Berry Phase & Geometric Phase: an

Introduction

Erik, Pillon

January 29, 2019

Acknowledgments
This set of lecture notes accompanies prof. Mukundan Seminar course on Geometric Phase,
taught from 10th to 15th March 2014.
The entire course is hosted on YouTube at the following address:

https://www.youtube.com/playlist?list=PLd9hKAUC3AZu1rM_6iFl1l7oYM-6r41Qq

These lecture notes are not endorsed by prof. Mukundan.

While I have tried to correct typos and errors made during the lectures (some helpfully
pointed out by YouTube commenters), I have also taken the liberty to add and/or modify
some of the material at various points in the notes. Any errors that may result from this are,
of course, mine.
Erik Pillon

i
Contents
1 Overview 1

2 Berry discovery of 1983-84 3

2.1 Simplified Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Adiabatic condition . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.1.2 Berry’s hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Berry original derivation with parameter space . . . . . . . . . . . . . . . . 8
2.2.1 Two fold degeneracies . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Aharonov-Anandan and Samuel-Bhandari generalizations 13

3.1 Aharonov-Anandan generalization (1987) . . . . . . . . . . . . . . . . . . . 13
3.2 First Mathematical Interlude . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Samuel and Bhandari generalization (1988) . . . . . . . . . . . . . . . . . . 16

4 The kinematic approach and the Bargmann’s Invariant 19

4.1 Bargmann’s Invariant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3 Second Mathematical Interlude . . . . . . . . . . . . . . . . . . . . . . . . . 20

Bibliography 21

iii
1 Overview
The work is given by a total amount of 6 sections; this first introductory section is an
overview of the subject. The second section is a discussion about the Berry original work:
first as a sempilified form and then as it was originally presented by prof. Berry in his
seminal paper Berry (1985).
The third section will cover the work of Aharonov and AnandanAharonov and Anandan
(1987) on the relaxation of the adiabatic hypothesis and, a year later, by Samuel and Bhandari
Samuel and Bhandari (1988) on the relaxation of the cyclicity condition. In this third section
we will also introduce a brief mathematical interlude and some geometrical consideration
about curvature and connection.
The fourth section is devoted essentially to the so called Kinematic Approach: we will bring
the concept of Bargman invariance and its connection to curvature and the interesting
application to the entire formalism.
Fifth section will be a second mathematical interlude on simplectic and riemannian mani-
folds.
Sixth section will be about the null phase idea.

1
2 Berry discovery of 1983-84
Berry discovery of 1983-841 was a new discovery on the context about adiabatic of quantum
mechanics and this work initiated a lot of work worldwide. In Berry’s derivation several
independent assumptions were made: initially this phase was called the Berry Phase for
everyone but over time this name changed to geometric phase and as we’ll see this concept
is relevant also in classical wave-optical situations. It’s quite remarkable that there are some
chances that this concept can be used in condensed matter context. On the other hand we
hope that the way we’re presenting this work can naturally point out the applications.
As we said many work have been done to relax the assumptions that Berry done in order
to defend Berry’s work under more general conditions. The first important step was taken
by Aharonov and Anandan (1987). The second important by S and A was taken in 1988
and a third successful step was taken in 1993 and these are the successful and successive
steps we’ll describe. Apart from these improvements, people were also looking for earlier
litteratures from different ideas much earlier then Berry. There are several of them but the
most important we will touch upon is the on of PancharatnamPancharatnam (1956) in 1957,
that is, 27 earlier than Berry work. The fact that the work of Pancharatnam was in the
direction of the geometric phase was pointed out by ... and ... in 1986. The other important
early work relevant in this subject was the one of Bargman and Valentine in 1964 mostly 20
years before Berry working on discuss inf Wigner theorem of 1931, a theorem that Wigner
had proved on how symmetric operators can be represented in quantum mechanics. So
the th itself is very early (1931); many people had tried to give alternative proof of Wigner
Theorem and one very important is given by Bargman in 1964, particularly elegant. The
fact that Bargman work was important in the discovery of the Berry phase was pointed out
and exploited by Syman and auth. in 1993
These lectures will describe all these thing and more mathematical relevant structures in a
more chronological structures, but will not be strictly chronological rigorous. You’ll find
that many features of QM which we we might be familiar with they will be re examined,
re-defined from the geometrical phase pov. When we will come to the kinematic approach
we will define some applications. This should give an overview and an idea of the scope of
these lectures.

2.1 Simplified Form

We are now going to present the original Berry’s work in a slightly simplified way.

1
We say 1983-1984 since the original work was apparently submitted in 1983 but was at first instance rejected.
However a preprint of his work must have been around since in 1983 another work on the Berry Phase
appeared on Physical Review Letter. That’s why sometimes the date can be misleading.

3
2 Berry discovery of 1983-84

Having a quantum mechanical system in mind and a general setting, we will mainly deal
with pure states H with a time dependent Hamiltonian H(t) governing the system and we
have a state vector describing the system ψ(t). The wavefunction must satisfy all times the
(time dependent) Schrödinger equation

d
i~ ψ(t) = H(t)ψ(t) (2.1) eq:2.1
dt
If the Hamiltonian had be time independent, a formal solution of the Schrödinger equation
would be easy to find because what we have to do is to formally take the Hamiltonian, find
a basis of our Hilbert space H by diagonalizing the Hamiltonian in its eigenfunctions and
eigenvalues and express our wavefunction as a linear combination of the basis element

Hψn = En ψn , n = 1, 2, . . . En real. {?}

For simplicity let us assume the spectrum to be discrete, non degenerate as well as constant
in time (we already know, though, En to be all real, because of the hermiticity of the
Hamiltonian).
What one has to do, then, in general, is to express the solution of the Schrödinger equation
ψ as a linear combination of the eigenfunctions of the Hamiltonian and for each element
add its evolution in time

cn e−iEn t/~ ψn .
X X
ψ= cn ψn → ψ(t) = (2.2) eq:2.3
n n

In principal this procedure is easy and really straightforward. ψn are called the stationary
states of the system and of course the ψn form a complete set of orthonormal vector basis
X
|ψn i hψn | = 1 {?}
n
(ψn , ψk ) = δnk {?}

where each ψn is defined up to an independent phase factor.

Let us now discuss the case of a time dependent Hamiltonian: generalizing what written
above
H(t)ψ(t)n = En (t)ψn (t), n = 1, 2, . . . En (t) real {?}

and as the Hamiltonian changes in time, then its eigenvalues do. Of course at each time the
eigenvalues form again a complete orthonormal set.
Remark 1. Every eigenfunction ψn (t) is obtained now by diagonalizing at each time t the
the Hamiltonian H(t). Even if ψn (t) are not anymore stationary states, the definition of
the eigenfunctions still holds up to a time dependent phase factor. This factor can be both
dependent on time and on n.

Now, there is no hope to recover the exact solution, even though you have solved the
eigenvalue problem for each time. What we can do is to generalize eq.(2.2) with
i
Rt
En (t0 ) dt0
cn (t)e− ~
X
ψ(t) = 0 ψn (t) (2.3) eq:2.4
n

and it will reduce to (2.2) in the case of time independence.

4
 2.1 Simplified Form

Now plugging (2.3) in (2.1) we get

X i Rt 
i 0 0
i~ c˙n (t)ψn (t) − cn (t)En (t)ψn (t) + cn (t)ψ˙n (t) e− ~ 0 En (t ) dt {?}
n ~
i
Rt
En (t0 ) dt0
= cn (t)En (t)ψn (t)e− ~ 0 {?}

so, erasing the equal terms, we obtain

i
Rt
En (t0 ) dt0
X 
c˙n (t)ψn (t) + cn (t)ψ˙n (t) e− ~ 0 = 0. ?eq:2.5?
n

and now we take the scalar product with the vector ψk (t)
i
Rt
(En (t0 )−Ek (t0 )) dt0
 
ck (t)e− ~
X
ċk = − 0 ψk (t), ψ̇n (t) ∀k {?}
n

Remark 2. The last equation is exact! There are no approximations involved so far!
 
Let’s focus for the moment on the term ψk (t), ψ̇n (t)

H(t)ψ(t)n = En (t)ψn (t), n = 1, 2, . . . {?}

and we differentiate with respect to time on both sides

! !
∂ ∂
H(t) ψn (t) + H(t)ψ̇n (t) = En (t) ψn (t) + En (t)ψ̇n (t) {?}
∂t ∂t

and again we take again the scalar product with a generic state ψk (t)
!
∂H(t)    
ψk (t), ψn (t) + Ek (t) ψk (t), ψ̇n (t) = Ėn (t)δnk + En ψk (t), ψ̇n (t) {?}
∂t

so it is appropriate to collect some terms, creating an energy difference, obtaining

!
  ∂H(t)
(En (t) − Ek (t)) ψk (t), ψ̇n (t) = −Ėn (t)δnk + ψk (t), ψn (t) ∀k, n {?}
∂t

and the previous result is true in general! There are no approximations involved at all!
If we restrict ourselves at the specific case k = n
!
∂H(t)
Ėn (t) = ψn (t), ψn (t) {?}
∂t

while in the case k 6= n we can straightforwardly derive from the eigenvalue problem 2
 
  ψk (t), ∂H(t)
∂t
ψn (t)
ψk (t), ψ̇n (t) = {?}
(En (t) − Ek (t))
So we found an explicit expression relating the relative energy gap in time, the time derivative
of the Hamiltonian and the scalar product of the time derivative of the wavefunction with
any other wavefunction. We remark that each time dependent wavefunction ψn (t) is defined
2
Remember that we assumed non degeneracy in energy spectrum in all the time

5
2 Berry discovery of 1983-84

up to a phase factor that can depend on n and may depend on time. With this freedom we
are left all alone.
 
From the normalization condition ψn (t), ψ̇n (t) = 1 we can already say conclude
 
< ψn (t), ψ̇n (t) = 0 ∀n

So from now on we agree to restrict the phase factor to be for each n such that
 
ψk (t), ψ̇n (t) = 0, ∀n (2.4) eq:requir

Remark 3. Making use of the requirement (2.4), the phase freedom is eliminated. Once we
chose a phase factor at time t = 0 for ψn (0), no more flexibility is left.

Rt
ωkn (t0 ) dt0
 
i
X
ċk (t) = − cn (t)e 0 ψk (t), ψ̇n (t) {?}
n6=k
!
X cn (t) i R t ωkn (t0 ) dt0 ∂H(t)
= e 0 ψk (t), ψn (t) ∀k (2.5) eq:2.11
n6=k ~ωkn (t) ∂t

Ek (t)−En (t)
where we defined ωkn (t) = ~
.
Remark 4. We stress again that Eq.(2.5) is obtained manipulating the Schrödinger equation
only letting the Hamiltonian to vary in time.

2.1.1 Adiabatic condition

Now we go to the adiabatic situation. The so called Adiabatic Theorem is a result originally
proven by Born&Fock in 1928 (see Born (1928) for further readings).
The main assumption we’ll do is that the Hamiltonian we’re considering is a slowly varying
operator, that is to say
∂H
(t) is "small" {?}
∂t
The term "small" will be quantitatively clear later on. Physically, it’s reasonable to say
that the quantities ψn (t), En (t) and cn (t) are expected to slowly change in time so the the
ordering on the energy levels will be maintained (i.e., no crossing levels are allowed).
Let’s then continue with the original work of Fock and Born, back to 1928: suppose to have
as initial condition ψ(0) = ψn (0), that is to say that the original initial state is equal to a
particular eigenvector of the Hamiltonian at time t = 0. So we have ck (0) = δkn (the same
chosen in (??))
Notice that our reasoning is perfectly consistent in the framework of the first order pertur-
bation theory; since the term ∂H
∂t
is explicitly appearing in (2.5), we can neglect the time
dependency of all the other terms, having that, for k 6= n,
!
1 iωkn t ∂H(t)
ċk (t) ' e ψk , ψn {?}
~ωkn ∂t

6
 2.1 Simplified Form

while for the n-th term we have cn (t) ' 1: it starts at 1 and stay fixed for all times. The
ċk (t) can be instead integrated and
!
i  iωkn t  ∂H(t)
ck (t) ' − 2 e − 1 ψk , ψn {?}
~ωkn ∂t
and so, while the term cn remains close to 1, all the other terms start from zero and oscillate
in time:
So,if !
1 ∂H
ψk , (t)ψn  ~ωkn , ∀k 6= n (2.6) eq:2.14
ωkn ∂t
is satisfied, then R t0
− ~i En (t) dt
ψ(t) ' e 0 ψn (t) {?}
and this ends the statement of the adiabatic theorem of quantum mechanics.
Remark 5. Eq.(2.6) is the quantitative statement of what we mean by "adiabatic condition".

2.1.2 Berry’s hypothesis

Now comes the step taken by Berry; Let’s take the hypothesis that the Hamiltonian H(t) is
cyclic, that is, H(0) = H(T ) for some T 3 .

Question: How does the approximate solution behave? Are they cyclic in a
similar sense?

Answer: the solution must be cyclic in some sense. Because of the non degeneracy and no
crossing levels, we have that the eigenvalues of H(T ) are the same of H(0) so
En (T ) =En (0) {?}
i
RT
ψ(0) = ψn (0) ⇒ ψ(T ) ' e− ~ 0
En (t) dt
ψn (T ) {?}

from the Adiabatic Theorem. And now we ask: is this a cyclic solution? The answer is still
yes but we have that the equality ψ(T ) = ψ(0) is only true apart from a phase, i.e.,
ψn (T ) = (n-dependent phase) × ψn (0)
?eq:2.18?

Now we stop for a while for commenting the following statements: 4 .

(n)
ψn (T ) = eiϕgeom ψn (0) (2.7)eq:geometric_phas
eq:geometric_ph
i
RT (n) (n)
(n)
ψ(T ) ' e− ~ 0
En (t) dt+iϕgeom
ψn (0) = eiϕtot ψ(0) i.e., ϕtot = arg (ψ(0), ψ(T )) (2.8)eq:total_phase
eq:total_phase
(n) (n) (n) (n)
ϕtot = ϕ(n)
geom + ϕdyn i.e. ϕ(n)
geom = ϕtot − ϕdyn (2.9)eq:Berry_discover
eq:Berry_discov

where we defined
iZT (n)
=− En (t) dt
ϕdyn {?}
~ 0
So, in the end, this is the original work of Berry presented in a slightly different language.
3
Cyclic condition on the Hamiltonian
4
They’re all approximate in the sense of the adiabatic theorem, but we put the equality sign with no confusion

7
2 Berry discovery of 1983-84

Remark 6. Eq. (2.7) defines how every phase of every wavefunction evolves in time provided
that condition (2.4) is satisfied. We recall that (2.4) is only a convention: given the general
expression for (ψk , ψ̇n ) in the particular case k = n, there’s no way for controlling the
generic phase between those two, but if (2.4) is satisfied, then a formula like (2.7) must exist.
Eq. (2.9) (and Eq. (2.8) as well) is regarded as the original discovery of Berry. Eq. (2.8) is a
consequence of the adiabatic theorem, the assumption we made in (2.4) and the cyclicity
condition of the Hamiltonian. If there are no degeneracies and no crossing levels, then every
approximate solution given by the adiabatic condition will also be cyclic.
Remark 7. One could argue that changing the convention (2.4) then the definition for (2.7)
must change. This is absolutely reasonable, but nonetheless we will see that the geometric
phase will not change under different assumptions: this invariance is one of the most
important properties of the Berry Phase.

2.2 Berry original derivation with parameter space

We want now to present the work performed by Berry in his original work in 1983 (see
Berry (1985)) in the spirit of the parameter space.
The original assumption was that the Hamiltonian, apart from being hermitian, depends
on a set of real external parameters, let’s say R, that is H ≡ H(R) with R belonging to
a multidimensional real parameter space. Then Berry imagined the following situation:
suppose to have a spin 12 particle immersed in a magnetic field, which is parametrically
dependent on time. Imagine then the real parameter space being slowly dependent in time,
that is to say, R itself to be time dependent, so that the magnetic field, and the Hamiltonian as
well, to be explicitly time dependent H(R(t)). The parameter now that varies adiabatically
is of course R(t). We will denote the closed loop in the parameter space by

C = {R(t) | 0 ≤ t ≤ T } (2.10) eq:Curve_

where we assumed R(0) = R(T ).

Figure missing! Place here as soon as possible!
The curve C is the curve traced by R(t) letting t varying. C is of course cyclic.
The curve C is then our domain of interest and for each point in C we have an associated
Hamiltonian.

H(t) |n; Ri = En (R) |n; Ri {?}

hn0 ; R | n; Ri = δn0 n {?}
X
|n; Ri hn; R| = 1 {?}
n

and we have of course an orthonormal basis at each point of the multidimensional parameters
space.
Remark 8. To be precise, we should refer to the eigenvalues En (R) not as eigenvalues of the
Hamiltonian but as eigenvalues of the Hamiltonian at a certain point R of the parameter space.
We recall moreover that each wavefunction |n; Ri is defined up to a phase factor that may
depend on n and on R.

8
 2.2 Berry original derivation with parameter space

Remark 9. Thanks to the non degeneracy of the eigenvalues of the Hamiltonian, we have
no ambiguity in defining the eigenvectors of the Hamiltonian, that is to say, there are not
crossing levels of energy. Let’s say for example that we start from the point R(0) in tha
parameter space, then we can recover a set of well defined single valued wavefunctions for
the Hamiltonian H(R(0)). Letting now t varying from 0 to T we have a complete set of well
defined single valued vawefunctions for every point and when t = T we have the same set
of eigenfunctions as we had in t = 0. Each eigenfucntion remains, anyway, defined up to an
arbitrary phase.

At this stage Berry recover the Adiabatic Theorem and tries to solve the classical Schrödinger
equation in the case in which the initial state is the n-th eigenstate of the Hamiltonian at
t = 0: 
i~ψ̇(t) = H(R(t))ψ(t)
{?}
ψ(0) = |0; R(0)i

which, for intermediate times, within the validity of the Adiabatic Theorem, gives the
solution i
Rt 0 0
ψ(t) ' e− ~ 0 En (R(t )) dt +γn (t) |n; R(t)i (2.11) eq:gen_solution
and we easily see that γn (0) = 0, since it has to match with the initial condition.
The first term in the exponential of eq.(2.11) is the dynamical phase while the term γn (t)
is the geometric phase. This latter phase is non integrable and so far we are not ready to
handle it.
So the next step to take is to take the (general) solution (2.11), to plug it into the Schrödinger
equation itself and try to extract an equation for γn (t).
Remark 10. The set |n; Ri form a set of single-valued eigenvectors in the parameter space.
Pay attention that this condition is the crucial difference in spirit with
 the derivation
 given
in the previous section. There the original assumption was to put ψn (t), ψ̇n (t) = 0 in
order to eliminate any arbitrary freedom on the choice of the initial wavefunction .
Remark 11. What Berry did is the following: he applied the Adiabatic Theorem and obtained
an approximated solution for ψ(t). He then put ψ(t) itself into the Schrödinger Equation
and derived an equation of motion for γn (t)

So, putting eq.(2.11) into the Schrödinger equation, is equivalent to write (as a consequence
of the Adiabatic Theorem)
d
γ̇n (t) |n; R(t)i ' i |n, R(t)i {?}
dt
and then sandwiching with a Bra hn, R(t)|,
d
γ̇n (t) ' i hn; R(t)| |n; R(t)i {?}
dt
= i hn; R(t)| ∇ |n; R(t)i Ṙ(t)5 {?}

Remark 12. We notice that the result

d
γ̇n (t) ' i hn; R(t)| |n : R(t)i {?}
dt
is a consequence of the Schrödinger Equation and not a condition imposed on the system.

9
 2 Berry discovery of 1983-84

Remark 13. We have |n; R(0)i = |n; R(T )i since they’re globally well defined.

Performing the circuitation along the curve C defined in (2.10), we have

I I
γn (T ) ≡ γn (0) = i hn; R(t)| ∇ |n; R(t)i Ṙ dt = i hn; R(t)| ∇ |n; R(t)i · dR(t) (2.12) eq:2.26
C C

and this is exactly what we wrote above.

In the end, we see that

hn; R(t)| ∇ |n; R(t)i = i= hn; R(t)| ∇ |n; R(t)i (2.13) eq:2.27

and so plugging (2.13) into (2.12) we obtain

I
γn (C) = −= hn; R(t)| ∇ |n; R(t)i · dR(t) {?}
x
= −= ∇ ∧ hn; R(t)| ∇ |n; R(t)i · dS 6 {?}
S
x
= −= (∇ hn; R(t)|) ∧ (∇ |n; R(t)i) · dS {?}
S
x
 
X
= −=  ∇ hn; R(t)| |m; R(t)i ∧ hm; R(t)| ∇ |n; R(t)i {?}
S m6=n

and the term m = n is neglected since gives a zero contribution: a pure imaginary term
times a pure imaginary term gives an only real contribution, so neglected by the operator
=.
Then from H(R) |n; Ri = En (R) |n; Ri, applying the gradient operator on both sides, we
obtain for m 6= n
hm; R| ∇H(R) |n; Ri
hm; R| ∇ |n; Ri = . {?}
En (R) − Em (R)
The final result is then x
γn (C) = − Vn (R) · dS {?}
S

where we defined Vn as
X hn; R| ∇H(R) |m; Ri ∧ hm; R| ∇H(R) |n; Ri
Vn (R) := = (2.14) eq:2.31
m6=n (En (R) − Em (R))2

Remark 14. Berry’s comment at this stage is that the result does not depend on R since
the gradient is only acting on H(R), so it is independent on the choice of the phase of the
wavefunction. The final take at home message is then that the geometric phase γn (C) does
not depend on the particular choice of the phase of the eigenvector |n; Ri of the Hamiltonian
at each point of the parameter space, under the assumption that the eigenvectors are globally
well defined.
hrem:freedomi
Remark 14 leads to an enormous amount of freedom in the choice of the global phase χ(R)

|n; Ri → eiχ(R) |n; Ri {?}

6
Following Berry’s argument, we are now restricting ourselves in the case in which the parameter space is a
3D space, so the simplest form of Stoke’s Theorem can be applied.

10
 2.2 Berry original derivation with parameter space

2.2.1 Two fold degeneracies

Let us suppose that we are in the proximity of a two fold degeneracy. By degeneracy we
mean that two levels of the Hamiltonian happen to be degenerate of a certain point of the
phase space.
We can suppose in R = 0 we have a degeneracy for the Hamiltonian. So if we are far from
zero, we have o problems at all, but if we are close to 0, we have to work in a different way.
The two crossing levels are the most important, so from a Quantum Mechanical point of
view, the problem is only a two level problem.

H(0) = 0; H(R) = |±; Ri = E± (R) = |±; Ri {?}

E+ (R) > E− (R), R 6= 0 {?}
E+ (0) = E− (0) = 0. {?}

So, now, without any loss of generality, we can make use of the Pauli Matrices for defining
the Hamiltonian to be in the following way
1 1
H(R) = ; E± (R) = ± R, R = |R| {?}
2 2
1 
|+; Ri h+; R| = 1 ± R̂ · σ (2.15) eq:2.33
2
And so we have to plug (2.15) into (2.14): the summation becomes an only single term and,
recognizing that En (0) − Em (0) = + 12 R − (− 21 R), we obtain

Rj
V + (R) = (2.16) eq:2.35
2R3

So now comes the result of Berry: the phase gained by the wavefunction is exactly given by
the integration of (2.16) and is the solid angle shifted by the curve C. Berry refereed to this
result as the magnetic field of a "magnetic monopole"
1
γ+ (C) = − Ω [C] , Ω [C] = solid angle at 0 {?}
2

11
3 Aharonov-Anandan and
Samuel-Bhandari generalizations

3.1 Aharonov-Anandan generalization (1987)

The first relaxation was taken in Aharonov and Anandan (1987) where the authors considered
the Adiabatic Hypothesis: they showed that the latter is not needed in order to get a geometric
effect.
Let us take a t-dependent Hamiltonian

d
i~ ψ(t) = H(t)ψ(t) {?}
dt
and assume that we found a solution for the problem that is cyclic in the sense given in the
previous chapter, that is,
ψ(T ) = eiψtot ψ(0) (3.1) eq:3.1
No assumptions on the cyclicity of the Hamiltonian is assumed in (3.1), nor no assumption
on the adiabaticity of the Hamiltonian has been made neither and we also don’t need the
fact that H(T ) = H(0), but if we have a periodicity on the solution we can associate to the
wavefunction a new concept of geometric phase1 :

ϕgeom = ϕtot − ϕdyn {?}

where
iZT
ϕdyn = − hψ(t)| H(t) |ψ(t)i dt (3.2) eq:phi_dyn
~ 0
Before exploit extensively the result achieved by Aharonov and Anandan, we can summarize
them briefly:
1. Clear definition of geometric phase: under the condition of cyclicity of the solution
ψ(t), it is possible to derive the quantity ϕtot . Then, given the prescription for the
definition of a dynamical phase, Eq. (3.2), is possible to derive the geometric phase
understood as a difference between two terms. In particular, in the case of an adiabatic
condition for the Hamiltonian, the calculation reduces to the ones already performed
by Berry.
2. The geometric phase depends only on the projection on the ray space: it is not
anymore important the wavefunction itself but it is only important the movement
of the trace of the solution projected on the space of density matrices. Being a pure
1
from now on we’ll distinguish the Berry phase from the present geometric phase since the concept we’re
dealing right now is quite different form the initial idea given by Berry: the concept of cyclicity and
adiabaticity are now not needed.

13
3 Aharonov-Anandan and Samuel-Bhandari generalizations

state, well defined, single valued unitary vector in Hilbert space for each time, ψ(t)
can be uniquely determined by the corresponding 1 dimensional projection operator,
or, as we will call it, a point in the ray space. In a certain sense, Berry already stated
this result, but the A-A were able to state this result in a more general perspective,
stating that the geometric phase lives in the ray space.
3. The parameter space concept is not needed to arrive to tha concept og geo-
metric phase: we’re not required to know the Hamiltonian acquire its dependence
on the Rvector.

3.2 First Mathematical Interlude

So far, we’ve been concerning an Hilbert space H of dimension dim H = N (where N
could be eventually infinite) and where all the possible states are defined by ψ. We define
now the unit sphere of this space, defined by all vectors that have unit norm:

B = {ψ ∈ H : hψ, ψi = 1} ⊂ H {?}

We notice immediately that B is not a linear vector space and that its dimension is N − 1.
The group U (1) of phase factors act on this set in an obvious way:

ψ ∈ B ⇒ ψ 0 = eiα ψ ∈ B, 0 ≤ α ≤ 2π {?}

We now want to introduce the concept of ray space, that we will denote by R, as the quotient
of B with the action of U (1). The idea is that two vectors that differ only by a phase factor
should be regarded as equivalent. We can also define now the equivalence class by fixing ψ
and letting α varying:
n o
R := ρ(ψ) = |ψi hψ| or ψ † ψ | ψ ∈ B . {?}

We denoted by ρ the well known projective operator, while we have to take care that the
above set is not a linear vector space. We notice that if we denote by N the dimension of the
Hilbert space H , then we have that dim(R) has 2(N − 1) real dimensions (or equivalently
N − 1 complex dimensions), or, in mathematical formalism, the space CPN −1 . We define
as well the inverse of the projection operator, the operator π, in the following way:

π: B → R {?}
ψ → ρ(ψ) = ψ † ψ ∈ R. {?}

Points in the real space are in 1 − 1 correspondence with the pure real state of the system.
The mathematical description is
n o
ρ(ψ) ∈ R → π −1 (ρ(ψ)) := ψ 0 = eiα ψ ∈ B | ψ fixed , 0 ≤ α ≤ 2π ⊂ B {?}

We now consider curves on the space B. Take a curve C

C := {ψ(s) ∈ B|s1 ≤ s ≤ s2 } ⊂ B {?}

14
 3.2 First Mathematical Interlude

with suitable smoothing conditions on it that will depend on the use we will make of this
curve. We consider its projection into the real space R through π into the set
n o
C = π [C] = ρ(ψ(s)) = ψ(s)ψ † (s) ∈ R|s1 ≤ s ≤ s2 ⊂ R. {?}

So for our purposes the most general lift will be

n o
C 0 = ψ 0 (s) = eiα(s) ψ(s)|ψ(s) ∈ C, s1 ≤ s ≤ s2 ⊂ B, {?}

that denotes all the possible lifts of the original projected curve.
Now, given a general curve C, comes natural to define the tangent vector

d
u(s) = ψ(s) = ψ̇(s) (3.3) eq:3.11
ds
(ψ(s), ψ̇(s)) = =(ψ(s), u(s)) (3.4) eq:3.12
0 iα(s)
u (s) = e (u(s) + iα̇(s)ψ(s)) {?}
u⊥ (s) = u(s) − ψ(s)(ψ(s), u(s)) = u0⊥ (s) = eiα(s) u⊥ (s) {?}
{?}

where it easy to show that Eq. (3.4) is a trivial consequence of Eq. (3.3). The general length
of a curve can be evaluated explicitly as
Z s2
L [C] = (u⊥ (s), u⊥ (s))1/2 · ds (3.5) eq:3.15
s
Z 1s2 n     o1/2
= ψ̇(s), ψ̇(s) − ψ(s), ψ̇(s) − ψ̇(s), ψ(s) ds {?}
s1

and we see that the length of the curve C does not depend on the particular lift we choose
since in the calculation tha particular α we choose disappears in the product (u⊥ (s), u⊥ (s)).
At this stage, some quick remarks are needed:
1. The practical way to evaluate the length of a curve C is to chose a particular lift,
the one that is most suitable for, plug it into formula Eq.(3.5) and evaluate directly.
Since, therefore, it is a quantity independent from gauge transformation, it is indeed a
quantity defined in the ray space R.
2. The quantity L [C] is reparametrization invariant: this is clear in view of formula (3.5)
since the integrand is homogeneous of degree 1 in velocity. This property is stated
saying that L [C] is a geometrical object.
Given then the functional (3.5), using calculus of variations we can derive an equation for
an extremum of functional length, in particular a minimum, that we will call geodesic.
i Suppose we take two point in ray space, let’s say ρ1 = ρ(ψ10 ), ρ2 = ρ(ψ20 ) and suppose
they’re not orthogonal, in the sense that Tr(ρ1 ρ2 ) > 0, i.e., |(ψ1 , ψ2 )| 6= 0, then exists
an unique geodesic C0 connecting ρ1 to ρ2 in the ray space.
ii For our purpose is then useful to define ψ1 ∈ π −1 (ρ1 ), ψ2 ∈ π −1 (ρ2 ) in such a way that
their inner product is real and positive definite,

(ψ1 , ψ2 ) = cos θ, 0 ≤ θ ≤ π(α) {?}

15
3 Aharonov-Anandan and Samuel-Bhandari generalizations

In general the inner product between two complex quantities gives a complex quantity.
If they’re real and positive definite, they’re said to be in phase one each other in the
Pancharatnam sense. We will see later on why this name is given. Pay attention that the
concept of "Pancharatnam phase" is applied only to Hilbert space vectors and not to ray
space points.
iii If we consider the geodesic C0 from ρ1 to ρ2 and then we take the lift
sin s
 
C0 = ψ0 (s) = ψ1 cos(s) + (ψ2 − ψ1 cos θ) |0≤s≤θ ⊂B {?}
sin θ
cos s
ψ̇0 (s) = u0 (s) = −ψ1 sin s + (ψ2 − ψ1 cos θ) {?}
sin θ
(u0 (s), u0 (s)) = 1, (ψ0 (s), u0 (s)) = 0, {?}
u⊥ (s) = u0 (s), π [C0 ] = C0 {?}
L [π [C0 ]] = θ {?}

in particular a parameter s is chosen in the ’affine parametrization’ spirit

In principle, no upper bounds are imposed on the value of L [C]; it could be even infinite.
Now we want to introduce another geometrical object: we don’t have time for exhaustively
talk about forms, but we want at least to give to the reader the formal definition.

A = one-form on B = −iψ † dψ {?}

On every C ⊂ B (recalling equation (3.4))

Z s2 ! Z s2
d
i ψ(s), ψ(s) ds = = (ψ(s), u(s)) ds {?}
s1 ds s1

and the last integral does not depend on how we parametrize it; it’s a geometric object. In
the language og differential geometry it is
Z
A {?}
C0

and its integral along any curve is the dynamical phase.

3.3 Samuel and Bhandari generalization (1988)

The previous mathematical section was aimed at the explanation of Samuel and Bhandari
(1988)
The main result Aharonov and Anandan (1987) achieved was that only a cyclic condition
was needed for the presence of a geometric phase. Samuel and Bhandari showed that even
the cyclic condition is not needed. Let us suppose to have a generic curve in R whose only
requirement is to be a geodesic

C0 = geodesic in R ⇒ C0 = special lift of C0 to B {?}

⇒ C = general lift of C0 to B {?}

16
 3.3 Samuel and Bhandari generalization (1988)

The result they achieved is that the phase gained along a path is equal to the integral of the
one-form defined above: in particular, the result is independent of the lifts chosen.
Z Z θ
A== (ψ(s), u(s) dS = arg(ψ(0), ψ(θ)) {?}
C0 0

So how’s they did: take the hermitian time dependent Hamiltonian H(t)

d
i~ ψ(t) = H(t)ψ(t) {?}
dt
and now consider the set of points

{0 ≤ t ≤ T : ψ(t) is a solution} ≡ C ⊂ B {?}

We are now defining a path that start from ψ(0) and ends at ψ(T ) for an arbitrary T . From
ψ(T ) we want now to come back to ψ(0) following the unique geodesic connecting those
two points. The final path will be then given by

C 0 = {Schrödinger evolution from ψ(0) to ψ(T )} ∪ {Any geodesic from ψ(T ) to ψ(0)} {?}

That is, the path C 0 is a closed loop in the Hilbert space. The main result is then that the
geometric phase of the wavefunction is given by
I
ϕgeom [Energetic Schrödinger equation] = A. {?}
C0

This is the tool in order to relax the assumption of the cyclic Hamiltonian.

17
4 The kinematic approach and the
Bargmann’s Invariant
The so called Kinematic approach has been developed by Mukunda and Simon (1993). The
key idea of the work was to look at all the possible parametrized curves in the space B.
They’re divided essentially in two groups of transformations:the first one was already seen
changing the phase of ψ(s) at each point (gauge transformation) while the second one is the
ones given by reparametrization. The leading question in this work was: what is the simplest
structure we can define on the set of curves that are invariant under gauge transformations
and that are invariant under reparametrization?
Let us start by defining what is meant by reparametrization: we take a curve C and we
substitute in place of the parameter s a function dependent on s, let’s say f (s)
( )
df (s)
C → C 0 = ψ 0 (s0 ) = ψ(s) | s0 = f (s), ≥0 {?}
ds

ϕgeom [C] = ϕtot [C] − ϕdyn [C] (4.1) eq:parametrizat

ϕtot [C] = arg(ψ(s1 ), ψ(s2 ) (4.2) eq:4.3
Z s2 ! Z s2 !
∂dψ(s) ∂dψ(s) Z
ϕdyn [C] = = ψ(s), ds = −i ψ(s), ds = A ds {?}
s1 ∂ds s1 ∂ds C

And the quantity in (4.1) needs to be reparametrization invariant and gauge invariant.
The beauty of this formulation stands in the fact that the expression holds for any open
path.
1. We can now make a subtle connection with the two derivation of Berry work in
Section 2. We said that we have an infinite freedom in choosing the initial phase of
the wavefunction. This freedom can be connected at the freedom we have in choosing
the lift we prefer for the evaluation of formula (4.2).
2. It does not take any effort to define the geometric phase for any open loop.
3. This approach is purely kinematic: there’s no Hamiltonian, no Schrödinger equation.
We close this section with the statement

ϕgeom [any geodesic in R] = 0. {?}

19
4 The kinematic approach and the Bargmann’s Invariant

4.1 Bargmann’s Invariant

This terminology was introduced by Bargmann (1964) working on a proof of the Wigner
Theorem. He gave the simplest non trivial Bargmann invariant. Take 3 generic unit vectors
and we ensure none of them are mutually orthogonal:

ψj , j = 1, 2, 3, ∈ B {?}

So the definition of the third order Bargmann invariant is the following:

∆3 (ψ1 , ψ2 , ψ3 ) = (ψ1 , ψ2 ) (ψ2 , ψ3 ) (ψ3 , ψ1 ) {?}

= Tr(ρ1 ρ2 ρ3 ) {?}

Where by ρi , i = 1, 2, 3 we denoted the density matrices and we recall that their product is
a ray space quantity.
It turns out that if dim H ≥ 2 then the above quantity is in general complex, the phase is
non trivial.
So in general the phase of the Bargmann quantity can be shown to be a geometric phase:
first of all it can be splitted in a sum of different phases and since every path is a geodesic,
then the argument of the inner product (ψi , ψj ) is equal to the dynamical phase acquired
along the path connecting the two points:

arg(∆3 (ψ1 , ψ2 , ψ3 )) = arg(ψ1 , ψ2 ) + arg(ψ2 , ψ3 ) + arg(ψ3 , ψ1 ) {?}

= ϕdyn [C12 ] + ϕdyn [C23 ] + ϕdyn [C31 ] {?}
= ϕdyn [C12 ∪ C23 ∪ C31 ] {?}
= −ϕgeom [C12 ∪ C23 ∪ C31 ] {?}

Notice that the definition of Bargmann invariant can be easily generalizable to any arbitrary
dimension.

4.2 Examples
! !
Ex E1
E= = ∈ R2 {?}
Ey E2
and the EM field can be described by the intensity I = E † E and by the polarization
n̂ = N1 E † τ E ∈ SP2 oincaré

4.3 Second Mathematical Interlude

We can imagine the tangent vector as the tangent of a smooth curve passing for the point.
We can therefore condifer a curve superimposed to the fiber. The tangent space generated
will be called vertical space

Vψ B = {iaψ | a ∈ R} ⊂ Tψ B {?}

20
 Bibliography
hase Yakir Aharonov and J Anandan. Phase change during a cyclic quantum evolution. Physical
Review Letters, 58(16):1593, 1987.
note Valentine Bargmann. Note on wigner’s theorem on symmetry operations. Journal of
Mathematical Physics, 5(7):862–868, 1964.
ical MV Berry. Classical adiabatic angles and quantal adiabatic phase. Journal of physics A:
Mathematical and general, 18(1):15, 1985.
928m M Born. M. born and v. fock, z. phys. 51, 165 (1928). Z. Phys., 51:165, 1928.
ntum N Mukunda and R Simon. Quantum kinematic approach to the geometric phase. i. general
formalism. Annals of Physics, 228(2):205–268, 1993.
ized Shivaramakrishnan Pancharatnam. Generalized theory of interference and its applications.
44(6):398–417, 1956.
eral Joseph Samuel and Rajendra Bhandari. General setting for berry’s phase. Physical Review
Letters, 60(23):2339, 1988.

21