You are on page 1of 24

QUANTUM MECHANICS

1 Vector Spaces and Operators


In classical mechanics the ”state” of a system at a given time t = t0 is spec-
ified by a point in the phase pace (qi , pi ), i = 1, . . . , n. Physical observables
are functions of phase space (and possibly time) F (pi , qi , t). Once a state is
given at t = t0 then the time evolution at an arbitrary time t is determined
by the hamiltonian H = H(pi , qi ) via the equations of motion:
∂ ∂
q˙i = {qi , H} = H ṗi = {pi , H} = − H, (1.1)
∂pi ∂qi
where
P {.} is the Poisson bracket and we take the standard symplectic form
dqi dpi . More generally
d ∂
F = {F, H} + F. (1.2)
dt ∂t
In quantum mechanics instead a state is associated to a vector |vi (ket)
in a complex vector space V , which we assume for simplicity to be of finite
dimension d. This means that there are at most d linearly independent
P vec-
tors (|v1 i, . . . , |vn i) and any |vi ∈ V can be written as |vi = i ci |vi i with
unique complex numbers ci , |vi i form a basis for V .

Dual Vector Space


Quite generally, given a vector space V there is a natural dual vector space V̂ ,
given by the linear functionals f on V , whose elements we denote by hf | (bra),
that by definition associate to any |vi a complex number f (|vi) ≡ hf |vi, with
property hu|(a|vi + b|wi) = ahu|vi + bhu|wi. By definition, linearity holds
also among the bras. Given a basis |vi i of V a natural dual basis, hv i | of V̂
defined by hv i |vj i = δji .

Scalar Product
We will be considering vector spaces endowed with a scalar product, i.e. a
map V × V → C, denoted temporarily by (|ui, |vi) with the following prop-
erties:
i)linearity in the second argument
ii) Anti-linearity in the first argument
iii) Hermiticity: (|ui, |vi) = ((|vi, |ui) (the bar stands for the complex con-
jugate).
iv) Positive semidefinitness: (|vi, |vi) ≥ 0 and (|vi, |vi) = 0 iff |vi = 0.

2
The last property implies non-degeneracy of the scalar product: one way to
characterise it is through the metric tensor gij = (|vi i, |vj i) defined for an
arbitrary basis. It amounts to say that gij is a non-degenerate (invertible)
matrix. In this case we can identify V̂ with V itself: take in V an orthonor-
mal basis (this is always possible by the Gram-Schmidt procedure)1 i.e. such
that in the above gij = δij . An element hf | of V̂ is characterised by its
valuesPon the basis elements |vi i of V , hf |vi i = fi . Consider then the ket
|f i = i f¯i |vi i. Clearly (|f i, |vi i = fi , so we identify

hf |vi i ≡ (|f i, |vi i). (1.3)

In practice one uses representations of abstract kets and bras: once the
standard orthonormal basis |ei i is chosen, then |vi is identified with the
column vector given by its components vi along the |ei is, vi =Phei |vi, and
hv| is P
the complex conjugate transposed row vector. If |vi = vi |ei i and
|ui = ui |ei i then X
hu|vi = ūi vi (1.4)

On a vector space with a positive


p semidefinite scalar product on defines a
norm of vectors: |||vi|| ≡ hv|vi ≥ 0. Positivity of the scalar product
implies the Cauchy-Schwarz inequality: for any |ui, |vi ,

|||ui|||||vi|| ≥ |hu|vi|. (1.5)


hu|vi
The C-S inequality follows from considering |||ui − hv|vi
|vi||2 ≥ 0.
The C-S inequality implies the triangle inequality :

|||ui|| + |||vi|| ≥ |||ui + |vi||. (1.6)

Spaces with a norm as above are also termed metric spaces: the norm gives a
notion of distance between two elements of the space, d(|vi, |ui) = |||ui−|vi||.

Operators and Observables


In Quantum Mechanics, on the other hand, physical observables are asso-
ciated to linear operators A from V to V , i.e. A|vi ≡ |Avi ∈ V for any
1
Given an arbitrary basis |ui i, i = 1, . . . , d , suppose that one has found k, k < d
orthogonal vectors |ũi i, i = 1, . . . , k. then a k + 1-th orthogonal vector |ũk+1 i is given by
|uk+1 i − i=1 hũhũi |u k+1 i
Pk p
i |ũi i |ũi i. Finally one can normalise the |ũi i by |ũi i/ hũi |ũi i

3
|vi ∈ V . and the action obeys the obvious linearity property. One can take
linear combinations and products of operators, so they form an algebra. We
can define for any A its matrix elements hu|A|vi. For every operator A the
adjoint or hermitian conjugate A† is defined by the relation

hu|Avi ≡ hA† u|vi = hv|A† ui (1.7)

The following relations can be proved, (AB)† = B † A† , (aA+bB)† = āA† +b̄B †


for any a, b ∈ C. An operator A is self-adjoint (hermitian) if A† = A.

As for vectors, one uses an explicit representation for operators: given


an orthonormal basis |ei i, hei |ej i = δij , the action of A on it is given
A|ei i = Aji |ej i for a d × d matrix Aji = hej |A|ei i. The product of oper-
ators becomes matrix multiplication. Hermitian conjugation of A becomes
hermitian
P conjugation of the corresponding matrix A†ij = Āji . P Writing
ci |ei i and |v 0 i = A|vi on sees that |v 0 i = c0i |ei i and c0i = j Aij cj
P
|vi =
is the standard action on column vectors.

For any operator A one has the notion of eigenvector and eigenvalue:

A|λi = λ|λi (1.8)

for eigenvalue λ ∈ C and eigenvector |λi.


For hermitian operators one can easily prove the following very important
facts:
i) λ is a real number.
ii) Eigenvectors corresponding to distinct eigenvalues λ1 6= λ2 are orthogonal
hλ1 |λ2 i = 0.
These facts follow from the identity hλ1 |A|λ2 i = hλ2 |A|λ1 i valid for an her-
mitian A.
Moreover, hermitian operators can always be diagonalised: one can find a
basis given by the eigenvectors , in which they become diagonal matrices
with eigenvalues λi on the diagonal, appearing with multiplicities gi ≥ 1. If
gi i1 then λi does not specify uniquely (up to a scalar) the eigenvector and
one has to introduce an additional label |λm
i i, m = 1, . . . , gi . Orthonormality
m n
implies hλi |λj i = δij δmn .
This can be rephrased by the statement for any hermitian operator A there
exists a similarity transformation (with U actually unitary, U † = U −1 ) such
that U −1 AU = D where D is the diagonal matrix above. The unitary matrix

4
U is given explicitly: its columns are the orthonorrnal eigenvectors of A.
In the finite dimensional case, after choosing a matrix representation for A,
the eigenvalues and their multiplicities are found from the secular equation:

det(A − λI) = 0, (1.9)

with I the d × d identity matrix, which is a polynomial equation of degree d.


Notice that this equation is invariant under a similarity transformation of A,
A → U −1 AU . This makes sure that the eigenvalues are independent of the
basis chosen to represent A.

Simultaneous Diagonalisation
What can be said about a set of hermitian operators about their simultaneous
diagonalisation ? Take two hermitian operators A, B, then it is to see that
they can be put simultaneously to diagonal form if and only if they commute
[A, B] ≡ AB − BA = 0. The necessity is obvious, since U −1 AU = DA and
U −1 BU = DB with diagonal DA,B and U −1 [A, B]U = [DA , DB ] = 0 . In the
other direction, suppose U −1 AU = DA with DA having distinct eigenvalues
first. From
[DA , X]ij = (λA A
i − λj )Xij (1.10)
(no sum on i, j) it follows that Xij = 0 for i 6= j. Since X = U −1 BU it fol-
lows that U diagonalises also B. This can be easily generalised to the case of
non trivial multiplicities and to a larger set of mutually commuting operators.

Therefore we have simultaneous orthonormal eigenstates of A, B,. . . ,

A|λi , µi , . . . i = λi |λi , µi , . . . i, B|λi , µi , . . . i = µi |λi , µi , . . . i, (1.11)

In the finite dimensional case the orthonormal eigenstates of an hermitian


operators constitute always basis for V , we define in general an observable
to be an hermitian operator whose orthonormal system of eigenstates form
a basis for the (possibly infinite dimensional) vector space V .
For a maximal set of commuting observables specifying their eigenvalues
specify uniquely (up to a phase) the corresponding normalized eigenstate.
In the angular momentum theory, for example, a state is specified by the
~ 2 , ~2 (j + 1) and that of L3 , ~m, and the corresponding state
eigenvalue of L
is denoted by |j, mi, j ∈ N and m ranging from −j to j in steps of 1. So

5
~ 2 is 2j + 1-degenerate.
each eigenvalue L

Completeness Relation
We have seen that for any |ui,|vi, hu|vi is a complex number. On the other
hand |uihv| is in fact an (linear) operator , acting on any |wi ∈ V as |uihv|wi
(this is also seen using column vector representation for a ket and row vector

representation for a bra). Clearly (|uihv|) P = |vihu|. Consider now an or-
thonormal basis |eP i i. For any |vi, |vi = ci |ei i, where ci = hei |vi, therefore
for any |vi, |vi = |ei ihei |vi, which implies
X
|ei ihei | = I (1.12)

where I is the identity operator on V . This relation is called resolution of


the identity or completeness relation.
An important class of hermitian operators are the projectors Π defined by
the condition: Π2 = Π. For an Porthonormal basis |ei i, i = 1, . . . , d one can
verify that the partial sums m i=k |ei ihei | are indeed projectors , with the
obvious geometrical interpretation.

Change of Basis
Clearly the explicit representation of vectors and operators discussed before
depends on the chosen (orthonormal) basis. Indeed let two orthonormal basis
|ei i and |fi i be related
|fi i = U |ei i. (1.13)
|fk ihk | and it easily verified that U † U = I i.e. U is unitary.
P
Clearly U =
Its matrix elements in the |ei i basis are

Uij = hei |fj i. (1.14)

The components of |vi in the two basis are ci = hei |vi and c0i = hfi |vi
Presp. To
find the relation between c0 and c one inserts the identity operator |ek .ihek |
in the last relation to find
c0i = Uij† cj (1.15)
By similar manipulations on can work out the transformation of operator
matrix elements under change of basis

A0ik = Uij† Ajl Ulk (1.16)

6
Infinite Dimensions
We have been assuming that the vector space V is finite dimensional, but
this is hardly the case for interesting physical systems where typically V
is actually an infinite dimensional Hilbert space 2 . For example, in the
case of the harmonic oscillator the relevant space is that of square integrable
functions on the real line. In case where the spectrum of the relevant opera-
tor (typically the Hamiltonian H ) is discrete, then formally all the above
expressions remain valid when d → ∞. Of course in writing for example
|vi = ∞
P
i=0 i |ei i, where |ei i is an orthonormal basis,
c P∞ one 2has to make sure
that this infinite series makes sense, i.e. hv|vi = i=0 |ci | < ∞. Actually,
there can be cases (i.e. the Coulomb potential), where the spectrum of H has
a continuous part, the scattering states, and a discrete part, the bound
states. The continuous states are labelled by say a real number ξ and the
orthonormality condition has to be extended by the Dirac’s δ-function

hξ|ξ 0 i = δ(ξ − ξ 0 ), (1.17)

so |ξi strictly speaking has not a finite norm. Nevertheless, one can use
the basis |ξi to constructR honest normalizable states, the analog of wave
packets. Indeed if |vi = dξc(ξ)|ξi, then, from the above equation,
Z
hv|vi = dξ|c(ξ)|2 , (1.18)

so it is sufficient to choose a square integrable profile function c(ξ) to get a


normalizable |vi.
The resolution of identity in the continuous plus discrete case takes the form
Z X
dξ|ξihξ| + |ei ihei | = I. (1.19)

2
A Hilbert space is a metric space (not necessarily finite dimensional) which is complete
in the sense that any Cauchy sequence converges to a point of the space.

7
2 The Axioms of Quantum Mechanics
1) At time t = t0 the state of a physical system is specified by a vector |ψi
in the space of states V .
Since we are dealing with linear spaces any linear combination of states is
again a state (superposition principle).
2) Every measurable physical quantity is represented by an observable A act-
ing on V
3) The possible results in a measurement of a physical quantity are the eigen-
values of the corresponding observable A.
Suppose we are in a state |ψi and hψ|ψi = 1, then consider an observable
A: its orthonormal eigenstates (assume no degeneracy)
P |ai i with eigenval-
P ai2 form a basis and we can expand |ψi =
ues ci |ai i with ci = hai |ψi and
|ci | = 1. Then the fourth axiom says
4) For a system in a state |ψi, the probabiility P (ai ) of obtaining ai in a
measurement of the physical quantity corresponding to A is
P (ai ) = |hai |ψi|2 (2.1)
The probability is understood as defined from the frequency with which ai
appears when we make a very large number of measurements of A, with the
system is always in the same state |ψi.
In the case of non trivial degeneracies gi of ai , with eigenstates |ani i, then
gi
X
P (ai ) = |hani |ψi|2 (2.2)
n=1

We understand the meaning of requiring the state to be normalized to unit


norm from the probabilistic Pinterpretation above: probabilities P (ai ) must
be non-negative and obey P (ai ) = 1, since the probability of finding any
eigenvalue must be 1.
Clearly if the system is in an eigenstate of A corresponding to some eigen-
value ai then P (ai ) = 1 and P (aj ) = 0 for j 6= i.

Expectation Values
The above probabilistic interpretation brings in naturally the notion of av-
erage value or expectation value hAi of an observable: it is the sum of
eigenvalues weighted with the respective probabilities:
X
hAi = P (ai )ai (2.3)

8
This expression can be rewritten as

hAi = hψ|A|ψi (2.4)

as can be checked by expanding |ψi in eigenstates of A (we assume as usual


|ψi normalized).
A quantity that measures the spread of the distribution of eigenvalues around
hAi is given by the variance ∆A . Clearly hA − hAii = 0, on the other hand
p p
∆A = h(A − hAi)2 i = hA2 i − hAi2 (2.5)

does precisely the job we want (recall the gaussian distribution), it measures
the width of the distribution of eigenvalues of A around the mean value, i.e.
the ”size” of quantum fluctuations in the state.|ψi (actually better take di-
mensionless ∆A/hAi).

Heisenberg Uncertainty Relation


Given two observables A and B, we can define ∆A, ∆B. Heisenberg’s un-
certainty relation gives a lower bound on the product ∆A∆B in terms of the
commutator [A, B]:
∆A∆B ≥ |h[A, B]i|/2 (2.6)
The proof of this uses C-S inequality applied to the states |αi = (A − hAi)|ψi
and |βi = (B − hBi)|ψi. We have from C-S that

hα|αihβ|βi ≥ |hα|βi|2 (2.7)

Now the expectation value of hermitian operators is real, whereas that of


anti-hermitian ones (A† = −A) is purely imaginary. On the rhs of the above
equation (A − hAi)(B − hBi) can be rewritten as

(A − hAi)(B − hBi) = 1/2([A − hAi, B − hBi] + {A − hAi, B − hBi}) (2.8)

where {.} denotes here the anticommutator. The commutator is anti-hermitian


whereas the anticommutator is hermitian, so the lhs is the modulus squared
of a real plus an imaginary term, which therefore is ≥ the modulus squared
of the imaginary part i.e. 1/4|h[A − hAi, B − hBi]i|2 = 1/4|h[A, B]i|2 . The
lhs is (∆A∆B)2 , so taking the square root gives the result. Anticipating
what will come later, In the context of relativistic QFT, this will require op-
erators which are space-like separated to commute (or anti-commute in case

9
of Fermi statistics), because they are not supposed t influence each other
(microcausality).

Rays
From the axioms of Quantum Mechanics we can conclude with a more pre-
cise definition of state: we have seen that an overall phase in a vector in
the Hilbert space of states is physically irrelevant, therefore we can identify
vectors differing by a phase and introduce the notion of ray for the corre-
sponding equivalence class. A unit ray is a ray with unit norm. This notion
is useful in Wigner’s analysis of symmetries in Quantum Mechanics, culmi-
nating in Wigner’s classification. Symmetries are required to be realized as
(linear or anti-linear) operators acting on the Hilbert space by preserving
physical quantities like transition probabilities. It states that symmetries are
realized either as linear, unitary or anti-linear, anti-unitary operators3 acting
on the Hilbert space of states.

Position Eigenstates Basis


In most of quantum mechanical problems one uses a basis given by eigenstates
of the position operator, say X:
X|xi = x|xi (2.9)

Clearly, hx|x0 i = δ(x − x0 ) and −∞ dx|xihx| = I, so that for any state
R
Z ∞
|ψi = dx|xihx|ψi (2.10)
−∞

and if hψ|ψi = 1 then


Z ∞
dx|ψ(x)|2 = 1 (2.11)
−∞
R
where ψ(x) = hx|ψi is the wave function. Then D dx|ψ(x)|2 is the prob-
abilitiy of finding the particle in the region D of the real line. This can be
extended in an obvious way to the 3-dimensional space X|~ ~ xi = ~x|~xi ,and
ψ(~x) = h~x|ψi with Z
d3 x|ψ(~x)|2 = 1. (2.12)
R3
3
Anti-linearity involves complex conjugation of the coefficients of linear combinations
and U being anti-unitarity means that hU u|U vi = hv|ui as opposed to hU u|U vi = hu|vi
for U unitary.

10
Of course one can represent operators in this basis and the matrix elements
then become: Z
hψ|A|φi = dxdx0 ψ̄(x)hx|A|x0 iφ(x0 ) (2.13)

The Momentum Operator


We introduce the momentum operator as the generator of space translations,
postulating an operator (we consider the one-dimensional case, but general-
isation to 3D is obvious)

Ta = exp(−iaP/~) (2.14)

where ~ ∼ 10−34 J sec in MKS inits, is the reduced Planck constant. P is


hermitian, so T is unitary and being a a length, we need a constant hav-
ing dimensions of an action, ~. Moreover Ta Tb = Ta+b , T−a = Ta−1 , so T
are elements of an abelian group and P is the infinitesimal generator. By
definition
Ta |xi = |x + ai (2.15)
Now, working infinitesimally in a, T = 1 − iaP/~ we have:

XT |xi = (x + a)|x + ai, T X|xi = x|x + ai (2.16)

implying
[X, P ] = i~ (2.17)
which generalises to
[Xi , Pj ] = i~δij (2.18)
for i, j = 1, 2, 3, to 3D. Also [Xi , Xj ] = [Pi , Pj ] = 0.
We then have the familiar form of Heisenberg uncertainty relation

∆X∆P ≥ ~/2 (2.19)

Let us now derive the form of P in position space:


Z Z
(1 − iP δx/~))|ψi = dx|x + δxihx|ψi = dx|xihx − δx|ψi (2.20)

Expanding to first order in δx we obtain


Z

P |ψi = dx|xi(−i~ ψ(x)) (2.21)
∂x

11
Projecting on hx0 |, we obtain hx0 |P |ψi = −i~ ∂x∂ 0 ψ(x0 ). in particular


hx0 |P |xi = −i~ δ(x0 − x) (2.22)
∂x0
since hx0 |xi = δ(x0 − x). and more generally
Z

hφ|P |ψi = dxφ̄(x)(−i~ ψ(x)) (2.23)
∂x
which says that P is hermitian on the space of functions vanishing at ∞.
One could have chosen a basis of eigenstates of P instead , P |pi = p|pi. Since


hx|P |pi = phx|pi = −i~ hx|pi (2.24)
∂x
we can solve for < x|pi:
1
hx|pi = √ eipx/~ (2.25)
2π~
where the normalization is fixed by consistency with the δ-function represen-
tation Z
1
δ(x) = dkeikx (2.26)

From hx|pi we get that the change of basis from x to p representations and
viceversa is just the Fourier transform:
Z Z
1
ψ(x) = dphx|pihp|ψi = √ eipx/~ ψ(p)dp (2.27)
2π~
and Z Z
1
ψ(p) = dxhp|xihx|ψi = √ e−ipx/~ ψ(x)dx (2.28)
2π~
Thus we see that if we have a classical Hamiltonian of the form H = p~2 /2m+
V (~x), in the quantum mechanical setting it becomes the differential operator:

~ 2 /2m + V (~x)
H = −~2 ∇ (2.29)

12
Time Evolution
So far we have been implicitly assumed all states to be at some given time
t = t0 , but it is clearly of fundamental importance to determine how states
evolve in time. From classical mechanics we know that in a suitable sense
the hamiltonian of a system generates time evolution through Hamilton equa-
tions. These are equations of motion for physical observables, we are after
equations governing time evolution of states in Quantum Mechanics. We can
reason like in the case of the momentum operator and postulate that there
exist a unitary operator U (t, t0 ) such that:

|ψ(t)i = U (t, t0 )|ψ(t0 )i (2.30)

We require:
i) U (t, t) = I
ii) U (t2 , t1 ) = U −1 (t1 , t2 )
iii) U (t3 , t2 )U (t2 , t1 ) = U (t3 , t1 )
iv) Finally we require the hamiltonian, which can be time dependent, to be
infinitesimal generator of time translations, in the sense that :

U (t + dt, t) = I − idtH(t)/~ + o(dt2 ) (2.31)

Using the above properties we can derive a differential equation for U (t, t0 ):

U (t + dt, t0 ) − U (t, t0 ) U (t + dt, t) U (t, t0 ) − U (t, t0 )


= = −iH(t)U (t, t0 )/~
dt dt
(2.32)
Therefore:
d
i~ U (t, t0 ) = H(t)U (t, t0 ) (2.33)
dt
and
d
i~ |ψ(t)i = H(t)|ψ(t)i (2.34)
dt
which is the Schrödinger equation for states.
The simplest case is when H is time independent, which gives U (t, t0 ) =
e−iH(t−t0 )/~ . This will be the case we will be interested. For eigenstates of
H with eigenvalue E the time evolution is simply |ψ(t)i = e−iE(t−t0 )/~ |ψ(t0 )i
(stationary state), and more generally
X
|ψ(t)i = e−iEn t/~ cn |ni (2.35)
n

13
where |ni are eigenstates of H and t0 = 0.
In this framework, states are time dependent whereas operators are kept
independent of time (Schrödinger picture) and expectation values are in
general time-dependent

hAi(t) = hψ(t)|A|ψ(t)i (2.36)

except for stationary states, where hAi is time-independent. There is an


equivalent picture, Heisenberg picture where states are kept fixed and
operators vary with time, in such a way that expectation values are the
same. If we denote by AS (t) the operator in Schrödinger picture, including
a possible explicit time dependence, then in Heisenberg picture AH (t) =
U † (t, t0 )AS (t)U (t, t0 ) and we can derive the following differential equation
for AH (t) using that for U :

d H 1 ∂
A (t) = [AH , H H ] + ( AS )H (2.37)
dt i~ ∂t
with the obvious meaning of the superscript H. Notice the similarity with
the equation of motion for the classical observables at the beginning of Sec-
tion 1: the Poisson bracket {, } there is replaced by the commutator [, ]/i~
in the quantum case.

Homework

1) The operator
 √ 
0
√ 2 √0
L = a  − 2 0√ 2 
0 − 2 0

with a = −i~/2 is hermitian. Find its eigenvalues and eigenvectors.

2) Consider the two operators


 
1 0 0
A = a  0 −1 0 
0 0 −1

14
 
1 0 0
B = b 0 0 1 
0 −1 0
in the standard orthonormal basis (ei )j = δij . Find the basis in which
both are diagonal. Verify the completeness relation of the new orthonormal
basis.

3) Let |ei i the standard orthonormal basis in 3 dimensional space. Con-


sider the states |ψi = √12 |e1 i+ 2i |e2 i+ 21 |e3 i and |φi = √13 |e1 i+ √i3 |e3 i compute
the projector operators on |ψi and |φi.

4) If H = p2 /2m + V (x) has eigenstates |ni with eigenvalues En compute


a in
hn|P |n0 i = ahn|X|n0 i (2.38)
Hint: consider [H, X].

5) With |ni as above verify that any operator A, with matrix elements
Anm = hn|A|mi, can be written as
X
A= Anm |nihm| (2.39)
n,m

6) With the usual Pauli matrices ~σ :


 
0 1
σ1 =
1 0
 
0 −i
σ2 =
i 0
 
1 0
σ3 =
0 −1
evaluate
ei~n.~σ (2.40)
where ~n is an arbitrary vector.

15
~ · J~ for some
7) Compute dtd (ψ̄ψ) and verify that it can be written as ∇
current J~ if ψ satisfies Schrödinger equation.

8) Wave packets for the free particle: consider the normalized wave
function (here ~ = 1)

ψ(x) = N exp(ikx − x2 /2d2 ) (2.41)

Determine |N |. Compute ∆x and ∆p and verify that Heisenberg uncertainty


relation is saturated. Verify that in momentum space ψ(p) = N 0 exp(−(p −
k)2 d2 /2). What would be ψ(p, t) at time t ?

3 The Two-Level System


The simplest quantum mechanical system involves a two-dimensional vector
space of states and it is relevant, for example, for describing spin 1/2 particles
in an external magnetic field if one neglects the orbital dynamics. A general
hamiltonian for a two-level system can be written as
~ · ~σ = E0 1 + E n̂ · ~σ
H = E0 1 + E (3.1)

~ and ~σ are the


n̂ = (sinθcosφ, sinθsinφ, cosθ) is a unit vector n̂2 = 1, E = |E|,
Pauli matrices . This is because any hermitian 2 × 2 matrix can be expanded
as above with real parameters E’s. For the case of a spin 1/2 particle in an
external magnetic field parallel to n̂ only the second term would be present.

The first question is how to diagonalize H. Since it is a 2 × 2 matrix this


can be done by brute force, of course. However we use the fact that n̂ · ~σ can
always be obtained by conjugating σ3 with an SU (2) element g 4 : in other
words there is an SU (2) group element g such that

n̂ · ~σ = g −1 σ3 g (3.2)

and one can verify that g = eiθσ2 /2 eiφσ3 /2 does the job. This corresponds
to first doing a (counterclockwise) rotation by θ around the 2nd axis and
then by φ around the 3rd one. Now if |±i are the standard eigenstates of
Recall that for g ∈ SU (2), g −1 σi g =
4
P
j R(g)ij σj where R(g) is an SO(3) group
element.

16
σ3 with eigenvalues ±15 then |±in̂ = g −1 |±i are eigenstates of n̂ · ~σ with
eigenvalues ±1. The states |±in̂ diagonalize H also. The eigenvalues of H
will be E± = E0 ± E and we will denote its orthonormal eigenstates |±in̂ by
|E± i.
Computing g explicitly one obtains
 −iφ/2
cos(θ/2) −e−iφ/2 sin(θ/2)

−1 e
g =
eiφ/2 sin(θ/2) eiφ/2 cos(θ/2)

from which
e−iφ/2 cos(θ/2)
 
|+in̂ =
eiφ/2 sin(θ/2)
and
−e−iφ/2 sin(θ/2)
 
|−in̂ =
eiφ/2 cos(θ/2)
For the general two level case suppose that at t = 0 the system is in the
state |+i . This is not an eigenstate of H so it will have a non trivial time
evolution obtained by applying e−iHt/~ to it. We can expand |+i = |ψ(0)i in
eigenstates of H, |E± i, inverting the previous relations and, up to an overall
phase, we get:
|+i = cos(θ/2)|E+ i − sin(θ/2)|E− i (3.3)
Then applying the time evolution operator:

|ψ(t)i = cos(θ/2)e−iE+ t/~ |E+ i − sin(θ/2)e−iE− t/~ ||E− i (3.4)

We can then ask what is the probability amplitude that the system be at
time t in the state |−i, and this is given by the matrix element:

h−|ψ(t)i = sin(θ)sin[(E+ − E− )t/2~] (3.5)

up to an overall phase. Therefore, the transition probability is P+→− (t) =


−E−
|h−|ψ(t)i|2 = sin2 (θ)sin2 ( E+2~ t) which has an oscillatory behavior with
E+ −E−
frequency ω = 2~ . Notice also that for θ = 0 this is zero because then
|+i = |ψ(0)i is an eigenstate of H. This is the quantum manifestation of
the classical precession (Larmor ) of the magnetic moment in an external
magnetic field.

5
A general normalized state α|+i + β|−i is also called a qubit state.

17
Homework

1) The above calculation can be extended to the case where the magnetic
field has a component which rotates say in the plane 1−2, with corresponding
hamiltonian

H = ~(ω0 σ3 + ω1 (cos(ωt)σ1 + sin(ωt)σ2 ))/2 (3.6)

Setting |ψ(t)i = a+ (t)|+i + a− (t)|−i one writes the equations for ȧ± (t)
but then observe that defining b± (t) = e±iωt/2 a± (t) one gets time inde-
pendent equations for ḃ± (t), with time independent hamiltonian H̃. Then
|ψ(t)i = e−iωtSz /~ |ψ̃(t)i where |ψ̃(t)i = e−itH̃/~ |ψ̃(0)i. Therefore one can com-
pute the transition amplitude h−|ψ(t)i like before.

2) Show that the hamiltonian H of the previous exercise can be written


as H = g −1 (t)Ĥg(t) where Ĥ is time independent (related to the previous
Ĥ), and g(t) is a time dependent SU (2) group element. Find g(t) and Ĥ.

4 The Quantum Harmonic Oscillator


The Hamiltonian of the one-dimensional harmonic oscillator is given by
1 2 1
H= p + mω 2 x2 (4.1)
2m 2
corresponding to a Lagrangian
1 1
L = mẋ2 − mω 2 x2 (4.2)
2 2
m being the mass and ω the frequency. We have p = δL/δ ẋ = mẋ. The
allowed values of the energy vary continuously from 0 to ∞.

This is also the quantum hamiltonian, but x and p are operators obeying
[x, p] = i~.
To discuss the quantum version it is convenient, using Planck’s constant ~,
to introduce dimensionless variables
r
mω 1
x̂ = x, p̂ = √ p (4.3)
~ mω~

18
obeying [x̂, p̂] = i. We then form a pair of hermitian conjugate operators
a and a† , where a = x̂+ip̂
√ , which obey [a, a† ] = 1. In terms of these the
2
quantum Hamiltonian becomes
1
H = ~ω(a† a + ) (4.4)
2
One has [N, a] = −a and [N, a† ] = a† .

We want to analyze the Hilbert space on which these operator act. No-
tice that H is a positive definite operator, since N = a† a is itself positive
semidefinite.

Now consider an eigenstate of N , N |µi = µ|µi. Then one proves that


µ ≥ 0, and if µ = 0 then a|0i = 0. This is due to the positivity of Hilbert
space inner product. Moreover from the commutator above it follows that
N (a|µi) = (µ − 1)a|µi and N (a† |µi) = (µ + 1)a† |µi and for this reason a and
a† are called annihilation (lowering) and creation (raising) operators, respec-
6 0 and a† |µi =
tively. One can also prove that if µ > 0 then a|µi = 6 0 for µ ≥ 0.

Putting together these informations one can see that µ must be a non-
negative integer n ∈ N, for otherwise by acting on |µi with a repeatedly
one would get states of negative eigenvalues for N . The vector |0i is the
lowest energy (ground) state of the system and all energy eigenstates |ni are
obtained by applying (a† )n to the ground state.
Assuming h0|0i = 1, one gets normalized eigenstates hn|mi = δnm by
a†
n √ †

|ni = √ n!
|0i and then a|ni = n|n − 1i, a |ni = n + 1|n + 1i.

The energy levels the are En = ~ω(n + 21 ), n ∈ N . Therefore the ground


state has energy ~ω/2, and, together with the discreteness of the spectrum,
this is to be contrasted with the classical case.

One can easily obtain from this algebraic approach the wave functions
in the x space ψn (x) = hx|ni . From a|0i = 0 one gets hx̂|(x̂ + ip̂)|0i = 0
2
d
and since hx̂|p̂|0i = −i dx̂ ψ0 (x̂) from which it follows that ψ0 (x̂) = C0 e−x̂ /2
where C0 is a normalization constant. The wave functions corresponding to
the excited states are obtained the raising operators and in this way one re-
produces the well known eigenfunctions of H involving Hermite polynomials

19
Hn times the gaussian factor.
r
mω 2
ψn (x) = Cn Hn ( x)e−mωx /2~ (4.5)
~
for suitable normalisation factor Cn .

5 Coherent States
In the previous section we have diagonalised the quantum Hamiltonian of the
harmonic oscillator in terms of the energy eigenstates |ni, n = 0, 1, 2, . . . . We
want now to study a different class of states |αi, α ∈ C being an arbitrary
complex number, called coherent states. They are expressed in terms of
|ni’s since anyway these are a complete set, but they, contrary to |ni, are
very close to describe the classical harmonic oscillator.

Remember that we can describe the classical motion by using the vari-
ables a, a† , but to emphasize their classical nature we will use α, ᾱ, where
now ᾱ is the complex conjugate of α.

The classical Hamiltonian is then H = ~ωαᾱ (notice there is no 1/2


shift). The solutions of the equations of motion will be α(t) = α0 e−iωt and
the value of the energy E = ~ω|α0 |2 .

We are after states |ψi which are close to this classical situation, that is
states for which, at t = 0 and |α0 | large,

hψ(0)|a|ψ(0)i = α0 hψ(0)|H|ψ(0)i = ~ω|α0 |2 (5.1)


hx̂i+ihp̂i
Clearly α = √ and therefore the expectation values of x and p are re-
2 q p
covered from α: hxi = 2mω ~
(α + ᾱ) and : hpi = ~mω/2(α − ᾱ)/i

The second condition implies that hψ(0)|a† a|ψ(0)i = |α0 |2 . But these two
conditions together imply that the state |φi = (a − α0 )|ψi is a null state,
i.e. it has zero norm hφ|φi = 0, as it can be easily verified. This means that
|φi = 0, that is,
a|ψ(0)i = α0 |ψ(0)i (5.2)

20
So |ψi is an eigenstate of the annihilation operator a with eigenvalue
P∞ α0 .
This condition can be solved: we assume that |ψ(0)i = n=0 cn (α0 )|ni
and then
√ find that the coefficients cn obey the recursion relation cn+1 =
α0 cn / n + 1 giving √
cn (αo ) = (α0 )n c0 (α0 )/ n! (5.3)
The normalisation hψ(0)|ψ(0)i = 1 fixes |c0 | and finally one obtains the (re-
named) state


−|α0 |2 /2
X α0n
|α0 i = e √ |ni (5.4)
n=0 n!
up to an overall phase. Thus to obtain the state with the sought properties
we see that we must excite all the energy levels up to infinity.

One important observation is the following: we have determined the state


at some fixed time (say t = 0), but what about it’s time evolution? For
that we just need to apply the time evolution operator e−iHt/~ to it and use
e−iHt/~ |ni = e−iωt/2 e−iωnt/ |ni . Thus we see that, up to the overall phase
e−iωt/2 , this amounts to replace α0 → α0 e−iωt/ . But this is just the time
evolution of α we discussed at the beginning of this section.

To justify calling the states |αi as ”semiclassical” one would expect that
the variances of various operators computed on these states should be small
compared to their expectation
p values. Indeed if one computes the variance
of the energy ∆H = hH i − hHi2 in these states one finds ∆H = ~ω|α|,
2

therefore ∆H/hHi = 1/|α| which is small for large |α| (recall that |α| is di-
mensionless) .

The coherent states involve an infinite linear combination of energy eigen-


states and in fact they can be obtained by applying to the ground state |0i an
operator involving an exponential of the creation operator a† . More precisely,
consider the operator

D(α) = eαa −ᾱa (5.5)
One sees that D is unitary: D† = D−1 . Moreover, using eA eB = eA+B+[A,B]/2 ,
valid if [A, B] commutes with A and B, one verifies the following identity:
2 /2 †
D(α)|0i = e−|α| eαa |0i = |αi (5.6)

21
This way of representing the coherent states is useful because it makes it easy
to compute the corresponding wave function Ψα (x) = hx|D(α)|0i. Indeed
using the relation between α and hxi, hpi and of a with the operators x and
p one can verify the following identity:

Ψα (x) = eihpix/~ ψ0 (x − hxi) (5.7)

where ψ0 (x) is the ground state wave function. The interpretation is clear:
Ψα (x) is a gaussian wave packet centered at hxi and moving with momentum
hpi. Contrary to wave packets obtained from superimposing plane waves,
time evolution does not deform this wave packet, because all it does is to
make hxi and hpi time dependent according to the equations of motion of
the classical harmonic oscillator.

As an application of coherent state techniques it is relatively easy to


evaluate the propagator for the harmonic oscillator, defined as

K(x, x0 ; t) = hx|e−iHt/~ |x0 i. (5.8)

Indeed, using the completeness of the |αi basis:


Z Z
−iHt/~ 0 −iHt/~
hx|e 2
|x i = d αhx|e |αihα|x i = d2 αe−iωt/2 Ψα(t) (x)Ψα (x)∗
0

(5.9)
where α(t) = e−iωt α. The α integral can easily be done6 .

6
P∞ Hn (x)Hn (y) n
Alternatively one can use Mehler’s formula: n=0 2n n! w = (1 −
2 2 2
Hn (x)exp(−x2 /2)
w2 )−1/2 exp[ 2xyw−(x
1−w2
+y )w
], together with the fact that √ √ are orthonormal.
2n n! π

22
Homework

1) Suppose you make measurements in a state of an harmonic oscillator


at t = 0 and you find that the ground state energy E0 as well as the first
excited E1 . appear with probability 1/2 . Can you determine completely the
state ? Suppose the mean value of the position operator hx̂i = 1/2. Does
this fix completely the state ? What will be then the value of hP i ?

2) A harmonic oscillator carries charge q and is subject to a constant elec-


tric field E giving an additional potential −qEx. Find the energy levels. How
are the wave functions compared to those of the standard harmonic oscillator.

3) Compute ∆x and ∆p in the state |ni of the harmonic oscillator. The


same in the coherent state |αi.

4) Compute ∆H/hHi in the coherent state for large |α|

5) Prove the completeness relation


Z
1
d2 α|αihα| = I (5.10)
π
The set |αi is overcomplete. Indeed for any |βi
Z
1
|βi = d2 α|αihα|βi (5.11)
π
but hα|βi =
6 δ(α − β). So the states |αi, α ∈ C are not linearly independent.
6) Consider the Lagrangian
1 1 ẋ22
L = ẋ21 + (5.12)
2 2 (x21 + 1)
i) Find the corresponding hamiltonian H.

ii) Noticing that p2 is a constant of motion (why ?), find the energy levels
of H.

7) Show that the propagator K(x, x0 ; t) for a free particle satisfies the
equation
~2
i~∂t K(~x, ~x0 ; t) = − ∆x K(~x, ~x0 ; t) (5.13)
2m
23
with boundary condition K(~x, ~x0 ; 0) = δ(~x − ~x0 ) (this is a heat equation with
imaginary temperature).

24

You might also like