Lecture 4 8.

321 Quantum Theory I, Fall 2017 21

Lecture 4 (Sep. 18, 2017)

4.1 Measurement
4.1.1 Spin- 12 Systems
Last time, we said that a general state in a spin- 12 system can be written as
|ψi = c+ |+i + c− |−i , (4.1)
where |+i, |−i ∈ H are eigenstates of S z and c+ , c− ∈ C. We noted that only the relative phase
between c+ and c− was physically relevant.
Assume that |ψi is normalized, so that hψ|ψi = 1, which implies that
|c+ |2 + |c− |2 = 1 . (4.2)
Now consider what happens when we measure Sz . We find
 
z ~
Prob S = + = |h+|ψi|2 = |c+ |2 ,
2
  (4.3)
z ~
Prob S = − = |h−|ψi|2 = |c− |2 .
2
We now want to consider a measurement of spin along an arbitrary axis n̂. In general, we have
  D E2
~ ~ ~ · n̂ = ± ~ ψ
Prob S · n̂ = ± = S 2 , (4.4)
2
where E
~ · n̂ = ± ~
S (4.5)
2

is the eigenket of the operator S ~ · n̂ with the eigenvalue ± ~ . This uses the fact that the eigenvalues
2
of the spin projection along any axis are always ± ~2 . We can deduce this by noting that we could
have chosen any axis originally as the quantization axis, but you should also try to prove it by
expressing a general S ~ · n̂ as a matrix and computing its eigenvalues.
For example, the eigenkets of S x are
1
S x = + ~2 = √ (|+i ± |−i) , (4.6)
2
with eigenvalues ± ~2 . We can prove this using the fact that S x corresponds to the matrix
~σ x
 
x ~ 0 1
S ↔ = . (4.7)
2 2 1 0
We can then compute
 
x ~ 2
Prob S = + = S x = + ~2 ψ
2
2
1
= √ (h+| + h−|)(c+ |+i + c− |−i)
2 (4.8)
2
1
= √ (c+ + c− )
2
1
= |c+ + c− |2 .
2
Lecture 4 8.321 Quantum Theory I, Fall 2017 22

4.1.2 The Stern–Gerloch Filter Revisited

Recall the experiment discussed in the first lecture, in which we feed a beam of spin- 12 atoms through
a region of spatially-varying magnetic field, resulting in the beam splitting into two beams, one of
which has S z = + ~2 and the other of which has S z = − ~2 . We discussed several thought experiments
using such filters oriented along various axes:

1. First, we discussed feeding the beam through a z-axis filter, and then taking the S z = + ~2
output and feeding it through yet another z-axis filter. The result is that all of the atoms
come out of the filter with S z = + ~2 . This corresponds to the collapse of the wavefunction;
the wavefunction has collapsed after the first measurement, and so the second measurement
returns the same result without further altering the wavefunction.

2. Next, we considered feeding the beam through a z-axis filter, and then taking the S z = + ~2
output and feeding it through an x-axis filter. The result is that 50% of the atoms come out
with each of S x = ± ~2 . We can see this from the calculation in Eq. (4.8), with c+ = 1, c− = 0.

3. Finally, we considered feeding the beam through a z-axis filter, feeding the S z = + ~2 output
through an x-axis filter, and then feeding the S x = + ~2 output through another z-axis filter.
The result is again that 50% of the atoms come out with each of S z = ± ~2 . In this context,
we reached the conclusion that we cannot simultaneously measure S z and S x . These are
incompatible observables.

4.2 Compatible and Incompatible Observables

We have seen that S z and S x cannot be simultaneously sharp (just like x and p). When can two
observables be simultaneously sharp? Let A, B be two Hermitian operators corresponding to two
observables. If we measure A, then the wavefunction collapses to one of the eigenstates |ai of A
with eigenvalue a. We next want to measure B and see what happens. Suppose that |ai is also an
eigenstate of B with eigenvalue b; in this case, measuring B will return the corresponding eigenvalue
b. If we remeasure A, then we will once again find the result a. Any measurements of these two
operators will always return sharp values, without changing the state after the first measurement.
Thus, the operators A, B can be simultaneously measured to have sharp values if and only if
all eigenstates of A are also eigenstates of B. Two such observables are called compatible, and
two observables not satisfying this relation are called incompatible. We can make this statement
more concise. If the eigenstates of A are exactly those of B, then A and B are simultaneously
diagonalizable, which occurs if and only if [A, B] = 0 (we proved this in the last lecture). This is the
concise condition for compatibility: two observables are compatible if and only if the corresponding
operators commute.
As an example, for the spin- 21 system, the operators S 2 = Sx2 + Sy2 + Sz2 and S z (or the spin
along any given axis) are compatible operators, while S z and S x are incompatible. Note that

[S z , S x ] = i~S y 6= 0 . (4.9)

Given a Hilbert space, we can ask what is the maximum number of mutually compatible observ-
ables we can find. A complete set of commuting observables is a set of observables {A, B, C, . . . }
such that all pairwise commutators vanish,

[A, B] = [A, C] = [B, C] = · · · = 0 , (4.10)

Lecture 4 8.321 Quantum Theory I, Fall 2017 23

and such that for any a, b, c, . . . , there is at most one solution to the eigenvalue equation

A|αi = a|αi ,
B|αi = b|αi ,
C|αi = c|αi , (4.11)
..
.

4.3 The Generalized Uncertainty Relation

Consider two observables A and B. If we measure A, we want to discuss the variance of the
probability distribution of possible outcomes. We do this in the standard way: we define the
variance of A as
∆A2 = (A − hAi)2
= A2 − hAi2 (4.12)
= ψ A2 ψ − (hψ|A|ψi)2 .
If |ψi = |ai is an eigenket of A with eigenvalue a, then

ψ A2 ψ = a2 ,
(4.13)
hψ|A|ψi = a ,

so ∆A2 = 0. In general, however, the variance will not

√ vanish.
We further define the standard deviation ∆A = ∆A2 . The uncertainty relation states that
1
∆A∆B ≥ |h[A, B]i| . (4.14)
2
(Note that this is not the strongest possible form of the uncertainty relation, but it is the most
commonly seen form.)

Proof. We will make use of the Cauchy–Schwarz inequality: for all |ai, |bi ∈ H,

|ha|bi|2 ≤ ha|aihb|bi . (4.15)

For a general |ψi, define the operators

δA = A − hAi ,
(4.16)
δB = B − |Bi ,

and the states

|ai = δA|ψi ,
(4.17)
|bi = δB|ψi .
Applying the Cauchy–Schwarz inequality, we then have

|hψ|δAδB|ψi|2 ≤ ψ (δA)2 ψ ψ (δB)2 ψ . (4.18)

Note that δA and δB are Hermitian, but there is no guarantee that their product is Hermitian.
However, we know that we can write
δAδB + δBδA δAδB − δBδA
δAδB = + . (4.19)
2 2
Lecture 4 8.321 Quantum Theory I, Fall 2017 24

The first term on the right-hand side is Hermitian, while the second is anti-Hermitian. We can
write this more concisely as
1 1
δAδB = {A, B} + [A, B] . (4.20)
2 2
Thus,
1 1
hδAδBi = h{δA, δB}i + h[δA, δB]i . (4.21)
2 2
The first term on the right-hand side must be real, as the eigenvalues of any Hermitian operator are
real, and the second term on the right-hand side must be imaginary, because any anti-Hermitian
operator can be written as i times a Hermitian operator. Thus, the squared modulus of this quantity
will simply be the sum of the squared moduli of each of these terms. We then have
1
|hδAδBi|2 = |h{δA, δB}i|2 + |h[δ A, δB]i|2 .


(4.22)
4
This statement leads to the strongest form of the inequality, but also implies the weaker statement
1
|hδAδBi|2 ≥ |h[δA, δB]i|2 . (4.23)
4
Now, note that
[δA, δB] = [A − hAi, B − hBi] = [A, B] , (4.24)
so we have
1
|hδAδBi|2 ≥ |h[A, B]i|2 . (4.25)
4
Returning to Eq. (4.18), we then see that
1
∆A2 ∆B 2 ≥ |hδAδBi|2 ≥ |h[A, B]i|2 , (4.26)
4
which completes the proof.

4.4 Position and Momentum

We will now move on to study observables with continuous eigenvalues. In order to do so, we
need to work with infinite-dimensional Hilbert spaces. Consider any Hermitian operator ξ with a
continuous spectrum
ξ ξ0 = ξ0 ξ0 , ξ0 ∈ R . (4.27)
How should we regard the overlap of two states, hξ 0 |ξ 00 i? For a discrete spectrum, we know
that the overlap of two distinct (normalized) eigenkets of a Hermitian operator are orthonormal.
We want to appropriately generalize this statement. For a discrete spectrum, the orthonormality
condition is stated as
a a0 = δaa0 . (4.28)
In the continuous case, we generalize this to

ξ 0 ξ 00 = δ ξ 0 − ξ 00 ,


(4.29)

where the Dirac delta “function” δ(x) is defined to be the object satisfying
ˆ ∞
dx δ(x) = 1 (4.30)
−∞
Lecture 4 8.321 Quantum Theory I, Fall 2017 25

6 0. This has the property

with δ(x) = 0 for all x =
ˆ ∞
dx δ(x)f (x) = f (0) . (4.31)
−∞

We can think of this as the limit of a sequence of increasingly peaked functions centered at x = 0,
each with total area 1.
We will also generalize the complete relation

1 = |aiha|
X
(4.32)
a

that is familiar from the discrete spectrum case. For continuous spectra, the equivalent statement
is ˆ ∞
1= dξ 0 ξ 0 ξ 0 . (4.33)
−∞

This is called resolution of the identity. For an arbitrary state |ψi, we then have
ˆ ∞
|ψi = dξ 0 ξ 0 ξ 0 ψ . (4.34)
−∞

We can express an arbitrary inner product as

ˆ ∞
0
ψ ψ = dξ 0 ψ 0 ξ 0 ξ 0 ψ . (4.35)
−∞
MIT OpenCourseWare
https://ocw.mit.edu

8.321 Quantum Theory I

Fall 2017

For information about citing these materials or our Terms of Use, visit: https://ocw.mit.edu/terms.