STERN-GERLACH EXPERIMENT AND BASIC PRINCIPLES OF

QUANTUM MECHANICS

Stern-Gerlach experiment

The Stern-Gerlach (SG) experiment is discussed in the introductory quantum mechanics

(QM), see also Sec. 1.5 of the textbook. Here, following Sakurai, we will show how the
results of this experiment lead to the basic principles of QM.
The quantization of the electron angular momentum is the first important feature re-
sulting from the SG experiment. According to the classical physics, a beam of atoms in
inhomogeneous magnetic field would spread itself over a vertical distance corresponding to
the continuous range of orientation of atomic magnetic moment. Instead Stern and Gerlach
observed that the beam was splitting in exactly two corresponding to two possible projec-
tions of magnetic moment. Because of the linear relation between th magnetic moment M
and angular momentum S
M = −γS

this result leads to quantization of the projection of S. SG were using silver atom whose
angular momentum is due to the intrinsic angular momentum of the valence electron, or
spin. For electron spin the gyromagnetic ratio γ is

2µB e
γ= =
h̄ mc

where µB = eh̄/(2mc) (eh̄/(2m) in SI units) is the Bohr magneton. The result of SG

experiment shows that possible values of projection of S on any axis are ±h̄/2. This result
can be considered as a definition of the Planck constant which introduces us to the quantum
domain. Of course, historically the Planck constant was introduced by him in connection with
quantization of the blackbody radiation, and then used by Bohr to create the first quantum
theory of the hydrogen atom. This is usually discussed in introductory QM courses, and you
can read Chapter 1 of the textbook to refresh your knowledge on these topics.

1
 Sequential SG experiments lead to important basic concepts of QM, the state vector and
the superposition principle. In the sequential experiment the atomic beam goes through two
or more SG apparatuses in sequence. In the first arrangement we subject the beam coming
out of the oven to the arrangement shown in Figure 1.a, where SGẑ stands for an apparatus
with the inhomogeneous magnetic field in the z-direction. We then block the Sz -component
coming out of the first SGẑ apparatus and let the remaining Sz + component be subjected
to another SGẑ apparatus. This time there is only one beam component coming out of the
second apparatus - just the Sz + component. This is perhaps not so surprising since they are
expected to remain with spin up.
A little more interesting is the arrangement shown in Figure 1.b. Here the first SG
apparatus is the same as before, but the second one (SGx̂) has an inhomogeneous magnetic
field in the x-direction. The Sz + beam that enters the second apparatus is now split into
two components, an Sx + component and an Sx − component, with equal intensities. How
can we explain this? Does it mean that 50% of the atoms in the Sz + beam coming out of the
first apparatus (SGẑ) are made up of atoms characterized by both Sz + and Sx +, while the
remaining 50% have both Sz + and Sx −? It turns out that such a picture runs into difficulty,
as we will see below.
We now consider a third step, the arrangement shown in Figure 1.c, which most dra-
matically illustrates the peculiarities of quantum-mechanical systems. This time we add to
the arrangement of Figure 1.b yet a third apparatus, of the SGẑ type. It is observed ex-
perimentally that two components emerge from the third apparatus, not one; the emerging
beams are seen to have both an Sz + component and an Sz − component. This is a complete
surprise because after the atoms emerged from the first apparatus, we made sure that the
Sz − component was completely blocked. How is it possible that the Sz − component, which
we thought, we eliminated earlier, reappears? The model in which the atoms entering the
third apparatus are visualized to have both Sz + and Sx + is clearly unsatisfactory. This
example is often used to illustrate that in quantum mechanics we cannot determine both Sz
and Sx simultaneously. More precisely, we can say that the selection of the Sx + beam by
the second apparatus (SGx̂) completely destroys any previous information about Sz .

2
 Figure 1: Sequential Stern-Gerlach experiment (taken from Sakurai).

This all suggests that we might be able to represent the spin state of a silver atom by
some kind of vector in a new kind of two-dimensional vector space, an abstract vector space
not to be confused with the usual two-dimensional (xy) space. We represent the Sx + state
by a vector, which we call a ket in the Dirac notation to be developed fully further. We
denote this vector by |Sx ; +i, or |Sx ↑i, and write it as a linear combination of two base
vectors, |Sx ↑i and |Sx ↓i, which correspond to the Sz + and the Sz − states, respectively. So
we may postulate
1 1
|Sx ↑i = √ |Sz ↑i + √ |Sz ↓i (1)
2 2
1 1
|Sx ↓i = √ |Sz ↑i − √ |Sz ↓i (2)
2 2
Thus the unblocked component coming out of the second (SGx̂) apparatus of Figure 1.3c is
to be regarded as a superposition of Sz + and Sz − in the sense of (1). It is for this reason
that two components emerge from the third (SGẑ) apparatus.
The next question is about the representation of the Sy ± states. Symmetry arguments
suggest that if we observe an Sz ± beam going in the x-direction and subject it to an SGŷ
apparatus, the resulting situation will be very similar to the case where an Sz ± beam going in

3
the y-direction is subjected to an SGx̂ apparatus. The kets for Sy ± should then be regarded
as a linear combination of |Sz ; ±i, but it appears from (1), (2) that we have already used up
the available possibilities in writing |Sx ; ±i. How can our vector space formalism distinguish
Sy ± states from Sx ± states? We can use the analogy with construction of circularly polarized
light as a linear combination of two linearly polarized beams (for more details see Sakurai):

1 h i
Ecirc = √ x̂ei(kz−ωt) + iŷei(kz−ωt)
2
which suggests the following expression for Sy ± states:

1
|Sy ±i = √ (|Sz +i ± i|Sz −i) . (3)
2
We thus see that the two-dimensional vector space needed to describe the spin states
of silver atoms must be a complex vector space; an arbitrary vector in the vector space
is written as a linear combination of the base vectors |Sz ; ±i with, in general, complex
coefficients. The fact that the necessity of complex numbers is already apparent in such an
elementary example is rather remarkable.

Hilbert space in Quantum Mechanics

Here we will formulate some principles of QM following from the above discussion and
from the theory of the Hilbert space. A mathematical introduction to Hilbert space is given
in separate notes (’text-916-hilbert.pdf’), see also Chapter 5 of the textbook. Note, however,
that Bransden and Joachain use the position-space notations. All equations in Chapter 5
can be rewritten in the abstract Hilbert space without using the position coordinates.
In QM a physical state, for example, a silver atom with a definite spin orientation, is
represented by a state vector in a complex vector space (Hilbert space). Following Dirac,
we call such a vector a ket and denote it by |ψi. This state ket is postulated to contain
complete information about the physical state; everything we are allowed to ask about the
state is contained in the ket. A linear combination of two kets

|ψi = c1 |ψ1 i + c2 |ψ2 i

4
is just another ket (c1 , c2 are complex numbers). This is called the superposition principle.
One of the physics postulates is that |ψi and c|ψi, with c 6= 0, represent the same physical
state. In other words, only the “direction” in vector space is of significance. If in the product
c|ψi the complex number c is zero, the resulting ket is said to be a null ket.
An observable, such as momentum and spin components, can be represented by an oper-
ator A in the vector space in question. Quite generally, an operator acts on a ket from the
left,
A(|ψi) = A|ψi

which is yet another ket.

In general, A|ψi is not a constant times |ψi. However, there are particular kets of
importance, known as eigenkets of operator A, denoted by |ψ1 i, |ψ2 i,... with the property

A|ψ1 i = a1 |ψ1 i, A|ψ2 i = a2 |ψ2 i, ... (4)

where a1 , a2 ,... are just numbers. The set of these numbers more compactly is called the set
of eigenvalues of operator A. The eigenvalues and eigenvectors of a Hermitian operator and
their properties are discussed on pp. 4,5 of the notes on the Hilbert space.
The dimensionality of the vector space formed by eigenkets is determined by the number
of eigenkets which physically corresponds to the number of alternatives in SG-type exper-
iments. In the case of the SG experiment the number of alternatives is two, therefore the
space is two-dimensional. More generally we are concerned with an N -dimensional vector
space spanned by the N eigenkets of observable A. Any arbitrary ket |ψi can be written as
X
|ψi = cn |ψn i (5)
n

where cn , n = 1, 2...N are complex coefficients. The vector space can have an infinite
dimension.
Eq. (5) is called the completeness theorem in the theory of Hilbert space. If ψn are
eigenstates of a Hermitian operator, then the coefficients cn can be found as (compare to
(5.61) of the textbook and p. 6 of the notes on Hilbert space)

cn = hψn |ψi (6)

5
where hψn | is the vector dual, or conjugated, to |ψn i, the bra vector according to the Dirac
terminology, and h...|...i is the inner product of bra and ket which is a complex number.
According to the statistical interpretation of quantum mechanics, the coefficients cn are
named the probability amplitudes such that |cn |2 is the probability to obtain the eigenvalue
an as a result of measurements of an observable A if the system is in the state |ψi.
Eq. (5) can be rewritten now as

X
I= |ψn ihψn | (7)
n

where I is the identity operator. Eq. (7) is called the closure relation (compare to Eq. (5.63a)
of the textbook).
Consider now the operator A whose eigenstates are |ψn i. Acting by A on both sides of
Eq. (7) we obtain the spectral representation of the operator A

X
A= an |ψn ihψn |. (8)
n

Suppose we have a basis of kets |φn i which are not necessarily eigenkets of A. We can
form a matrix representation of A by constructing

Anm = hφn |A|φm i.

Matrix representation of a state ket |ψi is given by hφn |ψi which is usually written as a
column  
hφ1 |ψi
|ψi =  hφ2 |ψi  . (9)
 

...
The corresponding state bra is the Hermitian conjugate of this column
 
hψ| = hφ1 |ψi∗ hφ2 |ψi∗ ... . (10)

Expectation values

Using the probabilistic interpretation of the amplitudes cn , Eq. (6), we can calculate
the average values, or expectation values as they are called in QM, of physical quantities.

6
According to the probability theory, the average value of quantity A is

an |cn |2
X
hAi =
n

Using Eq. (6) we can write

X X
hAi = hψ|ψn ian hψn |ψi = hψ|A|ψn ihψn |ψi.
n n

Finally, using closure, Eq. (7), we obtain

hAi = hψ|A|ψi.

Spin space

In many cases the Hilbert space is a functional space of infinite dimension, and the state
vector is the wave function in the position (or momentum) space. But this is not always the
case. For example, in the general theory of angular momentum the state vector is represented
by a column, called spinor, with a finite size 2j + 1 determined by the angular momentum
quantum number j of the system. In particular the electron spin (j = s = 1/2) is represented
by a spinor with two elements, and the operators in the corresponding Hilbert space are
represented by 2x2 matrices. In the quantum field theory the systems are represented by
abstract state vectors acting in the occupation number space of infinite dimension.
As a first example, we will discuss the state vectors and operators in the spin space. In
the simplest case of spin 1/2 systems, the eigenvalue-eigenket relation (4) is expressed as

h̄ h̄
Sz |Sz ↑i = |Sz ↑i, Sz |Sz ↓i = − |Sz ↓i,
2 2

where |Sz ↑i, |Sz ↓i are eigenkets of operator Sz with the eigenvalues ±h̄/2.
From Eqs. (7), (8)
I = |Sk ↑ihSk ↑ | + |Sk ↓ihSk ↓ |

(k = 1, 2, 3 corresponding to the Cartesian coordinates x, y and z) and

Sk = h̄/2[|Sk ↑ihSk ↑ | − |Sk ↓ihSk ↓ |] (11)

7
Using the statistical interpretation of QM and the matrix representation of state kets, Eq.
(9), we can form the state vectors |Sz ↑i and |Sz ↓i as two-row column kets
! !
1 0
|Sz ↑i = |Sz ↓i = .
0 1

Now from (1), (2), (3) we have the following kets.

! !
1 1 1 1
|Sx ↑i = √ |Sx ↓i = √ , (12)
2 1 2 −1
! !
1 1 1 1
|Sy ↑i = √ |Sy ↓i = √ . (13)
2 i 2 −i
√
The normalization coefficient 1/ 2 results from the fact that the total probability is 1. The
corresponding bras are found as Hermitian conjugated of kets, according to Eq. (10)
   
hSz ↑ | = 1 0 hSz ↓ | = 0 1 ,

1   1  
hSx ↑ | = √ 1 1 hSx ↓ | = √ 1 −1 ,
2 2
1   1  
hSy ↑ | = √ 1 −i hSy ↓ | = √ 1 i .
2 2
Using these base kets and bras and Eq. (8), we obtain the matrix representation of the
operators Sk
! ! !
h̄ 1 0 h̄ 0 1 h̄ 0 −i
Sz = , Sx = , Sy = . (14)
2 0 −1 2 1 0 2 i 0

It is customary to write these operators as

h̄
Sk = σk
2

where σk , k = 1, 2, 3 are Pauli matrices

! ! !
1 0 0 1 0 −i
σz = , σx = , σy = . (15)
0 −1 1 0 i 0

8
 Wavefunction

The projection of a state vector |ψi on the eigenstate of the position operator r is called the
wavefunction. Formally
ψ(r) = hr|ψi.

Generally the wavefunction depends on time as well. In this case we will be using the capital
Ψ(r, t). Similar we can define the wavefunction in the momentum space as

φ(p) = hp|ψi.

The wavefunction plays a very important role in QM. In fact the Schrödinger’s wave mechan-
ics was formulated independently of the operator and Hilbert-space formulation of QM due to
Heisenberg and Dirac. However, for the complete formulation QM requires the Heisenberg-
Dirac approach.
The experimental confirmation of the wave properties of matter was given by the electron
diffraction experiments. In the limit of the short wavelength the QM description of matter
should turn to the classical description. To obtain the wavefunction in this limit we can
use analogy with the limit of the geometric optics in the electromagnetic theory. Any field
component u in the electromagnetic wave propagating in a medium can be written in the
form (see Sec. 8.10 of J. D. Jackson)

u(r, t) = ei[φ(r)−ωt]

where φ is the phase which varies fast if the wavelength is short. It can be written in the
form
ω
φ = i S(r)
c
where the function S(r) with dimension of length is called eikonal. In the short-wave-length
limit it satisfies an equation similar to the Hamilton-Jacobi equation for action in classical
mechanics
(∇S)2 = n2 (r)

9
where n(r) is the refraction index. The path of the rays in the limit of geometrical optics is
determined by the Fermat’s principle according to which the optical path length given by
the eikonal S(r) must take its least possible value. And in classical mechanics the Hamilton-
Jacobi equation can be obtained from the principle of least action.
On the basis of this analogy we can assume that the classical trajectory is obtained from
QM wavefunction by requirement that its phase has the least value. Therefore the phase of
the wave function should be
S(r, t)
φ=
const
where S(r, t) is the action, or the principal Hamilton’s function. The constant in this equation
has dimension of action, and the diffraction experiments show that it equals h̄ which is
consistent with the de Broglie theory of the matter waves. This is another way to introduce
the Planck constant, in addition to the way we introduced it in connection with the Stern-
Gerlach experiment. Classical limit happens when the angular momentum of a particle
becomes large compared to h̄ or the action associated with the particle’s motion is large
compared to h̄.
In this context let us look again at the Stern-Gerlach experiment and question what part
of it can be described classically. The atomic spin of the silver atom h̄/2 is comparable
with the Planck constant, therefore it is a pure QM quantity which does not have a classical
analog. On the other hand, the action S correponding to the classical path of the macroscopic
length, say, 1 cm, is of the order of 10−25 J·s (see Prob. 1.23) which is much larger than h̄.
Therefore the description of the atom motion along the classical trajectory is valid.
In conclusion, in classical limit

i
 
Ψ(r, t) = exp S(r, t)
h̄

In this limit we have

i
∇Ψ(r, t) = ∇S(r, t)Ψ(r, t).
h̄
From the classical mechanics we know

p = ∇S

10
where p is the momentum of the particle. We conclude that Ψ in this limit is an eigenstate
of the operator ∇ with the eigenvalue ip/h̄. We can therefore introduce the momentum
operator
h̄
pop = ∇
i
whose eigenvalue is momentum p. Similar using

∂ i ∂S(r, t)
Ψ(r, t) = Ψ(r, t)
∂t h̄ ∂t

and the classical relation

∂S
E=−
∂t
we introduce the energy operator
∂
Eop = ih̄ .
∂t
Using the classical relation between the energy and momentum for a free particle

p2
E=
2m

and replacing classical quantities by corresponding operators, we obtain the Schrödinger

equation for a free particle
∂Ψ h̄2 2
ih̄ =− ∇Ψ
∂t 2m
whose particular solution is the de Broglie’s plane wave

Ψ(r, t) = C exp[i(p · r − Et)/h̄] = C exp[i(k · r − ωt)]

where k = p/h̄, ω = E/h̄. It is important that, in contrast to the classical wave equations
for electromagnetic waves, the Schrödinger equation is complex, therefore its solution is
complex. In particular cos(k · r − ωt) is not a solution of the Schrödinger equation.

Functional Hilbert space

We will discuss now the wavefunctions in the position space in the context of the theory
of Hilbert space. For simplicity let us talk about one-dimensional systems, that is systems

11
described by one position coordinate x, and ignore the time dependence. Mathematically
wavefunctions are represented by vectors in the functional space f (x), g(x) etc. The inner
product in this space is defined as
Z b
hg|f i = g ∗ (x)f (x)dx, (16)
a

where [a, b] is an interval on the real axis where the functions are defined. In particular we
can have a = −∞, b = ∞. Therefore the norm of the function f (x) is
Z b !1/2
2
||f || = |f (x)| dx .
a

Sometimes it is more convenient to define the inner product as

Z b
hg|f i = g ∗ (x)f (x)w(x)dx, (17)
a

where w(x) is called a weight function.

An example of a complete set of vectors in a functional space is a set of monomials 1, x,
x2 , on the interval [-1,1] with the inner product defined by Eq. (16). However, this set is not
orthonormalized. To make it such, we can use the Gram-Schmidt orthonormalization method
familiar to us from the theory of the Euclidean space. This procedure is accomplished, for
example, on pp. 182-183 of the Boas book ’Mathematical Methods in Physical Sciences’. As
a result, we obtain a set of functions which turn out to be normalized Legendre polynomials.
Another example is a set of monomials defined on the interval (−∞, ∞) with the inner
product (17) with the weight function w(x) = exp(−x2 ). The Schmidt-Gram method leads
to a set of polynomials which are normalized Hermite polynomials.
Not all wave functions in quantum mechanics have a finite norm (or normalizable). A
typical example is the plane wave eikx , but generally any eigenstate corresponding to an
eigenvalue belonging to the continuum spectrum does not have a finite norm. In order to
apply the general theory of the Hilbert space in this situation, we can use two methods.
The first approach, that is widely used in the quantum field theory, is to treat the quantum-
mechanical system in a finite volume with some boundary conditions on the surface enclosing
this volume. The specific form of these boundary conditions does not affect the physics in

12
most cases. For mathematical convenience the periodic boundary conditions are usually
used. In one-dimensional case we can place the system in a box with boundaries x = −L/2
and x = L/2 where L is a length which is very large compared to the size of the system.
Applying the periodic boundary condition f (L/2) = f (−L/2) to the plane wave, we obtain

eikL/2 = e−ikL/2

or
kL = 2πn, n = 0, ±1, ±2, ...

that leads to the discrete eigenvalues kn = 2πn/L. Formally this means that all integrals
over k in the expansion over plane waves are replaced by sums, but in fact, since L is very
large, the spectrum of k is almost continuous (or, as we say, quasicontinuous), and in all
practical calculations the sum over k is replaced by the integral according to the prescription
X L Z
→ dk.
k 2π

With this boundary condition the functions

1
fn (x) = √ eikn x
L

form an orthonormal set on the interval (−L/2, L/2). This set is complete in the sense that
any function defined on the interval (−L/2, L/2) can be expanded in fn (x) (In fact this is
the complex form of the Fourier series). Similarly the functions

1
fk (r) = √ eik·r
V

form a complete orthonormal set in 3-d space if the periodic boundary conditions are applied
for each Cartesian coordinate to the sides of a cube of volume V = L3 . In this case the vector
k is defined by three integers nx , ny , nz so that

2πnx 2πny 2πnz

 
k= , .
L L L

Another way to solve the normalization problem for the continuous spectrum, which is
usually used in the nonrelativistic quantum mechanics, is to use the Dirac delta-function

13
normalization. Taking again the one-dimensional plane wave as an example, we use

1 Z ∞ i(k−k0 )x
e dx = δ(k − k 0 ) (18)
2π −∞

where δ(y) is a function with the following properties

(
∞ if y = 0
δ(y) =
0 otherwise

and
Z ∞
δ(y)dy = 1.
−∞

The fact that the function in the left-hand-side of Eq. (18) satisfies the definition of the
delta function can be proven by considering the function

1 Z L/2 iyx
δL (y) = e dx
2π −L/2

in the limit L → ∞. Other examples of the delta-function representation (the last four are
from the homework, Prob. 2.5)
Z ∞
−1 2
δ(x) = lim(2π) eikx−k dk
→0 −∞

δ(x) = lim
→0 π(x2 + 2 )
sin(x/)
δ(x) = lim
→0 πx

δ(x) = lim [1 − cos(x/)]
→0πx2
θ(x + ) − θ(x)
δ(x) = lim
→0 
where θ(x) is the Heaviside (step) function
With all these caveats for the continuum spectrum, all general theorems for the Hilbert
space can be applied to quantum-mechanical wave functions. In particular, if we have a set
of eigenstates (eigenvectors) φn (x) of a Hermitian operator, the completeness theorem gives
X
f (x) = cn φn (x) (19)
n

where
Z
cn = φ∗n (x)f (x)dx.

14
 Applying this theorem to the delta-function δ(x − x0 ) we obtain the closure identity (eq.
(5.63a) of the textbook)
δ(x − x0 ) = φ∗n (x0 )φn (x).
X

The integral representation of the delta-function

1 Z ik(x−x0 )
δ(x − x0 ) = e dk
2π

can be considered as a particular case of the closure identity when the summation over
discrete states is replaced by the integration over the continuum of k.
Specific applications of these fundamental principles will be discussed in this course.

15