Quantum Mechanics 1 Lecture 13:
Quantum Theory
and Classical physics
Massimiliano Grazzini
University of Zurich
Outline
Quantum information
- pure vs mixed states The density matrix
Entanglement EPR experiment
Hidden variables and the Bell theorem
- quantum mechanics vs local realism
Quantum information
The theory of quantum information deals with how a quantum system can be used to accumulate, transmit and elaborate information
In the following we just want to discuss the simplest questions in the case of the simplest quantum system, the two-state system (qubit)
How can we extract the information in quantum state?
How can we code it ?
How can we determine the state of a quantum system ?
Bit Qubit
|χ⟩ = a+| + ⟩ + a−| − ⟩ a+, a− ∈ ℂ
with
|a+|2 + |a−|2 = 1In Lecture 10 we have seen that an arbitrary normalised spin state
|χ⟩can be written as
|χ⟩ = cos(θ/2)| + ⟩ + eiϕ sin(θ/2)| − ⟩
with
0 ≤ θ ≤ πand
0 ≤ ϕ ≤ 2πThis parametrization in terms of only two real parameters is possible because the overall phase is irrelevant
This holds for a state
|ψ⟩of a general two-level system
We now ask ourselves how can we determine the state of the system having many copies of the system in the same state
|ψ⟩but not knowing how it was prepared
Suppose that we can measure an observable and let us choose the basis
corresponding to its eigenvectors
A | ± ⟩A| + ⟩ = λ+| + ⟩ A| − ⟩ = λ−| − ⟩
The expectation value of this observable on the state
|ψ⟩is
⟨ψ |A|ψ⟩ = ⟨A⟩ = P+λ+ + P−λ−
where
P+ = |a+|2 P− = |a−|2
These probabilities are fully determined by the value of and by the normalisation condition
P+ + P− = 1 ⟨A⟩Therefore we can fix the coefficient or, equivalently, the angle , but we
cannot say anything on the relative phase, since a measurement of does not depend on it
a+ θ
A
To learn more about our system we have to measure other observables which do not commute with
AIndeed if commutes with we can write
A′ AA′| + ⟩ = λ′+| + ⟩ A′| − ⟩ = λ′−| − ⟩
and thus
⟨A′⟩ = P+λ′+ + P−λ′−which still allows us to determine the probabilities but not the relative phase
The density matrix
Where the trace is defined as
For a system in a state
|ψ⟩the density matrix is the hermitian operator
ρψ ≡ |ψ⟩⟨ψ |
This matrix allows us to easily write the expectation value of an observable :
we have
A⟨A⟩ = Tr (Aρ)
To proceed further it is useful to introduce the density matrix
Tr(O) = ∑
i
⟨ei|O|ei⟩
and
|ei⟩is an orthonormal basis of the system
Indeed since the trace is independent on the basis, we can choose a basis of eigenstates of and we find
ATr(Aρ) = ∑
i
⟨ei|Aρ|ei⟩ = ∑
i
⟨ei|A|ψ⟩⟨ψ |ei⟩ = ∑
i
λi|⟨ei|ψ⟩|2 = ⟨A⟩
The density matrix has the property of having unit trace
Tr(ρ) = ∑
i
⟨ei|ρ|ei⟩ = ∑
i
⟨ei|ψ⟩⟨ψ |ei⟩ = ∑
i
|⟨ei|ψ⟩|2 = 1
The density matrix allows us to describe the case in which we don’t have
complete information on the state of a quantum system, but we just know the probabilities of the states
We assume that we have an ensemble of states and we distinguish two cases
The system or objects under investigation are all in the same state : we say that the system is in a pure state or pure ensemble (for example: beam of spin 1/2
particles all with the same polarisation)
|ψ⟩
Pure state
We can investigate such ensemble of identically prepared states Suppose that
|ψ⟩ = ∑
i
ci|ei⟩
where the orthonormal states
|ei⟩correspond to values of the observable
ai AIf we have identical copies of the state and we measure we can call the number of times we obtain the result
N |aψ⟩i A NiThe larger is and the better the ratio
N Ni/Nwill approach
|ci|2|ci|2 = lim
N→∞
Ni N
In this limit the expectation value of and the statistical average coincide
A⟨ψ|A|ψ⟩ = ∑
i
|ci|2ai = lim
N→∞
1
N ∑
i
Niai
Mixed state
Our system is a statistical superposition of states: we say that the system is in a mixed state, or a mixed ensemble (for example: beam of spin-1/2 particles of which we do not know the polarisation)
The system is in each of the states with probability : the density matrix is
defined as
|ψi⟩ Piρ ≡ ∑
i
Pi|ψi⟩⟨ψi|
Statistical average
Note that a mixed state actually corresponds to a classical mixture of pure states, not to their quantum superposition, and that the
|ψi⟩need not be orthogonal
The pure state is obtained back in the case in which for a given and
for
Pi = 1 iPj = 0 j ≠ i
The density matrix defined in this way allows us to compute the average of an observable as
Tr(Aρ) = ∑
i
⟨ei|Aρ|ei⟩ = ∑
ij
⟨ei|A|ψj⟩⟨ψj|ei⟩Pj = ∑
ij
⟨ψj|ei⟩⟨ei|A|ψj⟩Pj
= ∑
j
⟨ψj|A|ψj⟩Pj
that is, the weighted average of the expectation values over the states ,
which we assume to be normalised
|ψj⟩The density matrix is a semi-positive operator: indeed for any state
|ψ⟩we have
⟨ψ |ρ|ψ⟩ = ∑
i
Pi⟨ψ |ψi⟩⟨ψi|ψ⟩ = ∑
i
Pi|⟨ψ |ψi⟩|2 ≥ 0
A measurement of always produces the result
ATr(Aρpure) = 12 Tr
[(0 1
1 0) (1 1
1 1)] = ⟨A⟩ψpure = 1
We can illustrate the difference between a pure and a mixed state with a simple example
Let us consider the pure state defined as
|ψpure⟩ = 12 (| + ⟩ + | − ⟩)
ρpure = |ψpure⟩⟨ψpure| = 12 (1 1 1 1)
This is an eigenvector with eigenvalue 1 of the operator
A = (0 11 0) A|ψpure⟩ = |ψpure⟩
Pure vs mixed states
Let us now consider a mixed state, with 50% probability of being in and
50% probability of being in
| − ⟩ | + ⟩The corresponding density matrix is
Indeed we have
Tr(Bρpure) = ⟨B⟩ψpure = 0 Tr(Bρmixed) = 1
2 ⟨ + |B| + ⟩ + 1
2 ⟨ − |B| − ⟩ = 0 ρmixed = 12 | + ⟩⟨ + | + 12 | − ⟩⟨ − | = 12 (1 0
0 1)
The density matrix is
A measurement of gives in this case
ATr(Aρmixed) = 12⟨ + |A| + ⟩ + 12⟨ − |A| − ⟩ = 0
The result for the average value of is different !
ATherefore a suitable measurement allows us to distinguish a statistical mixture from a quantum superposition of the same states
We note, however, that if we now consider the operator which is diagonal on the basis we have chosen a measurement of gives the same results in both states
B = (1 0 0 −1) B
What are the differences in the density matrix in the case of pure and mixed states ?
This is the special case of a general result: in the case of absence of information the density matrix is a multiple of the identity
We conclude that
pure state
Tr(ρ2) = 1 Tr(ρ2) < 1
mixed state and that for a pure state the density matrix is a projection operator It is easy to see that in a pure state we have
ρ2 = ρand thus
Tr(ρ2) = 1Indeed
ρ2 = |ψ⟩⟨ψ |ψ⟩⟨ψ | = |ψ⟩⟨ψ | = ρOn the other hand in a mixed state we have
Tr(ρ2) = Tr ∑
i
Pi|ψi⟩⟨ψi| × ∑
j
Pj|ψj⟩⟨ψj| = ∑
i,j,k
⟨ek|ψi⟩⟨ψi|ψj⟩⟨ψj|ek⟩PiPj
= ∑
i,j
|⟨ψi|ψj⟩|2PiPj = ∑
i
Pi2 + ∑
i≠j
|⟨ψi|ψj⟩|2PiPj < ∑
i
Pi2 + ∑
i≠j
PiPj
= ∑
i,j
PiPj =
(∑
i
Pi )
2
= 1
where we have used
|⟨ψi|ψj⟩| < 1if
i ≠ jand the normalisation of the states
Tr(ρ2) < 1
We now ask ourselves what are the most general measurement and density
operator for our two-level system
In the case of a two-level system these objects can be fully characterised: both must indeed be
2 × 2hermitian matrices
A general observable is an arbitrary hermitian operator , while the most general density matrix is an arbitrary hermitian operator with unit trace
AρA
2 × 2hermitian matrix is fully determined by real parameters
Mij 4Indeed the diagonal elements must be real and the off diagonal elements must be one the complex conjugate of the other
Mii = M*ii i = 1,2 M12 = M*21
We can thus introduce four linearly independent hermitian matrices and write as a linear combination of them
M
A convenient choice is represented by the Pauli matrices and the identity matrix
σ1 = (0 1
1 0) σ2 = (0 −i
i 0 ) σ3 = (1 0 0 −1) σ0 = I = (1 0
0 1)
The most general observable is thus
A = a0I + a ⋅ σ
The most general density matrix takes the same form with the additional constraint that
Tr(ρ) = 2a0 = 1
ρ = 12 (I + n ⋅ σ) = 12 ( 1 + n3 n1 − in2 n1 + in2 1 − n3 )
We can write it in the form
Since is a semi-positive hermitian operator its eigenvalues must be real and positive or vanishing, which means that
ρdet(ρ) ≥ 0
On the other hand for a pure state must be a projection operator, which means that, in our case, in some basis it must take the form
ρρ = (1 0 0 0)
And, therefore for a pure state
det(ρ) = 0
(1)
Now, in the parametrisation the determinant of is
(1) ρdet(ρ) = 1
4 (1 − |n|2)
Therefore, since
det(ρ) ≥ 0the density matrices can be described as
ρ = 12 (I + n ⋅ σ) |n|2 ≤ 1
This is the Bloch sphere, i.e. the sphere of unit radius we have encountered in Lecture 10, whose surface ( and ) describes the most general pure state
|n| = 1 det(ρ) = 0ρmixed = 12 | + ⟩⟨ + | + 12 | − ⟩⟨ − |
We can now go back to the case of the mixed state considered before
In this case there is 50% of probability of being in and 50% probability of being in
| − ⟩: the density matrix is proportional to the identity
| + ⟩ρmixed = 12 I
Which corresponds to the case
n = 0in the parametrisation
(1)Therefore to determine the it is enough to measure three observables that are linearly independent when expanded in terms of Pauli matrices
niIn the case of a pure state two such measurements are sufficient, since the third component can be fixed through the condition
|n| = 1We can now answer our original question on how can we determine the state of a system assuming we can prepare many copies of it
The measurement of an observable whose operator is a Pauli matrix gives
σiTr(ρσi) = Tr (1
2 (I + n ⋅ σ) σi) = ni
𝒫± = 1
2 (I ± n ⋅ σ)
The above discussion also lead us to the conclusion that for
|n| = 1the operator
is the projector on a
±1/2state along the direction
n(2)
Entanglement
Its state vector lives in the direct product space of two Hilbert spaces
ℋ1 ⊗ ℋ2|ψ⟩ = ∑
m,n
cmn|m1⟩ ⊗ |n2⟩ |m1⟩ ∈ ℋ1 |n2⟩ ∈ ℋ2
We note that, while the basis vectors are a direct product of basis vectors in the two spaces, their superposition is not
|m1⟩ ⊗ |n2⟩A direct product vector would have the form
|ϕ⟩ =
(∑
m
cm|m1⟩
) ⊗
(∑
n
dn|n2⟩ )
But not all the states
|ψ⟩will have this form
A state that cannot be written in a factorised form is said to be entangle (in German verschränkt)
Entangled states reveal the non-local nature of the wave function
Suppose that we have a system which is described by two parts and
1 2We have seen that the formalism of the density matrix allows us to characterise the behaviour of a quantum system when a measurement is performed
A particularly interesting situation is when we carry out a measurement only on a part of the complete system
ρ = |ψ⟩⟨ψ| = ∑
m,n,m′,n′
c*m′n′cmn|m1⟩|n2⟩⟨m′1|⟨n2′|
where we have used the shorthand notation
|m1⟩ ⊗ |n2⟩ = |m1, n2⟩For the state
|ψ⟩ = ∑m,n
cmn|m1⟩ ⊗ |n2⟩
The density matrix is
Let us consider an observable acting only on the second subspace: its action
is diagonal on the subspace and therefore its expectation value is
1 A2The result for
⟨A2⟩can also be expressed as
⟨A2⟩ = Tr2 (A2ρ2)
where we have defined a density matrix reduced to the second subspace
ρ2ρ2 = Tr1ρ = Tr1 ∑
m,n,m′,n′
c*m′n′cmn|m1⟩|n2⟩⟨m′1|⟨n′2| = ∑
m,n,n′
c*mn′cmn|n2⟩⟨n2′|
⟨ψ|A2|ψ⟩ = ∑
m′,n′∑
m,n
c*m′n′cmn⟨m′1, n′2|A2|m1, n2⟩
= ∑
m,n,n′
c*mn′cmn⟨n′2|A2|n2⟩
This is of course a rather common situation: the entire universe is in principle
described by a unique wave function, but in practice a measurement is always
done on small subsystem of it
As an explicit example let us consider a system of non-identical spin-1/2 particles in the pure state
|ψ⟩ = 1
2 (| + ⟩1| − ⟩2 + | − ⟩1 + ⟩2)
Its density matrix is
ρ = |ψ⟩⟨ψ | = 12 (| + ⟩1| − ⟩2 + | − ⟩1 + ⟩2) (⟨1 + |⟨2 − | + ⟨1 − |⟨2 + |)
Suppose that we want to measure the spin of the second particle: we need to consider the density matrix obtained by taking the trace with respect to 1
ρ2ρ2 = Tr1ρ = 1
2 [⟨ + |1(| + ⟩1| − ⟩2 + | − ⟩1 + ⟩2) (⟨1 + |⟨2 − | + ⟨1 − |⟨2 + |)| + ⟩1] + 12 [⟨ − |1(| + ⟩1| − ⟩2 + | − ⟩1 + ⟩2) (⟨1 + |⟨2 − | + ⟨1 − |⟨2 + |)| − ⟩1]
= 12 [| − ⟩2⟨ − |2 + | + ⟩2⟨ + |2]
But this is the density matrix for a completely unpolarised mixed state (recall
the discussion at page )
11We can thus conclude that, even if the system is in a pure state, under a partial measurement the system may behave as a mixed state
In other words, a quantum superposition of states can become a classical mixture if we average over degrees of freedom that are not measured
One can show that a necessary and sufficient condition under which the density matrix of a pure state still corresponds to a pure state when a
measurement is performed over a subsystem is that the state can be written in a factorised form
If we can write
|ψ⟩ = |ψ1⟩|ψ2⟩
then
ρ22 = ρ2
This explains why in many cases it is possible to describe a quantum system in terms of the Hamiltonian of a part of it
Even if the entire universe is described by a single wave function, in practice, if
factorisation holds, one can focus on a subsystem of it !
Einstein-Podolsky-Rosen experiment
The EPR Gedankenexperiment considers an entangled system formed by two subsystems
A simplified version due to Bell can be stated as follows: we have a spin-zero particle which decays in two spin
1/2particles, say an electron and a positron Since the total angular momentum is null, the wave function is
|ψ0⟩ = 1
2 (| + ⟩| − ⟩ − | − ⟩| + ⟩)
The two particles move away from the source in such a way that time evolution does not modify the wave function, and when the particles are sufficiently distant spin measurements are carried out that are causally disconnected
Local realism states that, since the particles are causally disconnected, there is nothing that can happen to one of them as a consequence of what we do on the other, or, using the words of the original paper
..since at the time of measurement the two systems no longer interact, no real
change can take place in the second system in consequence of anything that may be
done in the first system.
Quantum mechanics contradicts these statements
Indeed, if we measure the spin of the first particle in the state and we find spin up, after the measurement the wave function of the system is projected onto
|ψ0⟩|ψ0′⟩ = | + ⟩| − ⟩
and the spin of the second particle is down
So, it seems that there is transfer of information at a speed larger than the speed of light and a violation of causality
Such conclusion is, however, not correct
We can understand it by using the example of Bertlmann socks Suppose that there are two brothers who choose the socks in a drawer that contains only two pairs: one is blue and the
other is not
Local realism states that a property of the system which is measurable must
necessarily be a property of the system before the measurement is done
If, during the day, when the brothers are far away, we observe that the socks of one of them are not blue, we can immediately conclude that the socks of the other are blue
Of course this does not violate local realism, even if the brothers are very far away:
this is because the colour of the socks was determined before being measured When they dress up we don’t know who will wear the blue socks and there is
probability that the socks of each brother are of one colour or of the other
50 %Let’s now go back to our state
|ψ0⟩: the problem comes when we realise that
|ψ0⟩ = 1
2 (| + ⟩| − ⟩ − | − ⟩| + ⟩)
can also be written as a superposition of spin states along an arbitrary axis For example, since
|ψ0⟩ = 1 (| + ⟩x| − ⟩x − | − ⟩x| + ⟩x)
we have
| ± ⟩x = 1
2 (| + ⟩ ± | − ⟩)
Therefore, if we measure the spin of the first particle along the axis, then the second particle has opposite spin along the axis. But if we do the measurement along the axis, then the second particle will have opposite spin along the axis
z z
x x
But the spin operators along the two directions do not commute
The second particle cannot be before the measurement in a state with given spin both along the and the direction
z xThe explanation used for Bertlmann socks does not work here, and thus quantum mechanics contradicts the hypothesis of local realism
Since for Einstein and collaborators the local realism is principle they don’t want to abandon, they conclude that Quantum Mechanics cannot be complete
Einstein expressed his rejection of Quantum Mechanics with the famous sentence:
“God does not play dice”
This expresses the non-local character of quantum theory: the non locality is a
consequence of the existence of correlated many-particle states and the fact that one
can linearly superimpose them
Hidden variables and the Bell theorem
By insisting on local realism, one can assume that a more complete theory exists and that a result of a measurement is determined by a property that the system possesses before the measurement
We now show that such description leads to results different from those of quantum mechanics
We can therefore introduce the hypothesis that the complete state of the system is described not only by the variables we use in the quantum-mechanical
description but also by some additional hidden variables unknown to us
If we knew such hidden variables we could predict with certainty the result of a measurement of any operator
We go back to the EPR experiment: we assume that the electron and the positron travel far apart in the state
|ψ0⟩ = 1
2 (| + ⟩| − ⟩ − | − ⟩| + ⟩)
α β
Alice Bob
ℏ/2
−ℏ/2 ℏ/2
−ℏ/2
Source
The density matrix for this state is
ρ = |ψ0⟩⟨ψ0| = 1
4 (I − σ1 ⋅ σ2)
where the and are Pauli matrices acting on the spin states of the two particles
σ1 σ2This result can be easily checked by showing that on an arbitrary state
|ψ⟩ = ∑
σ,σ′=±
cσσ′|σ⟩|σ′⟩
The operator
ρ = 1projects onto
4 (I − σ1 ⋅ σ2) |ψ0⟩ = 1
2 (| + ⟩| − ⟩ − | − ⟩| + ⟩)
Electron and positrons move away from the source and reach two detectors
operated by two physicists, Alice and Bob
The detectors measure the angular momentum along the directions
n1 = (0, sin α, cosα) n2 = (0, sin β, cosβ)
We ask ourselves what is the joint probability that Alice measures the spin of particle 1 up along the direction and Bob measures the spin of particle 2 down along the
n2 n1In quantum mechanics we can compute this probability We have
P+−(α, β) = |⟨ψ0| + ⟩n1| − ⟩n2|2This probability can be computed explicitly or by using the density operator
P+−(α, β) = |⟨ψ0| + ⟩n1| − ⟩n2|2 = ⟨ψ0| + ⟩n1| − ⟩n2⟨n1 + |⟨n2 − |ψ0⟩
where, by recalling the result in Eq. the projector
(2) 𝒫n1,+n2,−is
𝒫n1,+n2,− = I + n1 ⋅ σ1
2 ⊗ I − n2 ⋅ σ2 2
= ∑
k
⟨ψ0| + ⟩n1| − ⟩n2⟨n1 + |⟨n2 − |ek⟩⟨ek|ψ0⟩ = Tr (ρ𝒫n1,+n2,−)
More generally the four correlated probabilities read
P++(α, β) = P−−(α, β) = 14 (1 − n1 ⋅ n2) P−+(α, β) = P+−(α, β) = 14 (1 + n1 ⋅ n2)
In the case
α = βwe recover the previous result
P++ = P−− = 0and
P+− = P−+ = 1/2Then the required probability is
P+−(α, β) = Tr (ρPn1,+n2,−) = Tr ( 1
4 (I − σ1 ⋅ σ2) I + n1 ⋅ σ1 2
I − n2 ⋅ σ2
2 )
Since the Pauli matrices are traceless the only non trivial contribution comes from the term
(σ1 ⋅ σ2)(n1 ⋅ σ1)(n2 ⋅ σ2)and we get
P+−(α, β) = 1 4 1
4 × 4 + 1
4 × 1
4 × Tr ((σ1 ⋅ σ2)(n1 ⋅ σ1)(n2 ⋅ σ2))
= 14 + 14 × 1
4 × Tr ((σ1)i(σ2)i(n1)j(σ1)j(n2)k(σ2)k) = 14 (1 + n1 ⋅ n2) 2δij
2δik
We now introduce the correlation function defined as the probability of equal outcomes minus the probability of different outcomes
C(α, β) = ∑
σ=± (Pσσ − Pσ,−σ)
We find
C(α, β) = − n1 ⋅ n2
We now assume that the probabilistic nature of quantum mechanics is simply a result of our lack of knowledge of some hidden variable
λWe assume that the first and the second particle are characterised by
response functions for a measurement of the spin (in units of ) along the
directions and
n1 n2 ℏ/2σ1(α, λ) = ± 1 σ2(β, λ) = ± 1
To explain the observed anticorrelation we have to assume
σ2(α, λ) = − σ1(α, λ)This is what quantum mechanics predicts
The parameter is unknown and measurements correspond to averages over with
some probability density
λ ρ(λ)such that
λWe can now make predictions for the measurements
For example
P1(α, ± ) = ∫ dλρ(λ) 1 ± σ1(α, λ) 2
And this should be
1/2to agree with the quantum mechanical result Correlated probabilities can be computed analogously
Pσσ′ = ∫ dλρ(λ) 1 + σσ1(α, λ) 2
1 + σ′σ2(β, λ) 2
C(α, β) = P++ + P−− − P+− − P−+ = ∫ dλρ(λ)σ1(α, λ)σ2(β, λ)
The correlation coefficient becomes
∫ dλρ(λ) = 1
The model is completely defined by giving
ρ(λ), σ1(α, λ), σ2(β, λ)We can thus define the quantity
s(λ, α,α′, β, β′) = σ1(α,λ)[σ2(β, λ) − σ2(β′, λ)] + σ1(α′,λ)[σ2(β,λ) + σ2(β′,λ)]
Since and can only take the values we have
σ1 σ2 ±1s(λ, α, α′, β, β′) = ± 2
Therefore
−2 ≤ ∫ dλρ(λ)s(λ, α, α′, β, β′) ≤ 2
and introducing the correlations
C(α, β) = ∫ dλρ(λ)σ1(α, λ)σ2(β, λ)we find
|C(α, β) − C(α, β′) + C(α′, β) + C(α′, β′)| ≤ 2
This is the Clauser-Horne-Shimony-Holt (CHSH) inequality
According to local realism we can simultaneously measure the spin along different
directions , and , corresponding to angles , and ,
n1 n′1 n2 n′2 α α′ β β′It is easy to see that the quantum mechanical result
C(α, β) = − n1 ⋅ n2 = − cos(α − β)
violates such inequality
For example if we choose the angles as in the figure we obtain
π/8 C(α, β) − C(α, β′) + C(α′, β) + C(α′, β′)
= sin(π/8) − 3 cos(π/8) = − 2.38896
n1
n2
n′1 n′2
We must therefore conclude that quantum mechanics is indeed not compatible with local realism and with theories with hidden variables (Bell theorem)
Identities analogous to the CHSH identity had been derived earlier by Bell in 1964
For example, in the case
α′ = β′ ≡ γone can arrive to the alternative inequality
C(α, β) − 1 ≤ C(α, γ) + C(γ, β)