22 .
(16.8)
The relative space r22 is the plane E 2 0 with the points x and x identified.
This manifold can be seen as a circular cone as we see in figure 16.1.
425
Figure 16.2: The parallel transport of a tangent vector v around two different
closed curves on the cone.
A cone is globally curved, although it is locally flat everywhere except at the
singular vertex. Now we watch the parallel transport of a tangent vector. For
this we may map back onto the plane. The mapping is isometric (by definition),
and parallel transport on the cone becomes the familiar parallel transport in the
plane. Note, that a tangent vector v at the point x E 2 is identified with the
vector v at x. We see in figure 16.2 that the parallel transport around a
closed curve on the cone r22 changes v into (1)m v, m Z being the number of
revolutions or winding number of the curve around the vertex.
More generally speaking, we see, that in n dimensions, n N, there are two
classes of equivalent closed curves with respect to the transport of a tangent vector
v in the relative space r22 . One class does not change v, while the other class
changes v into v. A closed curve of the first class connects a point (x1 , x2 ) E 2n
continuously with itself, while a closed curve of the second class connects (x1 , x2 )
continuously with (x2 , x1 ) in E 2n .
Two identical particles in three dimensions
The physically most interesting case is of course X = E 3 . We have already seen
that r23 = (0, ) RP 2 , where RP 2 is the two-dimensional real projective space.
One can get it by identifying antipodal points on the two-dimensional sphere S 2 .
We get a visualization of RP 2 by dropping the southern hemisphere of S 2 and
identifying antipodal points on the equator (figure 16.3).
A tangent vector v r23 is changed into v by the parallel transport of a
closed curve q2 , which is a closed curve in RP 2 connecting opposite points on
426
Topology in Physics
Figure 16.3: The real projective space RP 2 can be represented as the northern
hemisphere with opposite points on the equator being identified.
Figure 16.4: Examples of a contractible (q1 ) and a non-contractible (q2 ) loop on

RP 2 .
the sphere. Such a curve encircles the singular point x = 0 once. If a closed
curve q1 encircles the singularity twice, it can be be continuously contracted to
a point without having to pass through the singularity, as we see in figure 16.4.
This is a completely different picture as in the two-dimensional case, where loops
with different winding numbers can never be continuously deformed into each
other. Now we also understand why the fundamental group of RP n1 , n > 2 has
got only two elements, i.e. 1 (RP n1 ) ' Z2 , whereas it is countably infinite for
n = 2.
Quantization of two-particle systems
How can we quantize two-particle systems? The configuration space of the identical particles is locally isometric to that of two nonidentical particles, except at
the singularity. The difference in global topology of the two systems does not
show up, unless we study a situation where the particles may physically inter-
427
change positions in the course of time evolution. For particles being far apart, it
is of no importance whether they are identical or not.
On the other hand, there is no obvious way to quantize a theory with a
curved configuration space, which even has singularities. We will try to follow
the simplest way, trying not to impose unnecessary restrictions on the theory.
We follow the Schrodinger quantization scheme and assume, that the state of the
system is given by a quadratically integrable function defined on the configuration
space. The problem is to define the free Hamiltonian, by taking properly care
of the physical effects of the singular points. As usual, an interaction between
particles will be described by adding a potential to the free particle Hamiltonian.
In two and more dimensions, the configuration manifold Mn2 is not flat, even if
all singular points are excluded. The presence of singularities is revealed through
a global curvature, which we studied in the section before in terms of the parallel
transport of tangent vectors around closed curves. In a similar way we introduce
the concept of parallel displacement of state vectors.
First, we introduce for each point x in the configuration manifold a corresponding, one dimensional Hilbert space hx , called a fiber. We assume the state
of the system to be described by a continuum of vectors (x) hx . is
assumed to be a single-valued function over the configuration manifold, whose
function value (x) at the point x is a vector in hx . If a normed basis vector x
is introduced for each space hx , then the complex-valued wave function (x) is
just the coordinate of the vector (x) relative to that basis:
(x) = (x)x .
(16.9)
It is clear that the function (x) will depend on the set of basis vectors, or
gauge {x }. A change in this set causes a gauge transformation:
x 0x = ei(x) x
0
(x) (x) = e
i(x)
(x)
(16.10)
(16.11)
In order to define a gauge-invariant derivative of the functions (x) we need

the concept of parallel displacement. Let us denote the linear operator which
transports the vectors of hx into hx0 parallelly by P (x0 , x). The transport is
along a continuous curve joining x to x0 . The parallel displacement in general may
depend on the curve, but we assume that the infinitesimal parallel displacement
P (x + dx, x) is uniquely defined. P (x0 , x) is assumed to be always a unitary
operator. Finally we assume that it is possible, at least locally, to choose the
gauge {x } in such a way that the rule of infinitesimal parallel displacement is
of the form

P (x + dx, x)x = 1 + i dxk bk (x) x+dx .
(16.12)
428
Topology in Physics
The covariant derivative is
ibk (x).
(16.13)
xk
The functions bk are determined partly by the dynamics of the systems and
partly by the choice of the gauge {x }. To make P (x + dx, x) unitary, they
must be real valued. The noncommutativity of the components of the covariant
derivative is measured by the gauge-invariant quantity
k =
fkl = i[k , l ] =
bk
bl
.
xk xl
(16.14)
In the case of parallel transport of tangent vectors, it corresponds to the curvature

tensor.
In the quantum theory of a charged particle in a magnetic field, we can find
a similar formulation, introduced by Weyl. The force field corresponds to the
antisymmetric tensor fkl , the vector potential to bk (x).
In the present case we do not want fkl to describe a force field. Therefore we
assume that fkl (x) = 0 for all x except at the singular points of the configuration manifold, where fkl is undefined. It follows, that a vector hx will be
unchanged by the parallel transport around any closed curve not encircling the
singularity. If is parallely transported around the singularity m times, it will
be transformed into (Px )m , where Px is a linear, unitary operator acting on hx .
Since the Hilbert space is one-dimensional, Px is just a phase factor,
Px = ei ,
(16.15)
where (R mod 2). Px transforms like

Px0 = P (x0 , x)Px P (x0 , x)1 = ei .
(16.16)
Therefore, the parameter must be independent of the position x, thus it is a

characteristic value of the given two-particle system.
We call the statistics of the system. As we will see below, it describes the
many body nature of the system. A system is bosonic or fermionic for = 0 or
= 1 respectively. In two dimensions intermediate states exist, which we call
anyons.
The field bk (x) has a dynamical effect through the gauge-invariant differentiation operator k . In the case we are considering, bk can be transformed into zero
by choosing the basis vectors x in a particular way. When the basis vector x
is given at some arbitrary point x, then we define the basis vectors for all other
points by parallel displacement of this x . When fkl vanishes, this procedure defines a gauge where bk vanishes. On the other hand, the complex wave function
(x) will be multivalued in this gauge, since all the basis vectors x , ei x ,
429
e2i x , etc. are generated by parallel transport of x around different closed

curves.
We see now that we have managed to transfer the dynamical effect of the singularities from the differentiation operator, and therefore from the Hamiltonian,
to the multivalued character of the wave function (x).
In one dimension the topology is trivial. For a solution of two particles moving
in one dimension, the interested reader may refer to [171], p. 11f.
An example in two dimensions
To discuss the two-dimensional case, we investigate two anyons in a harmonic
well [171], p. 15f and [173] p. 958.
The Hamiltonian of two anyons in an harmonic well is given by
2

~2
2
1
+ 2 + m 2 (x21 + x22 )
H=
(16.17)
2
2m x1 x2
2
where x1 and x2 are the coordinates of the anyons. When we write the
Hamiltonian in terms of the center of mass coordinates X and relative coordinates
x, it becomes

2 2

~2 2
~
1
2 2
2 2
H=
+ m X +
+ m x
(16.18)
4m X2
m x2 4
We observe that the center of mass part and the relative part completely split
off and therefore can be treated separately. As the center of mass space is topologically trivial, the motion of the center of mass is the usual harmonic oscillator
in two dimensions. The eigenfunctions should be quadratically integrable and
nonsingular at the origin. This problem can be solved as in [174], p. 102 ff.
For the discussion of the relative motion, we introduce polar coordinates r
and . The relative space is topologically nontrivial because of the singularity.
As we have seen above, the wavefunction obtains an additional phase when the
two particles are interchanged:
(r, + 2) = ei (r, ).
(16.19)
There is no obvious reason to restrict to the bosonic and fermionic cases

only. Therefore, we get a continuum of intermediate cases connecting those two.
With the gauge transformation
(r, ) 0 (r, ) = ei 2 (r, )
(16.20)
0 (r, + 2) = 0 (r, ).
(16.21)
we find
430
Topology in Physics
We can solve the problem, where the wave function behaves as in (16.21) in
the usual way. With the transformed Hamiltonian

2 !
2
2
~
1
H 0 = ei 2 Hei 2 =
+
+ 2
+i
+ m 2 r2 (16.22)
2
m r
r r r
2
4
the Schrodinger equation for the relative motion becomes
E 0 = H 0 0 .
(16.23)
We can split off the angular motion by choosing the eigenfunctions as

0 (r, ) = eil R(r),
(16.24)
where l is an integer. The radial function is then determined by the equation

2

d
1 d
4
2 1 m2 2 2 mE
+
l+
r + 2 R = 0.
(16.25)
dr2 r dr r2
2
4 ~2
~
Apart from the values of l + /2, this is the ordinary radial equation of the
harmonic oscillator in two dimensions which can be solved as above. We find the
allowed energies to be

1

En,l = 2~ n + l + +
(16.26)
2
2
where n N. We see that the value influences the energy spectrum of the
system. An example is shown in figure 16.5.
The three-dimensional case
We have seen before that a closed curve encircling the singularity twice can be
continuously contracted to a point without passing through the singularity. This
implies for the operator Px the additional condition
Px2 = 1.
(16.27)
Therefore, Px = 1, and the statistics can only take the values 0 and 1. This
means, that there can only be fermions and bosons in higher dimensions.
Artins braid groups
So far we have discussed only the cases with two particles. Now we will talk about
systems with more than two particles. As we have seen before, the configuration
manifold of an Euclidian system of N particles in n dimensions is given by
MnN = (E nN )/SN .
(16.28)
431
Figure 16.5: The energy spectrum of the relative motion of two anyons in an
harmonic well. It is shown for three different statistics, namely the bosonic
= 0, the fermionic = 1 and an intermediate case = 23 . The degeneracies of
each level are given in the figure.
432
Topology in Physics
The calculation of the fundamental groups of manifolds like MnN in different

dimensions n is an important topic of algebraic topology. It was solved in the
early 1960s [175].
Theorem 1. Let MnN = (E nN )/SN be the configuration manifold of a system
of N particles in n dimensions. Then the fundamental groups of this manifold
are
1 (M1N ) ' {1},
1 (M2N ) ' BN ,
(16.29)
(16.30)
1 (M3
N ) ' SN ,
(16.31)
where BN is the braid group of N strands (to be defined later) and SN is the
permutation group.
The result for n 3 is the result we expected. The bosonic case is described
via totally symmetric one-dimensional representations of the permutation group
SN . The fermionic case is described by totally antisymmetric one-dimensional
representations of SN . We observe that S2 ' Z2 .
The two-dimensional case is totally different from the higher-dimensional case.
The fundamental group here is Artins braid group of N strands.
Definition 4. The braid group of N strands, BN , is an infinite non-Abelian group.
It is generated by N 1 elementary moves i , satisfying the following properties:
i i+1 i = i+1 i i+1
(16.32)
i j = j i
(16.33)
for i = 1, ..., N 2 and
for |i j| 2. The inverse element of i is denoted by i1 , the identity by 1.

The elementary moves are best described by a pictorial representation as follows: Given N vertical strands, the generator i acts on them by simply braiding
the ith strand around the (i + 1)th in a definite way as shown on the left hand
side of figure 16.6. On the right hand side we see the move i1 , undoing i . One
easily sees, that B2 ' Z.
A generic braid is a word in the generators i and their inverses i1 , which
can be arranged using the relations (16.32) and (16.33). The usual convention is
that generators on the right act first.
In general i2 6= 1. If i2 = 1 for all i, then the braid group reduces simply to
the permutation group SN . Now we will look at an example.
Example 3. We assume a system of three identical particles. At the time t the
configuration is the one displayed on the left hand side of figure 16.7. We can describe it by listing the azimuthal angles of all possible pairs of particles measured
433
Figure 16.6: Pictorial representation of the elementary moves i and i1 respectively.
Figure 16.7: Configuration of the three identical particles at times t and t0 .
434
Topology in Physics
with respect to some arbitrary axes. In our case the angles are
12 (t) = 0,
13 (t) = ,
23 (t) = ,
where

ij = arctan
yj yi
xj xi
(16.34)
(16.35)
(16.36)

.
(16.37)
At a certain time t0 the particles should reach the positions shown on the right
hand side of figure 16.7. Now, the winding angles are
12 (t0 ) = + ,
13 (t0 ) = + ,
23 (t0 ) = .
(16.38)
(16.39)
(16.40)
Because of the indistinguishability the positions at times t and t0 are in fact

the same, even if the particles 1 and 3 have been interchanged. We notice that
ij (t0 ) ij (t) is not in general an integer multiple of . However, the relation
X
X
ij (t0 )
ij (t) = n
(16.41)
i<j
i<j
is always true, where n is an integer, in our example n = 3. The evolution from

t to t0 can be seen as the braiding 1 2 1 as shown in figure 16.8. The strands
are non intersecting world lines in [0, 1] R2 as described in [176], chapter 8.1.
Before we go on, we repeat some facts about path integrals. According to the
standard path-integral formulation of quantum mechanics, the amplitude for a
two dimensional system which evolves from the configuration q at time t to the
configuration q 0 at time t0 is given by the Kernel
Z q(t0 )=q0
R t0
i
0 0
0 0
)]
K(q , t ; q, t) = hq , t |q, ti =
Dq e ~ t d L[q( ),q(
(16.42)
q(t)=q
R q(t0 )=q0
where L(q, q)
is the Langrangian of the N -particle system. The symbol q(t)=q Dq
denotes the sum over all paths connecting q at time t with q 0 at time t0 . The
Kernel K(q 0 , t0 ; q, t) evolves the single-valued wave-function (q, t) according to
Z
0
(q , t ) =
Z
dq hq 0 , t0 |q, ti hq, t|i
(16.43)
dq K(q 0 , t0 ; q, t)(q, t).
(16.44)
M2N
=
M2N
435
Figure 16.8: The interchange of the particles 1 and 3, seen as braiding 1 2 1 .

Now, we set q 0 = q and hence describe loops. As all homotopic loops are
equivalent, we can write the sum over all loops in (16.42) into a sum over homotopic classes 1 (M2N ) and into a path integral in each class. Therefore,
(16.42) may be rewritten as
K(q, t0 ; q, t) =
K (q, t0 ; q, t)
1 (M2N )
X
1 (M2N )
q (t0 )=q 0
(16.45)
i
Dq e ~
R t0
t
d L[q ( ),q ( )]
(16.46)
q (t)=q
We can interpret this formula as a decomposition of the amplitude K(q, t0 ; q, t)

into the subamplitudes K (q, t0 ; q, t) to which only homotopic loops contribute.
With such a decomposition it is clear that in principle we can assign different
weights to the different subamplitudes K (q, t0 ; q, t), provided that we preserve
the conventional rules for the composition of probabilities. Thus, instead of
(16.46) we can write
0
K(q, t ; q, t) =
X
1 (M2N )
q (t0 )=q
()
Dq e ~
R t0
t
d L[q ( ),q ( )]
(16.47)
q (t)=q
where () is some complex number. If our amplitude should be a probability

amplitude, our weights () cannot be arbitrary. In fact, since we want to
436
Topology in Physics
maintain the usual rule for combining probabilities,

K(q 00 , t00 ; q, t) = hq 00 , t00 |q, ti
Z
=
dq 0 hq 00 , t00 |q 0 , t0 i h0 q,0 t|q, ti
2
M
Z N
dq 0 K(q 00 , t00 ; q 0 ,0 t)K(q 0 , t0 ; q, t)
=
(16.48)
(16.49)
(16.50)
M2N
the weight must satisfy

(1 )(2 ) = (1 2 )
(16.51)
for any i 1 (M2N ). Statement (16.51) can be read differently as well: The
weights () must be one-dimensional representations of BN .
They are given by
(k ) = ei
(16.52)
for any k = 1, ..., N 1, where is a real parameter defined mod 2. It will be
identified with the statistics. Since in general k2 6= 1, is an arbitrary number.
In the elementary move k all winding angles ij remain constant except for
k,k+1 which changes by . Thus we can rewrite (k )
(k ) = eik,k+1 = ei
(k)
i<j
ij
(16.53)
where we have introduced

(k)
(k)
(k)
ij := ij (t0 ) ij (t) = i,k j,k+1 .
(16.54)
With this notation, we can easily generalize (16.53) to an arbitrary braiding

. The representation can then be written as
() = ei
()
R t0
P
i<j
()
d
d d
ij ( )
(16.55)
()
where the increment ij (t0 ) ij (t) has been written as an integral over an
()
evolution parameter . In general, the functions ij are very complicated. They
can only be specified when the dynamics of the particle is taken fully into account.
If we substitute (16.55) into (16.47) we obtain
K(q, t0 ; q, t) =
X
1 (Mn
N)
q (t0 )=q
Dq e
R t0
t
d L[q ( ),q ( )]
()
d
ij
i<j
d
. (16.56)
q (t)=q
When we define
0
L := L ~
X d()
ij
i<j
(16.57)
437
we see that the Kernel K(q, t0 ; q, t) is decomposed into subamplitudes of the same
weight each with respect to L0 . This is as if we were describing bosons:
X Z q (t0 )=q
R t0
i
0
0
K(q, t ; q, t) =
Dq e ~ t d L .
(16.58)
1 (Mn
N)
q (t)=q
In other words, we can deal with bosons governed by the Langrangian L0 instead of
anyons governed by the Langrangian L. We can trade anyonic statistics for some
kind of fictitious force and describe anyons as ordinary particles (for example
bosons) with an additional statistical interaction. This statistical interaction is,
of course, very peculiar and intrinsically topological in nature. In our example, it
is a total derivative. The addition of a total derivative to L does not change the
equations of motion, but it does change the statistical properties of the particle,
which is related to the global topological structure of the configuration space.
One can generalize from a system on the plane to systems of N identical
particles on compact Riemannian surfaces with the corresponding braid group

()N
BN () 1
(16.59)
SN
It has been classified for all kinds of Riemannian surfaces . The topology
of plays a very important role because it restricts the possible values of the
statistics , the number of particles N and the number of components of the
wavefunction. The interested reader will find more information about this topics
in [172], p. 19 ff.
Fractional statistics in the Chern-Simons gauge
In this section we will discuss one example for a system with fractional statistics
and give a very short introduction to the Chern-Simons theory.
The cyon system and its symmetries
The cyon system is composed of a non-relativistic particle with mass m and
electric charge e and a magnetic flux across the origin. The flux is created by
an infinitely long and thin solenoid oriented along the z-axis and passing through
the origin. As the motion in z direction is free, the relevant motion happens on
the (x, y) plane. The Langrangian of this system is
mv2 e
+ vA(r)
(16.60)
2
c
for |r| > 0, where v = r . In a convenient symmetric gauge, the magnetic vectorpotential A is given by

x
y
A(r) =
x + 2
y
(16.61)
2 x2 + y 2
x + y2
L=
438
Topology in Physics
where is the magnetic flux and x and y are the unit vectors pointing in x and
y direction respectively. In this two-dimensional case, the magnetic field B is a
pseudo-scalar. We obtain
B = A = (2) (r)
and
(16.62)
d2 rB = .
(16.63)
The canonical momentum is derived from the Langrangian via

p=
L
e
= mv + A(r).
v
c
(16.64)
We note, that the canonical momentum is different from the kinetic momentum
mv. With a Legendre transformation we get the Hamiltonian H of the system:
H =pvL=
2 mv2
e
1
p rA =
2m
c
2
(16.65)
The magnetic field and the vector potential are invisible when the Hamiltonian
is written in terms of the kinetic momentum. In fact, the Hamiltonian of the
cyon is numerically equal to that of the free particle. The non-trivial relationship
between the kinetic and canonical momentum is the only effect of the vortex
in classical mechanics. The non-trivial features get important in the quantum
theory.
As the Lagrangian is rotationally invariant, the canonical orbital angular momentum Jc is a constant of motion,
e
Jc = r p = r mv + r A
c
e
= r p = r mv +
2c
e
=J+
,
2c
(16.66)
(16.67)
(16.68)
where J is the gauge invariant kinetic angular momentum. Formula (16.66) can
be written as
e
Jc = J + r A
cZ
Z
1
1
2
=J
d x x E(t, x)B(t, x) +
d2 x [E(t, x) x A]
c
c
(16.69)
(16.70)
where E(t, x) represents the electric field created by the moving charge. It satisfies
the Gauss law E(t, x) = e (2) (x r(t)), where r(t) is the position of the particle
439
at time t. By substituting (16.62), we find that the second term in the right hand
side of (16.70) identically vanishes, so that we are left with
Z
1
JC = J +
d2 x [E(t, x) x A]
(16.71)
c
for the cyon system. The second term in (16.71) is a boundary term which is
dissipated away to infinity. Therefore only the kinetic angular momentum is
present on the cyon.
The spectrum of the canonical angular momentum is represented by the quantum mechanical operator
Jc = i~
(16.72)
where is the polar angle on the plane. When this operator acts on a single
valued wave-function with angular dependence eim , it becomes simply
Jc = ~m
(16.73)
with m Z. The spectrum of Jc is always the conventional one.

The kinetic angular momentum operator is given by
J = Jc
e
= i~
.
2c
2c
(16.74)
When it acts on a single valued wave function with angular dependence eim , it
becomes

e
J =~ m
,
(16.75)
hc
m Z. Therefore, the spectrum of J consists of integers shifted by e
.
hc
We call the kinetic angular momentum the spin of the cyon. More precisely,
we have
J(m = 0)
e
s=
= .
(16.76)
~
hc
In general, s is neither integer nor half-integer. Therefore we should expect the
cyon to be an anyon in general, if a connection between spin and statistics exists.
To establish the statistical properties, we consider two identical cyons with a
wavefunction (1, 2). We assume, that the electric charge and the magnetic flux
are tightly bound on each particle. Now we slowly move one cyon around the
other by a full loop and neglect both charge-charge and vortex-vortex interactions.
As we know from the Aharonov-Bohm effect, the wave function acquires a phase
e
= ei ~c
drA
(16.77)
when we move cyon 1 around cyon 2 in a closed loop . We can rewrite (16.77)
in terms of the magnetic flux using Stokes theorem
e
= ei ~c
drA
= ei ~c
d2 r B
= e2i ~c .
(16.78)
440
Topology in Physics
If one particle is rotated around the other, there are actually two contributions
to the phase. One is due to the motion of the first particle in the vortex field of
the second, and one is due to the motion of the second particle in the vortex field
of the first. The total phase acquired by the wavefunction (1, 2) under a full 2
rotation is then
2e
e2i ~c .
(16.79)
Therefore, the statistics of the cyon is
=
2e
.
hc
(16.80)
The spin s and the statistics are related in the usual way
= 2s.
(16.81)
The Chern-Simons construction of fractional statistics

Now we will present a simple construction of fractional statistics of many body
systems. In the Chern-Simons formulation, fractional statistics is implemented by
means of a local, long range, Abelian gauge interaction taking place in a (2 + 1)dimensional spacetime. Rather than affixing an external vector potential A to
an ordinary particle by hand, one promotes A to be the space component of a
dynamical (2+1) dimensional U (1) gauge field A . Its action is the Chern-Simons
action
Z
Z
SCS = dt LCS =
d3 x A A
(16.82)
2c
where is the completely antisymmetric tensor density. We will use the convention that Greek indices take the values 0, 1, 2 and are contracted with the
metric = diag(1, 1, 1). The space components are labeled by Latin indices and take the values 1, 2. We denote the three-vector x (ct, xi ) by x, so
that d3 x = c dt d2 x.
The action (16.82) is gauge invariant even if the Langrangian contains an
undifferentiated gauge field A . Under gauge transformations
A A + ,
(16.83)
where is a space-dependent parameter, the Chern-Simons Langrangian changes

only by a total derivative
LCS =

[( )A ],
2c
(16.84)
so that the action remains invariant.

Now we will couple the gauge field A to a matter system consisting of N nonrelativistic point particles of mass m and charge e, whose coordinates rI (t) serve
441
as dynamical variables. Here, the capital Latin indices take the values 1, . . . , N
and label the particles. We introduce the current
j (x) =
N
X
evI (t) (2) (x rI (t)) (c, j),
(16.85)
I=1
with vI (c, vI (t)) with the velocity vI (t) = r I (t). It clearly satisfies the
continuity equation
j = t + = 0.
(16.86)
We clearly see the meaning of j when we write the time and space components
explicitly:
(x) =
j(x) =
N
X
I=1
N
X
e (2) (x rI (t)),
(16.87)
evI (t) (2) (x rI (t)).
(16.88)
I=1
and j are the conventional charge and current density for point particles located
at rI (t) and moving with velocity vI (t).
We couple the conserved current j to the gauge field A in the standard
minimal way
Z
1
Sint = 2 d3 x j (x)A (x)
(16.89)
c
( N
)
Z
X
1
=
dt
e [vI (t) A(t, rI (t)) A0 (t, rI (t))] .
(16.90)
c
I=1
As kinetic term for the N particles, we take the non-relativistic action
!
Z
N
X
1
Smatter = dt
mvI2 .
2
I=1
(16.91)
Therefore, the total action for our System is

Z
S = Smatter + Sint + SCS =
dtL
(16.92)
with the total Langragian

L=
N
X
I=1
mvI2
1
+
c

c
d2 x j A + A A .
2
(16.93)
442
Topology in Physics
The equations of motion of the gauge field A are given by

L !
L

= 0.
A
A
(16.94)
Hence we find
j =
c
c
( A A ) = F .
2
2
(16.95)
Instead of equations of motion we call them identities, because they relate the
fields E and B to the matter currents j and . When we write (16.95) in components, we get
1 ij j
j ,
c
1
B = ,
Ei =
(16.96)
(16.97)
with the Chern-Simons magnetic field

B = A = 1 A2 2 A1 F12
(16.98)
and the Chern-Simons electric field

1
E i = t Ai i A0 F0i .
c
(16.99)
Equations (16.96) and (16.97) are very important because they tell us, that
the Chern-Simons field is not arbitrary, but uniquely prescribed. This is a big
difference to the cyon system, where we affixed an arbitrary external magnetic flux
to the charged particle. Especial equation (16.97) has remarkable consequences.
If we integrate this equation over a small two-dimensional disk CI which
includes only the I-th particle the left hand side yields the magnetic flux attached
to that particle, while the right hand side gives its charge, namely
Z
e
I =
d xB =
CI
d2 x
CI
N
X
e
(2) (x rJ (t)) = .
J=i
(16.100)
Equation (16.100) tells us that a particle possesses a flux = e/, whenever it

possesses a charge e. The Chern Simons automatically binds charge and flux to
the particles.
We want to deepen our understanding with analysing the corresponding Hamiltonian H which we get in the usual way as in [177], p. 153 f.
Z
X
1
i
H=
pIi vI +
d2 x, A L
(16.101)
c
I
443
where is the field, canonical conjugated to the gauge field A , namely

=
L
c
c
=
0 A 0 A = 0 A .
0 A
2 0 A
2
. We find, that 0 = 0 and j =

pIi is
pIi =
c ij
Ai .
2
(16.102)
The canonical conjugated momentum
L
e
= mvIi + Ai (t, rI (t)).
i
vI
c
(16.103)
Combining (16.101), (16.102), (16.103) and (16.87) gives us

H=
N
X
1
I=I
Z
mvI2
d2 xA0 (x) (B(x) + (x)) .
(16.104)
We impose = B as a constraint. Hence H becomes numerically equal to

the Hamiltonian of N non interacting particles
H=
N
X
1
I=1

mvI2
(16.105)
As in the cyon system, the non-trivial dynamics resides entirely in the relation
between the canonical and the kinetic momenta.
We fix A using the Weyl gauge A0 = 0 and the Coulomb gauge i Ai = 0.
We solve the constraint = ij i Aj by using equation (16.87) and get
N
e X ji (2)
(x rI (t)) = i Aj .
I=1
(16.106)
Equation (16.106) has the solution

e X ij rIj rJj

.
2 J6=I |rI rJ |2
(16.107)
N
2
X
1
e
H =
pI AI (r1 , . . . , rN .
2m
c
I=1
(16.108)
AiI (rI , . . . , rN ) =
The Hamiltonian can be written as
0
The prime on H denotes that we used the solution given on (16.107). This
effective vector potential is non-local since it depends on the positions of all particles, and in particular, when N = 1, it vanishes. The magnetic field associated
to A(r1 , . . . , rN ) is
(16.109)
BI = ij i AjI (r1 , . . . , rN ).
rI
444
Topology in Physics
Each particle sees the (N 1) others as vortices carrying a flux = e

(16.100). The non-local potential AI (r1 , . . . , rN ) seems divergent at first sight.
But this is not the case as the coincident points (rI = rJ ), where the singularities
would occur, are excluded from the configuration manifold M2N of the system.
We observe, that the solution (16.107) can be written as
e ij X
ln |rI rJ |

2 rIj J6=I
e X
=
IJ
2 rIj J6=I
AiI =
(16.110)
(16.111)
as well. IJ is the winding angle of particle J with respect to particle I, such

that
2

xI x2J
1
IJ = tan
,
(16.112)
x1I x1J
as we discussed in (16.37). The Langrangian corresponding to (16.108) is then
N
X
1

e
L =
+ vI AI (r1 , . . . , rN )
2
c
I=1

N
N
X
1
e2 X X i
2
=
vI i IJ
mvI +
2
2c
rI
I=1
I=1 I6=J

N
X
1
e2 X i
2
=
mvI +
(vI vJi ) i IJ .
2
2c I<J
rI
I=1
0
mvI2
(16.113)
(16.114)
(16.115)
With
d
IJ =
dt

i
i
vI i + vJ i IJ = (vIi vJi ) i IJ
rI
rJ
rI
(16.116)
it becomes simply
L0 =
N
X
1
I=1

mvI2
e2
2c
X d
IJ
dt
I<J
!
.
(16.117)
When we compare (16.117) with (16.57), we can directly read off the statistics of
our system
e
2
e2
=
=
.
(16.118)
=
2~c
2~c
2~c
As in the cyon system, the statistics are not integer in general. Additionally, we
mention, that the statistics in the Chern-Simons system is twice the statistics in
the cyon system which is explained in detail in [172], p. 34 f.
Bibliography
[1] W. Fulton and J. Harris, Representation Theory, Springer, 2004.
[2] H. Georgi, Lie Algebras in Particle Physics, Benjamin/Cummings Publishing Company, Inc., 1982.
[3] M. E. Peskin and D. V. Schroeder, An introduction to quantum field theory,
chapter 15,20, Addison-Wesley, Massachusetts, 1995.
[4] H. Huang, Quarks, leptons and gauge fields, chapter 3, 4, World scientific,
Singapore, 1992.
[5] P. Goddard and D. I. Olive, Magnetic monopoles in gauge field theories,
Rep. Prog. Phys. 41, 1375 (1978).
[6] S. Coleman, Aspects of Symmetries, chapter 5.2, Cambridge University
Press, Cambridge, 1984.
[7] L. ORaifeartaigh, Group structure of gauge theories, chapter 8, Cambridge
University Press, Cambridge, 1987.
[8] W. Greiner, Eichtheorie der schwachen Wechselwirkung, chapter 1,2,3,4,
Verlag Harri Deutsch, Thun, 1995.
[9] A. A. Abrikosov, Fundamentals of the Theory of Metals, chapter 5, Elsevier
Science Publishers B.V., Amsterdam, 1998.
[10] J. Jackson, Klassische Elektrodynamik, chapter 6, de Gruyter Verlag, 2002.
[11] A. Wachter and H. Hoeber, Repetitorium Theoretische Physik, chapter 2,
Springer Verlag, 2005.
[12] Goddard, P. and Olive, D. I., New Developments in the Theory of Magnetic
Monopoles, Rept. Prog. Phys. 41, 1357 (1978).
[13] Harvey, J. A., Magnetic monopoles, duality, and supersymmetry, (1996),
hep-th/9603086.
445
446
Topology in Physics
[14] T. Broecker, Lineare Algebra und Analytische Geometrie, chapter 3,

Birkhauser Verlag, 2003.
[15] V. Rubakov, Classical Theory of Gauge Fields, Princeton University Press,
2002.
[16] N. Manton and P. Sutcliffe, Topological solitons, chapter 3,8, Cambridge
University Press, 2004.
[17] L. D. Faddeev, Nonlocal, nonlinear and nonrenormalisable field theories,
Proc. Int. Symp., Alushta (Dubana: Joint Institute for Nuclear Research)
, 207 (1976).
[18] S. Coleman, New Phenomena in Subnuclear Physics, Proc. 1975 Int. School
of Physics Ettore Majorana , 297 (1975), ed Zichichi, A.
[19] A. S. Schwarz, Magnetic Monopoles in Gauge Theories, Nucl. Phys. B112,
358 (1976).
[20] E. Corrigan, D. I. Olive, D. B. Fairlie, and J. Nuyts, Magnetic Monopoles
in SU(3) Gauge Theories, Nucl. Phys. B106, 475 (1976).
[21] J. Arafune, P. G. O. Freund, and C. J. Goebel, Topology of Higgs Fields,
J. Math. Phys 16, 433 (1975).
[22] A. Hatcher, Algebraic Topology, chapter 4, Cambridge University Press,
2002.
[23] N. S. Manton, The Force between t Hooft-Polyakov Monopoles, Nucl. Phys.
B126, 525 (1977).
[24] N. Ashcroft and N. Mermin, Festkoerperphysik, Oldenbourg Verlag, 1999.
[25] J. Friedel, Dislocations, Pergamon Press, New York, 1964.
[26] A. Cottrell, Theory of Crystal Dislocations, Gordon and Breach Science
Publishers, New York, 1965.
[27] D. Hull, Introduction to Dislocations, Pergamon Press, New York, 1965.
[28] M. Peach and J. Koehler, The Forces Exerted on Dislocations and the Stress
Fields Produced by Them, Phys. Rev. Lett. 80, 436 (1950).
[29] F. Nabarro and A. Kosevich, Dislocations in Solids Vol.1, North-Holland
Publishing Group, Amsterdam, 1979.
[30] P. Rosakis, Supersonic Dislocation Kinetics from an Augmented Peierls
Model, Phys. Rev. Lett. 86, 95 (2001).
References
447
[31] A. Rosakis, O. Samudrala, and D. Coker, Cracks Faster than the Shear
Wave Speed, Science 284, 1337 (1999).
[32] J. Eshelby, Supersonic Dislocations and Dislocations in Dispersive Media,
Proc. Phys. Soc. (London) B69, 1013 (1956).
[33] N. Manton and P. Sutcliffe, Topological solitons, chapter 1-5, Cambridge
University Press, 2004.
[34] http://maths.dur.ac.uk/ dma0saa/Solitons/solitons.pdf, Solitons, 2005.
[35] N. P. Ong, Transport studies near phase transitions in N bSe3 , Phys. Rev.
B 17, 3243 (1978).
[36] W. Fogle and J. H. Perlstein, Semiconductor-to-Metal Transition in the
Blue Potassium Molybdenum Bronze, K0.30 MoO3 ; Example of a Possible
Excitonic Insulator, Phys. Rev. B 6, 1402 (1972).
[37] E. B. Lopes, M. Almeida, and J. Dumas, Thermal conductivity of the
molybdenum blue bronze Rb0.3 M oO3 , Phys. Rev. B 42, 5324 (1990).
[38] G. Gr
uner, Density Waves in Solids, Addison-Wesley, 1994.
[39] A. J. Berlinsky, One-dimensional metals and charge density wave effects in
these materials, Rep. Prog. Phys. 42, 1243 (1979).
[40] W. P. Su, J. R. Schrieffer, and A. J. Heeger, Soliton excitations in polyacetylene, Phys. Rev. B 22, 2099 (1980).
[41] W. P. Su, J. R. Schrieffer, and A. J. Heeger, Erratum: Soliton excitations
in polyacetylene, Phys. Rev. B 28, 1138 (1983).
[42] W. P. Su and J. R. Schrieffer, Fractionally Charged Excitations in ChargeDensity-Wave Systems with Commensurability 3, Phys. Rev. Lett. 46, 738
(1981).
[43] T. L. Einstein and J. R. Schrieffer, Indirect Interaction between Adatoms
on a Tight-Binding Solid, Phys. Rev. B 7, 3629 (1973).
[44] V. Ginzburg and L. Landau, On the theory of superconductivity (in Russian), ZH. Eksp. Teor. Fiz. 20, 1064 (1950): Collected Papers of L.D.
Landau (ed. D. ter Haar), Gordon and Breach, New York, 1967.
[45] D. R. Tilley and J. Tilley, Superfluidity and Superconductivity, chapter 8,
Adam Hilger, Bristol, New York, 1990.
[46] G. H. Derick, Comments on nonlinear wave equations as models for elementary particles, J. Math. Phys. 5, 1252 (1964).
448
Topology in Physics
[47] R. S. Palais, The Principle of Symmetric Criticality, Commun. Math. Phys.

69, 19 (1979).
[48] N. Manton and P. Sutcliffe, Topological Solitons, chapter 1, 2, 3, 4, 7,
Cambridge University Press, Cambridge, 2004.
[49] L. Jacobs and C. Rebbi, Interaction energy of superconducting vortices,
Phys. Rev. B19, 4486 (1979).
[50] S. M. Girvin, Lecture Notes The Kosterlitz-Thouless Phase Transition,
chapter 2, 2000.
[51] P. M. Chaikin and T. C. Lubensky, Principles of Condensed Matter Physics,
chapter 5, 6, 9, Cambridge University Press, Cambridge, 1995.
[52] A. Tomita and R. Chiao, Observation of Berrys topological phase by use
of an optical fibre, Phys. Rev. Lett. 57, 937 (1986).
[53] M. Berry, Quantal phase factors accompanying adiabatic changes, Proc. R.
Soc. Lond. A 392, 45 (1984).
[54] A. Bohm, A. Mostafazadeh, H. Koizumi, Q. Niu, and J. Zwanziger, The
geometric phase in quantum systems, chapter 2,3,4, Springer, Berlin, 2003.
[55] Y. Aharonov and D. Bohm, Significance of Electromagnetic Potentials in
the Quantum Theory, Phys. Rev 115, 485 (1959).
[56] R. A. Webb, S. Washburn, C. Umbach, and R. Laibowitz, Observation of
h/e Aharonov-Bohm Oscillations in Normal-Metal Rings, Phys. Rev. Lett.
54, 2696 (1985).
[57] A. Bohm, The geometric phase in quantum systems, chapter 2.1-2.4, 4.1-4.3
and A.1, A.2, Springer-Verlag, Berlin, 2003.
[58] N. Manton, Topological solitons, chapter 2, Cambridge University Press,
2004.
[59] Y. Aharonov and A. Casher, Topological Quantum Effects for Neutral Particles, Phys. Rev. Lett. 53, 319 (1984).
[60] A. Cimmino, G. Opat, and A. Klein, Observation of the Topological
Aharonov-Casher Phase Shift by Neutron Interferometry, Phys. Rev. Lett.
63, 380 (1989).
[61] F. Schwabl, Quantenmechanik f
ur Fortgeschrittene, chapter 5.3 and 6.2,
Springer-Verlag, Berlin, 2004.
References
449
[62] Y. Ji et al., An elecronic Mach-Zehnder interferometer, Nature 422, 415

(2003).
[63] N. W. Ashcroft and N. D. Mermin, Solid State Physics, chapter 8,9,10,12,
Holt, Rinehart and Winston, Cornell University, 1975.
[64] M. Tabor, Chaos and Integrability in Nonlinear Dynamics, chapter 6, Wiley, Columbia University, New York, 1989.
[65] P. Kramer and M. Saraceno, Geometry of the Time-Dependent Variational
Principle in Quantum Mechanics, chapter 1, Springer-Verlag, Berlin, Heidelberg, New York, 1981.
[66] Q. Sundaram, G. Niu, Wave-packet dynamics in slowly perturbed crystals:Gradient corrections and Berry-phase effects, Phys. Rev. 59, 14915
(1999).
[67] Q. Bohm, A. Niu, The Geometric Phase in Quantum Systems, chapter 12,
Springer, Berlin Heidelberg New York, 2003.
[68] Z. Fang and N. Nagaosea, The Anomalous Hall Effect and Magnetic
Monopoles in Momentum Space, Science 302, 92 (2003).
[69] F. Haldane, Berry Curvature on the Fermi Surface: Anomalous Hall Effect
as a Topological Fermi-Liquid Property, Physical Review Letters 93, 206602
(2004).
[70] J. A. Lipa, J. A. Nissen, D. A. Stricker, D. R. Swanson, and T. C. P. Chui,
Specific heat of liquid helium in zero gravity very near the lambda point,
Phys. Rev. B 68, 174518 (2003).
[71] K. Huang, Statistical Mechanics, Wiley and Sons, New York, 1987.
[72] R. P. Feynman, Statistical Mechanics: A Set of Lectures, W. A. Benjamin,
Massachusets, 1972.
[73] H. Palevsky, K. Otnes, and K. Larsson, Excitation of Rotons in Helium II
by Cold Neutrons, Phys. Rev. 112, 11 (1958).
[74] A. J. Legget, Superfluidity, Rev. Mod. Phys. 71, 318 (1999).
[75] J. W. Negele and H. Orland, Quantum Many-Particle Systems, Addison
Wesley, New York, 1988.
[76] R. P. Feynman, Atomic Theory of the -Transition in Helium, Phys. Rev.
91, 1291 (1953).
450
Topology in Physics
[77] D. M. Ceperley, Path Integrals in the Theory of Condensed Helium, Rev.

Mod. Phys. 67, 279 (1995).
[78] E. L. Pollock and D. M. Ceperley, Path-Integral Computation of Superfluid
Densities, Phys. Rev. B 36, 8343 (1987).
[79] J. E. Hirsch, R. L. Sugar, D. J. Scalapino, and R. Blankenbecler, Monte
Carlo Simulations of One-Dimensional Fermion Systems, Phys. Rev. 26,
5033 (1982).
[80] M. Boninsegni, N. Prokofev, and B. Svistunov, Worm Algorithm for
Continuous-Space Path Integral Monte Carlo Simulations, Phys. Rev. Lett.
96, 070601 (2006).
[81] N. A. Nielsen and I. I. Chuang, Quantum Computation and Quantum
Information, Cambridge University Press, 2004.
[82] E.
W.
Weisstein,
Eulers
Totient
Theorem.,
http://mathworld.wolfram.com/EulersTotientTheorem.html 1, 1 (2003).
[83] U. Maurer, Vorlesungsskript Informationssicherheit und Kryptographie,
Unpublished, 2004.
[84] R. P. Feynman, Simulating physics with computers, Int. J. Theor. Phys.
21, 467 (1982).
[85] D. Deutsch, Quantum theory, the Church-Turing principle, and the universal quantum computer, Proc. Roy. Soc. London A400, 97 (1985).
[86] P. W. Shor, Algorithms for quantum computation: discrete logarighms and
factoring, Proceedings of the 35 Annual Symposium on the Foundations of
Computer Science, 1994, pp. 124-134.
[87] J. A. Walter, Kryptographie: Prinzipien und Anwendungen, Talk at University Bielefeld, 2003.
[88] S. Beauregard, Circuit for Shors algorithm using 2n+3 qubits, (quantph/0205095), 2002.
[89] L. K. Grover, Quantum mechanics helps in searching for a needle in a
haystack, Phys. Rev. Lett. 79, 325 (1997).
[90] C. Monroe, D. M. Meekhof, B. E. King, W. M. Itano, and D. J. Wineland,
Demonstration of a fundamental quantum logic gate, Phys. Rev. Lett. 75,
4714 (1995).
[91] M. H. Freedman, A. Y. Kitaev, M. J. Larsen, and Z. Wang, Topological
Quantum Computation, (quant-ph/0101025), 2002.
References
451
[92] A. Barenco et al., Elementary gates for quantum computation, (quantph/9503016), 1995.
[93] L. M. K. Vandersypen et al., Experimental realization of Shors quantum
factoring algorithm using nuclear magnetic resonance, (quant-ph/0112176),
2001.
[94] N. A. Gershenfeld and I. L. Chuang, Quantum computing with molecules,
Science 275, 350 (1997).
[95] D. P. DiVincenzo, The Physical Implementation of Quantum Computation,
(quant-ph/0002077), 2000.
[96] A. Imamoglu et al., Quantum information processing using quantum dot
spins and cavity-QED, (quant-ph/9904096), 1999.
[97] C. A. Sackett et al., Experimental entanglement of four particles, Nature
404, 256 (2000).
[98] R. Vrijen et al., Electron Spin Resonance Transistors for Quantum Computing in Silicon-Germanium Hetero-structures, (quant-ph/9904096), 1999.
[99] J. M. Kikkawa, I. P. Smorchkova, N. Samarth, and D. D. Awschalom,
Room-Temperature Spin Memory in Two-Dimensional Electron Gases, Science 277, 1284 (1997).
[100] T. Pellizzari, S. A. Gardiner, J. I. Cirac, and P. Zoller, Decoherence, Continuous Observation, and Quantum Computing: A cavity QED Model, Phys.
Rev. B 71, 224109 (2005).
[101] M. S. Sherwin, A. Imamoglu, and T. Montroy, Quantum Computation with
Quantum Dots and Terahertz Cavity Quantum Electrodynamics, (quantph/9904096), 1999.
[102] G. E. Moore, Cramming more components onto integrated circuits, Electronics 38 (1965).
[103] J. Preskill,
Topological Quantum Computation,
Lecture notes,
2000,
chapter 9 of Quantum Computation available online:
www.theory.caltech.edu/people/preskill/ph229.
[104] P. W. Shor, Scheme for reducing decoherence in quantum computer memory, Phys. Rev. A 52, 2493 (1995).
[105] J. Preskill, Fault-tolerant quantum computation, (quant-ph/9712048), 1997.
[106] D. Gottesman, A theory of fault-tolerant quantum computation, (quantph/9702029), 1997.
452
Topology in Physics
[107] A. M. Steane, Error correcting codes in quantum theory, Phys. Rev. Lett.
77, 793 (1996).
[108] A. R. Calderbank and P. W. Shor, Good quantum error-correction codes
exist, Phys. Rev. A 54, 1098 (1996).
[109] K. M. Svore, D. P. DiVincenzo, and B. Terhal, Noise Threshold for a FaultTolerant Two-Dimensional Lattice Architecture, (quant-ph/0604090), 2006.
[110] L. B. Ioffe et al., Topologically protected quantum bits using Josephson
junction arrays, Nature 415, 503 (2002).
[111] A. Y. Kitaev, Fault-tolerant quantum computation by anyons, (quantph/9707021), 2004.
[112] D. S. Rokhsar and S. A. Kivelson, Superconductivity and the Quantum
Hard-Core Dimer Gas, Phys. Rev. Lett. 61, 2376 (1988).
[113] P. W. Leung, K. C. Chiu, and K. J. Runge, Columnar dimer and plaquette
resonating-valence-bond orders in the quantum dimer model, Phys. Rev. B
54, 12938 (1996).
[114] R. Moessner and S. L. Sondhi, Resonating Valence Bond Phase in the
Traingular Lattice Quantum Dimer Model, Phys. Rev. Lett. 86, 1881
(2001).
[115] R. Moessner, S. L. Sondhi, and P. Chandra, Two-Dimensional Periodic
Frustrated Ising Models in a Transverse Field, Phys. Rev. Lett. 84, 4457
(2000).
[116] R. Moessner, S. L. Sondhi, and E. Fradkin, Short-ranged RVB physics,
quantum dimer models, and Ising gauge theories, Phys. Rev. B 65, 024504
(2001).
[117] K. S. Raman, R. Moessner, and S. L. Sondhi, SU(2)-invariant spin-1/2
Hamiltonians with RVB and other valence bond phases, Nature 414, 887
(2001).
[118] P. W. Anderson, Mater. Res. Bull. 8, 153 (1973).
[119] A. Ralko, M. Ferrero, F. Becca, D. Ivanov, and F. Mila, Zero-temperature
properties of the quantum dimer model on the triangular lattice, Phys. Rev.
B 71, 224109 (2005).
[120] M. Oshikawa and T. Senthil, Fractionalization, topological order, and quasiparticle statistics, (cond-mat/0506008), 2005.
References
453
[121] T. Ando, Y. Matsumoto, and Y. Uemura, Theory of Hall Effect in a TwoDimensional Electron System, J. Phys. Soc. Jpn 39, 278 (1975).
[122] K. v. Klitzing, G. Dorda, and M. Pepper, New Method for High-Accuracy
Determination of the Fine-Structure Constant Based on Quantized Hall
Resistance, Phys. Rev. Lett 45, 494 (1980).
[123] R. B. Laughlin, Quantized Hall conductivity in two dimensions, Phys. Rev.
B 23, 5623 (1981).
[124] J. P. Eisenstein and H. L. Stormer, The fractional quantum Hall effect,
Science 248, 1510 (1990).
[125] D. C. Tsui, H. L. Stormer, and A. C. Gossard, Two-Dimensional Magnetotransport in the Extreme Quantum Limit, Phys. Rev. Lett. 48, 1559
(1982).
[126] R. B. Laughlin, Anomalous Quantum Hall Effect: An Incompressible Quantum Fluid with Fractionally Charged Excitations, Phys. Rev. Lett. 50, 1395
(1983).
[127] X. G. Wen and Q. Niu, Ground-state degeneracy of the fractional quantum Hall states in the presence of a random potential and on high-genus
Riemann surfaces, Phys. Rev. B 41, 9377 (1990).
[128] D. Arovas, J. R. Schrieffer, and F. Wilczek, Fractional Statistics and the
Quantum Hall Effect, Phys. Rev. Lett. 53, 722 (1984).
[129] O. I. Motrunich and T. Senthil, Exotic Order in Simple Models of Bosonic
Systems, Phys. Rev. Lett. 89, 27704 (2002).
[130] G. Misguich, D. Serban, and V. Pasquier, Quantum Dimer Model on the
Kagome Lattice: Solvable Dimer-Liquid and Ising Gauge Theory in strongly
correlated systems, Phys. Rev. Lett. 89, 137202 (2002).
[131] R. Moessner and S. L. Sondhi, Ising models of quantum frustration, Phys.
Rev. B 63, 224401 (2000).
[132] T. Senthil and M. P. A. Fisher, Fractionalization, topological order, and
cuprate superconductivity, (cond-mat/0008082), 2000.
[133] T. Senthil and M. P. A. Fisher, Z2 gauge theory of electron fractionalization
in strongly correlated systems, Phys. Rev. B 62, 7850 (1999).
[134] T. Senthil and M. P. A. Fisher, Fractionalization in the cuprates: Detecting
the topological order, (cond-mat/0006481), 2000.
454
Topology in Physics
[135] D. A. Bonn et al., A limit on spin-charge separation in thigh Tc superconductors from the absence of a vortex-memory effect, Nature 414, 887
(2001).
[136] N. R. Cooper, N. K. Wilkin, and J. M. Gunn, Quantum Phases of Vortices
in Rotating Bose-Einstein Condensates, Phys. Rev. Lett. 87, 120405 (2001).
[137] R. Coldea, D. A. Tennant, and Z. Tyczynski, Extended scattering continua
characteristic of spin fractionalization in the two-dimensional frustrated
quantum magnet Cs2 CuCl4 observed by neutron scattering, Phys. Rev. B
68, 134424 (2003).
[138] M. H. Freedman, A. Y. Kitaev, M. J. Larsen, and Z. Wang, Topological
quantum computation, Bull. Amer. Math. Soc. 40, 31 (2002).
[139] M. H. Freedman, M. J. Larsen, and W. Z., A modular functor which is
universal for quantum computation, Comm. Math. Phys. 227, 587 (2002).
[140] M. H. Freedman, A topological modular functor which is universal for
quantum computation, Talk given at MSRI, 2000, available online:
http://www.msri.org/communications/
ln/msri/2000/qcomputing/freedman/1/index.html.
[141] A. Y. Kitaev, A Class of P, T -Invariant Topological Phases of Interacting
Electrons, (cond-mat/0307511), 2003.
[142] M. H. Freedman, C. Nayak, and K. Shtengel, A Line of Critical Points in
2+1 Dimensions: Quantum Critical Loop Gases and Non-Abelian Gauge
Theory, Phys. Rev. Lett. 94, 147205 (2005).
[143] N. E. Bonesteel, L. Hormozi, and G. Zikos, Braid Topologies for Quantum
Computation, Phys. Rev. Lett. 95, 140503 (2005).
[144] C. M. Dawson and M. A. Nielsen, The Solovay-Kitaev algorithm, (quantph/0505030), 2005.
[145] N. E. Bonesteel,
Braid Topologies for Quantum Computation,
Talk given at KITP, 2006,
available online:
http://online.kitp.ucsb.edu/online/qubit06/bonesteel/.
[146] F. E. Camino, W. Zhou, and V. J. Goldman, Aharonov-Bohm Superperiod
in a Laughlin Quasiparticle Interferometer, Phys. Rev. Lett. 95, 246802
(2005).
[147] S. Bravyi, Universal Quantum Computation with the = 5/2 Fractional
Quantum Hall State, Phys. Rev. A 73, 042313 (2006).
References
455
[148] M. H. Freedman, C. Nayak, and K. Walker, Towards Universal Topological

Quantum Computation in the = 5/2 Fractional Quantum Hall State,
(cond-mat/0512066), 2005.
[149] J. K. Slingerland and F. A. Bais, Quantum groups and nonabelian braiding
in quantum Hall systems, Nucl. Phys. B 612, 229 (2001).
[150] A. Y. Kitaev, Fault-tolerant quantum computation by anyons, Ann. Phys.
303, 2 (2003).
[151] J. Preskill, Introduction to quantum computation and information, World
Scientific, Singapore, 1998.
[152] S. A. Kivelson, D. S. Rokhsar, and J. P. Sethna, Topology of the resonating
valence-bond state: solitons and high-Tc superconductivity, Phys. Rev. B
35, 8865 (1987).
[153] X. G. Wen, Mean-field theory of spin-liquid states with finite energy gap
and topological orders, Phys. Rev. B 44, 2664 (1991).
[154] L. B. Ioffe and A. I. Larkin, Superconductivity in the liquid-dimer valencebond state, Phys. Rev. B 40, 6941 (1989).
[155] E. Manousakis, The spin-1/2 Heisenberg antiferromagnet on a square lattice
and its application to the cuprous oxides, Rev. Mod. Phys. 63, 1 (1991).
[156] L. Capriotti, A. E. Trumper, and S. Sorella, Phys. Rev. Lett. 82, 3899
(1999).
[157] W. Marshall, Proc. R. Soc. London, Ser. A 232, 48 (1955).
[158] G. Misguich and C. Lhuillier, Frustrated spin systems, World Scientific,
2005.
[159] B. Sutherland, Systems with resonating-valence-bond ground states: Correlations and excitations, Phys. Rev. B 37, 3786 (1988).
[160] G. Misguich, D. Serban, and V. Pasquier, Quantum dimer model with
extensive ground-state entropy on the Kagome lattice, Phys. Rev. B 67,
214413 (2003).
[161] P. Fendley, R. Moessner, and S. L. Sondhi, Classical dimers on the triangular lattice, Phys. Rev. B 66, 214513 (2002).
[162] A. Ioselevich, D. A. Ivanov, and M. V. Feigelman, Ground-state properties
of the Rokhsar-Kivelson dimer model on the triangular lattice, Phys. Rev.
B 66, 174405 (2002).
456
Topology in Physics
[163] V. Elser and C. Zeng, Kagome spin-1/2 antiferromagnets in the hyperbolic

plane, Phys. Rev. B 48, 13647 (1993).
[164] C. Zeng and V. Elser, Quantum dimer calculations on the spin-1/2 kagome
Heisenberg antiferromagnet, Phys. Rev. B 51, 8318 (1995).
[165] B. D. Josephson, Possible new effects in superconductive tunnelling, Phys.
Letters 1, 251 (1962).
[166] R. P. Feynman, The Feynman Lectures on Physics Vol. III, Pearson
Addison-Wesley, San Francisco, 2006.
[167] E. Simanek, Inhomogeneous Superconductors - Granular and Quantum
Effects, International Series of Monographs on Physics, Oxford University
Press, 1994.
[168] H. Grabert and M. H. Devoret, Single charge tunneling: coulomb blockade
phenomena in nanostructures, Plenum Press, New York, 1992.
[169] L. B. Ioffe, V. B. Geshkenbein, M. V. Feigelman, A. L. Fauchere, and
G. Blatter, Environmentally decoupled s-waved-waves-wave Josephson
junctions for quantum computing, Nature 398, 678 (1999).
[170] M. Nielsen and I. Chuang, Quantum Computation and Quantum Information, Cambrige University Press, 2000.
[171] J. M. Leinaas and J. Myrheim, On the Theory of identical particles, Il
Nuovo Cimento 37 B (1977).
[172] A. Lerda, Anyons, volume m14 of Lecture Notes in Physics, chapter 2, 3,
Springer Verlag, New York, 1992.
[173] F. Wilczek, Quantum Mechanics of Fractional-Spin Particles, Phys. Rev.
Lett. 49, 957 (1982).
[174] F. Scheck, Nichtrelativistische Quantentheorie, Theoretische Physik 2,
Springer Verlag, Berlin, Heidelberg, 2000.
[175] E. Fadell and L. Neuwirth, Configuration Spaces, Math. Scand. 10, 111
(1962).
[176] J. Frohlich, Vorlesungsskript Quantenmechanik II, chapter 8, 2005, given
at ETH Z
urich.
[177] F. Scheck, Klassische Feldtheorie, Theoretische Physik 3, Springer Verlag,
Berlin, Heidelberg, 2004.
References
457
[178] J. C. Davis and R. E. Packard, Superfuid He Josephson weak links, Rev.

Mod. Phys. 74, 741 (2002).
[179] E. Fradkin, Field Theories of Condensed Matter Systems, Frontiers in
Physics 82, Westview Press, 1991.
[180] L.-P. Levy, Magnetism and Superconductivity, Springer Verlag, Berlin,
2000.
[181] J. Blin, Energie Mutuelle de Deux Dislocations, Acta Metallurgica 3, 199
(1955).
[182] D. P. Landau and K. Binder, A Guide to Monte Carlo Simulations in
Statistical Physics, chapter 2, pages 726, Cambridge University Press,
Cambridge, 2000.
[183] J. Cardy, Scaling and Renormalization in Statistical Physics, chapter 1, 2,
3, 6, Cambridge University Press, Cambridge, 1996.
[184] M. Plischke and B. Bergersen, Equilibrium Statistical Physics, chapter 5,
pages 206213, Scientific Publishers, London, 1994.
[185] Y. Aharonov and D. Bohm, Significance of Electromagnetic Potentials in
the Quantum Theory, Phys. Rev. 115, 485 (1959).
[186] G. Gr
uner, The dynamics of charge-density waves, Rev. Mod. Phys. 60,
1129 (1988).
[187] K. Maki,
Creation of Soliton Pairs by Electric Fields in ChargeDensityWave Condensates, Phys. Rev. Lett. 39, 46 (1977).
[188] G. P. Collins, Computing with Quantum Knots, Scientific American April,
57 (2006).
[189] A. Y. Kitaev, Quantum computations: algorithms and error correction,
Russian Math. Survey 52:61, 1191 (1997).
[190] M. H. Freedman, C. Nayak, and K. Shtengel, Non-Abelian topological
phases in an extended Hubbard model, (cond-mat/0309120), 2003.

Katzgraber - Topology in Physics

Uploaded by

Copyright:

Available Formats

Katzgraber - Topology in Physics

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Katzgraber - Topology in Physics

Uploaded by

Copyright:

Available Formats

Topology in Physics

July 31, 2006

Cyril Stark (under the supervision of Urs Wenger)

2 Goldstone bosons and the Higgs mechanism . . . . 51

3 Magnetic monopoles: Dirac and Polyakov-tHooft

Philip Grech (under the supervision of Urs Wenger)

4 Dislocations in crystals . . . . . . . . . . . . . . . . . . . . 103

6 Solitons in density wave systems . . . . . . . . . . . . . 151

8 BKT transition in spin systems . . . . . . . . . . . . . . 197

9 Berry Phase I: Mathematical Concepts . . . . . . . . 227

10 Berry phase II: The Aharonov-Bohm effect . . . . 271

11 Berry Phase III: Berry phase and Bloch states . . 293

12 Superfluidity and winding numbers . . . . . . . . . . 311

13 Introduction to quantum computing . . . . . . . . . 339

14 Topologically protected quantum computing . . . 361

15 Topological quantum computing implementations 393

16 Braid group statistics in quantum mechanics . . . 421

c 2006 H. G. Katzgraber (Theoretische Physik ETH Z

Lie Algebras: a crash course

Definition of a Lie Group

Definition 1. Let G and H be two Lie groups. A homomorphism between Lie

where GL(G) denotes the set of all automorphisms on G. We observe

Lie Algebras: a crash course

This identity still explicitly depends on (g). Thus, we differentiate a second

Therefore, we can consider the map ad to be a bilinear map

Lie Algebras: a crash course

Definition (Representation). A representation of a Lie algebra g on a vector

Lie Algebras: a crash course

: Let be irreducible and let W be a complex linear subspace invariant

real Lie algebra su(n) and sl(n; C)

Properties of the exponential map:

Lie Algebras: a crash course

The bracket operation on g induces a bracket operation on the quotient space

Dk g := [g, Dk1 g].

The subalgebras Dk g are ideals in g.

Dk g := [Dk1 g, Dk1 g].

The subalgebras Dk g are ideals in g and Dk g Dk g for all k.

Theorem 3. Let g be an arbitrary complex Lie algebra and let gsemisimple =

Lie Algebras: a crash course

Proposition 5. The following is equivalent:

The goal of this section is to analyze the irreducible complex representations of

Assume that is an arbitrary irreducible finite-dimensional representation of

With the decomposition of V into eigenspaces of (H) we have brought a first

We summarize these results in the following

From the 3 facts that

Lie Algebras: a crash course

Consider the case m = 0:

(Y ) (X)((Y )(v)) + [(X), (Y )]((Y )(v))

Repeating these calculations yields

Since (Y )(m1) (v) is by assumption not equal to zero we get:

We can summarize the former results in the following theorem.

Lie Algebras: a crash course

Theorem 4. Let be an irreducible representation of sl(2, C) with representation

Definitions and first Observations

Lie Algebras: a crash course

where ranges over a finite subset of h as explained in remark 3. The subspaces

for every H h and X g .

for every H h, X g and Y g . This crucial calculation proves the following

for all H h, X g and v V . We deduce