Summary of Lecture #1: fundamental concepts

Goals and Requirements:

• Reflect on what you have learned about the basic objects in quantum mechanics:
wavefunctions & operators.

• Establish the basic picture about the math structure of quantum mechanics
(NOTE: these are not mathematically rigorous)

Hilbert space H linear space equipped with an inner product

quantum states (‘ket’) |ψi elements in the linear space H
‘bra’ hψ| linear functionals defined on H: H 7→ C
quantum mechanical operators linear mappings: H1 7→ H2 .

• Be familiarized with the general description of a quantum state: the density matrix.
Get some taste of quantum entropy and quantum entanglement if time permits.

• By the end of this lecture, you should feel comfortable about

dealing with abstract quantum states without reference to the wavefunctions, and
dealing with abstract quantum operators without reference to matrices.

• NOTE: statements with ? are advanced topics/challenge questions/extra exercises

(NOT required).

References:
J.J. Sakurai, Modern Quantum Mechanics, Chapter 1.
P.A.M. Dirac, The Principle of Quantum Mechanics, Sections I.5-6, Chapter III.
J. von Neumann, Mathematical Foundations of Quantum Mechanics, Chapters I,II.

Advanced Quantum Mechanics, Fall 2018 1/15

 I THE HILBERT SPACE

I. THE HILBERT SPACE

A. The Wavefunction

• ψ(x) is a complex-valued function defined on some “coordinate” space V (x ∈ V ).

R
• Strictly speaking, ψ(x) shall be normalizable: |ψ(x)|2 dV < ∞.
2
• Probability of the system being in “volume element” dV is R|ψ(x)| dV . − Max Born.
|ψ(x)|2 dV

• Normalized ψ(x) has “dimension”(unit) of “volume”−1/2 (usually not dimensionless).

• Be careful when you parametrize the coordinate space. You might need to absorb the
Jacobian into the wavefunction (depending on the definition of dV ), and/or introduce
artificial “boundary conditions”.

• We may need non-normalizable wavefunctions when the space V is not compact

R
( dV = ∞), e.g. plane waves in open space. We usually regularize this problem
by first taking a finite V , and finally taking the limit that the volume goes to infinity.

• Most wavefunctions we will deal with are continuous, and (piecewise) smooth.

• Example:
V is the unit sphere S 2 parametrized by polar and
azimuth angles x = (θ, ϕ) ∈ [0, π] × [0, 2π], dV = sin θ dθdϕ,
legitimate ψ(θ, ϕ) are normalizable complex functions
R 2π R π
( ϕ=0 θ=0 |ψ(θ, ϕ)|2 sin θ dθdϕ < ∞)
with the “boundary condition” ψ(θ, 0) = ψ(θ, 2π), ∀θ
and ψ(0, ϕ) = ψ(0, 0), ψ(π, ϕ) = ψ(π, 0), ∀ϕ.
A basis of such wavefunctions are spherical harmonics Y`m (θ, φ).

B. The Hilbert Space

• The Hilbert space H(V ) defined on a coordinate space V is the complex linear space
formed by normalizable wavefunctions defined on V .

– Being a linear space: if ψ1 and ψ2 are elements of H (legitimate wavefunc.), then

so does λ1 ψ1 + λ2 ψ2 , for any complex numbers λ1 and λ2 . Exercise: prove this.

Advanced Quantum Mechanics, Fall 2018 2/15

 I THE HILBERT SPACE

• There is a natural inner product(“overlap”) of two wavefunctions φ and ψ,

(φ, ψ) = φ∗ (x) ψ(x) dV , satisfying the three important “axioms”
R

– Hermiticity: (φ, ψ) = (ψ, φ)∗ .

– Linearity: (φ, λ1 ψ1 + λ2 ψ2 ) = λ1 · (φ, ψ1 ) + λ2 · (φ, ψ2 ), for λ1 , λ2 ∈ C.

∗ Above two properties lead to, (λ1 φ1 + λ2 φ2 , ψ) = λ∗1 · (φ1 , ψ) + λ∗2 · (φ2 , ψ).

– Positive definiteness: (ψ, ψ) ≥ 0, and (ψ, ψ) = 0 if and only if “ψ = 0”.

λ∗i · (ψi , ψj ) · λj ≥ 0 ⇒
P P P
• One consequence: ( i λi ψi , j λj ψj ) = i,j

the Gram matrix (ψi , ψj ) (where i is row-index, j is column-index) is Hermitian and

positive semi-definite.

– The Gram determinant det[(ψi , ψj )] ≥ 0.

– n = 2 case is the Cauchy-Schwarz inequality, (ψ1 , ψ1 )(ψ2 , ψ2 ) ≥ |(ψ1 , ψ2 )|2 .

– The states ψi are linearly dependent if and only if the matrix (ψi , ψj ) is singular,
or equivalently det[(ψi , ψj )] = 0.

C. Combining Hilbert Spaces: Direct Sum & Tensor Product

• Direct sum of two Hilbert spaces 

H1 (V1 ) ⊕ H2 (V2 ):
 ψ (x), x ∈ V ;
1 1
the wavefunction (ψ1 ⊕ ψ2 )(x) =
 ψ2 (x), x ∈ V2 .
The inner product becomes (φ1 ⊕ φ2 , ψ1 ⊕ ψ2 )V1 ⊕V2 = (φ1 , ψ1 )V1 + (φ2 , ψ2 )V2 =
R ∗
φ1 ψ1 dV1 + φ∗2 ψ2 dV2 .
R

– Note: you need consistent definition of the “volumes” of V1,2 .

• Tensor product of two Hilbert spaces H1 (V1 ) ⊗ H2 (V2 ):

the wavefunction (ψ1 ⊗ ψ2 )(x1 , x2 ) = ψ1 (x1 ) · ψ2 (x2 ), for x1 ∈ V1 and x2 ∈ V2 .
The inner product becomes (φ1 ⊗ φ2 , ψ1 ⊗ ψ2 )V1 ⊗V2 = (φ1 , ψ1 )V1 · (φ2 , ψ2 )V2 .

– Entanglement: wavefunctions in H1 ⊗ H2 may not be a direct product ψ1 ⊗ ψ2 .

– Identical particles: will be treated later.

Advanced Quantum Mechanics, Fall 2018 3/15

 I THE HILBERT SPACE

• Example:
direct sum: vs. tensor product:
one particle in two potential wells two inequivalent particles in two wells
need to know if it is in left or right need to know both left & right particles’ state

H1 ⊕ H 2 H1 ⊗ H2

D. The Dirac Notation

• ‘kets’ |ψi: element(quantum state) in Hilbert space corresponding to wavefunction ψ.

• ‘bras’ hψ|: linear functional defined on the Hilbert space: H 7→ C, φ 7→ (ψ, φ).

ψ ∗ φ dV .
R
– Short-hand notation: hψ|φi ≡ (ψ, φ) =

– ‘bra’ is a linear functional: hψ|λ1 φ1 + λ2 φ2 i = λ1 hψ|φ1 i + λ2 hψ|φ2 i.

– ‘bras’ form an anti-linear space: λ∗1 hψ1 | + λ∗2 hψ2 | = hλ1 ψ1 + λ2 ψ2 |.

– Any ‘continuous’ linear functional f : H 7→ C, φ 7→ f (φ), corresponds to a

wavefunction ψf so that f = hψf |, f (φ) = (ψf , φ). − Riesz-Fréchet theorem.
P
With complete orthonormal basis |ei i, hψf | = i f (ei ) hei |,
so |ψf i = i f (ei )∗ |ei i.
P

• Short-hand notation: |ψ1 i + |ψ2 i with ψ1 ∈ H1 , ψ2 ∈ H2 means the direct sum state
ψ1 ⊕ ψ2 ∈ H1 ⊕ H2 .

• Short-hand notation: |ψ1 i|ψ2 i with ψ1 ∈ H1 , ψ2 ∈ H2 means the tensor product state
ψ1 ⊗ ψ2 ∈ H1 ⊗ H2 .

• Short-hand notation: |ψ1 ihψ2 | with ψ1 ∈ H1 , ψ2 ∈ H2 is a linear operator: H2 7→

H1 , φ 7→ (ψ2 , φ)ψ1 ≡ |ψ1 ihψ2 |φi.

• Other labels like quantum numbers or just an index, are often used in ‘bras’ & ‘kets’:
e.g. |L = 2, Lz = 0i, |0i.

Advanced Quantum Mechanics, Fall 2018 4/15

 I THE HILBERT SPACE

E. Complete Orthonormal Basis

• (Discrete) Orthonormal basis ei (i = 0, 1, . . . ) satisfy hei |ej i = δij .

P
• Complete orthonormal basis: for any ψ ∈ H, |ψi = i |ei ihei |ψi.

– For finite dimensional Hilbert space, this is just

‘number of orthonormal basis’=‘dimension of Hilbert space’.

– For infinite dimensional Hilbert space, completeness is usually very hard to prove.

• Resolution of identity: 1 =
P
i |ei ihei |, the sum is over a complete orthonormal basis.

– We will see the resolution of identity in terms of overcomplete basis later.

P P P
– Application: change of basis, for |ψi = j c̃j |ẽj i = i ci |ei i = i,j ci |ẽj ihẽj |ei i,
P
coefficients c̃j = i ci hẽj |ei i, where ẽj are another complete orthonormal basis.

∗ The matrix Uji = hẽj |ei i is a unitary matrix. (U · U † )jk = i Uji (U † )ik =
P

i hẽj |ei ihei |ẽk i = hẽj |ẽk i (by resolution of identity) = δjk = (1)jk . Above
P
P
relation is c̃j = i Uji ci or c̃ = U · c in short form.
∗ Conversely, given a unitary matrix U and complete orthonormal basis |ei i,
P
then |ẽj i ≡ i Uji |ei i form a new set of complete orthonormal basis.
Exercise: is the “U ” here same as the “U ” in previous item?

• Example: Fourier series.

Particle moving on a ring parametrized by angle θ, legitimate
wavefunctions are normalizable ψ(θ) with period 2π, ψ(θ + 2π) = ψ(θ).
A complete orthonormal basis is en (θ) = √1
2π
exp(inθ) for n ∈ Z.

• Basis of composite Hilbert space:

if |ei i (i = 1, . . . , n) are the basis of H1 , and |e0j i (j = 1, . . . , m) are the basis of H2 ,

– the (n + m) basis of H1 ⊕ H2 can be chosen as

(|e1 i, |e2 i, . . . , |en i, |e01 i, |e02 i, . . . , |e0m i).

– the (n × m) basis of H1 ⊗ H2 can be chosen as |ei i ⊗ |e0j i, namely

(|e1 i|e01 i, |e1 i|e02 i, . . . , |e1 i|e0m i, |e2 i|e01 i, . . . , |e2 i|e0m i, · · · , |en i|e01 i, . . . , |en i|e0m i).

Advanced Quantum Mechanics, Fall 2018 5/15

 II QUANTUM MECHANICAL OPERATORS

II. QUANTUM MECHANICAL OPERATORS

A. Quantum Mechanical Operators

• Linear operators: linear mappings between two (often the same) Hilbert spaces:
Ô|ψi ∈ H2 for |ψi ∈ H1 , and Ô|λ1 ψ1 + λ2 ψ2 i = λ1 Ô|ψ1 i + λ2 Ô|ψ2 i.

• Anti-linear operators: replace the last condition of linear operators by

Ô|λ1 ψ1 + λ2 ψ2 i = λ∗1 Ô|ψ1 i + λ∗2 Ô|ψ2 i.

– Example: the operator of “taking complex conjugate” K: φ(x) 7→ φ(x)∗ .

• Hermitian conjugate (adjoint) of linear operators: Ô† is a linear operator satisfying

(Ô† ψ, φ) = (ψ, Ôφ) for any ψ & φ, or hÔ† ψ| = hψ|Ô for any ψ.

– (Ô† )† = Ô.
“Proof”: for any ψ & φ, by definitions of inner product and hermitian conjugate,
(ψ, ((Ô† )† )φ) = (((Ô† )† )φ, ψ)∗ = (φ, (Ô† )ψ)∗ = ((Ô† )ψ, φ) = (ψ, Ôφ). Then
((Ô† )† ) and Ô must be the same.

– (λÔ)† = λ∗ Ô† , (ÂB̂)† = B̂ † † .

Exercise: try to “prove” these as the “proof” for the previous relation.

– Hermitian operators: those satisfy Ô† = Ô.

Anti-Hermitian operators: Ô† = −Ô.

– Any operator is the sum of its Hermitian & anti-Hermitian part:

Ô+Ô† Ô−Ô†
Ô = 2
+ 2
, the 1st term is Hermitian, 2nd term is anti-Hermitian.

• Matrix representation: under a complete orthonormal basis |ni, the operator Ô has
‘matrix elements’ Omn ≡ hm|Ô|ni. (O† )mn = (Onm )∗ .

– Matrix represention under non-orthogonal/overcomplete basis can also be useful.

– Expectation value of Ô in state ψ: hψ|Ô|ψi/hψ|ψi.

P
• Trace: TrÔ = n hn|Ô|ni, summing over a complete orthonormal basis.
The result of trace is independent of the choice of basis.
Cyclic property: Tr(ÂB̂) = Tr(B̂ Â), for ‘finite’ operators Â, B̂ (e.g. finite dimensional)

Advanced Quantum Mechanics, Fall 2018 6/15

 II QUANTUM MECHANICAL OPERATORS

• Eigenvalue λ and eigenstate |Ô = λi of operator Ô: defined by Ô|Ô = λi = λ|Ô = λi.

– Eigenvalues of Hermitian operators are real.

• (Not required) Singular value decomposition (SVD):

P
– Any operator can be written as Ô = n |ñi ρn hn|, where n labels the singular
value ρn ≥ 0, and the two sets of orthonormal basis |ni & |ñi are eigenstates of
Ô† Ô & ÔÔ† respectively.

– In complete orthonormal basis |ei i, the above relations becomes

Oij = hei |Ô|ej i = n hei |ñi ρn hn|ej i = (U · ρ · V † )ij ,
P

where the unitary matrices Uin = hei |ñi and Vjn = hej |ni,
and the diagonal matrix ρ has diagonal elements ρn . And
(O† O)ij = hei |Ô† Ô|ej i = n hei |ni ρ2n hn|ej i = (V · ρ2 · V † )ij , and
P

(OO† )ij = hei |ÔÔ† |ej i = n hei |ñi ρ2n hñ|ej i = (U · ρ2 · U † )ij .
P

• Projection operators: operators P̂ : H 7→ H, satisfying P̂ P̂ = P̂ .

P
– Hermitian projection operators P̂ = i |ei ihei |, ei are a set of orthonormal basis,
and P̂ have eigenvalues 1 and 0 only.

– 1 − P̂ is also a projection operator. Exercise: check that (1 − P̂ )(1 − P̂ ) = (1 − P̂ ).

• Inverse of an operator: Â−1 must satisfy Â−1 ·  = 1 and  · Â−1 = 1

– ? In infinite dimensional Hilbert space, there can be cases with B̂ Â = 1 while

ÂB̂ 6= 1, B̂ shall not be called Â−1 .

• Unitary operators: linear operators with

(Û ψ, Û φ) = (ψ, φ), ∀ψ, φ. Or equivalently Û † Û = 1.
P
– Unitary operators are of the form i |ẽi ihei |, where ei is a set of complete or-
thonormal basis, ẽi is another set of orthonormal basis.

– If Ĥ is Hermitian, then exp(iĤ) is unitary. Exercise: is the converse true?

• Anti-unitary operators: anti-linear operators with

(Û ψ, Û φ) = (ψ, φ)∗ , ∀ψ, φ.

Advanced Quantum Mechanics, Fall 2018 7/15

 II QUANTUM MECHANICAL OPERATORS

B. Abstract Calculations with Operators

• Commutator & anti-commutator of  & B̂: [Â, B̂] ≡ ÂB̂ − B̂ Â, {Â, B̂} ≡ ÂB̂ + B̂ Â.

– For notation simplicity, define ‘Lie derivative’ LÂ B̂ ≡ [Â, B̂].

• Elementary functions of operators may be defined by their power series expansion, e.g.
exp(Â) = ∞ n
P
n=0 (Â) /n! (let’s not worry about convergence).

– Note: Â · f (B̂) · Â−1 = f (Â · B̂ · Â−1 ) for such functions f that can be defined as
power series, because  · (B̂)n · Â−1 = ( · B̂ · Â−1 )n .

• Jacobi identity: [Â, [B̂, Ĉ]] + [B̂, [Ĉ, Â]] + [Ĉ, [Â, B̂]] = 0. Or [LÂ , LB̂ ]Ĉ = L[Â,B̂] Ĉ.

• ‘Leibniz’s rule’:
[Â, B̂1 B̂2 · · · B̂n ] = [Â, B̂1 ]B̂2 · · · B̂n + B̂1 [Â, B̂2 ] · · · B̂n + · · · + B̂1 B̂2 · · · [Â, B̂n ]. Or
LÂ (B̂1 B̂2 · · · B̂n ) = (LÂ B̂1 )B̂2 · · · B̂n + B̂1 (LÂ B̂2 ) · · · B̂n + · · · + B̂1 B̂2 · · · (LÂ B̂n ).

• Baker-Hausdorff formula: e B̂e− = B̂ + [Â, B̂]/1! + [Â, [Â, B̂]]/2! + . . .

Or formally e B̂e− = exp(L )B̂.

– A heuristic “proof”:
define fˆ(t) = et B̂e−t , then fˆ(0) = B̂.
d tÂ
Take derivative with respect to t, note that dt e = Âet = et Â,
then d fˆ(t) = Â · fˆ(t) − fˆ(t) · Â = [Â, fˆ(t)] = L fˆ(t).
dt Â

The formal solution of this ordinary differential equation is then fˆ(t) = etLÂ fˆ(0),
so e B̂e− = fˆ(1) = exp(L )B̂. Â

• Direct sum & tensor product of operators are defined similarly to wavefunctions:
for operator  defined on H1 , and B̂ defined on H2 ,  ⊗ B̂ is an operator defined on
H1 ⊗ H2 , such that (Â ⊗ B̂)|ψ ⊗ φi = (Â|ψi) ⊗ (B̂|φi), for states ψ ∈ H1 and φ ∈ H2 .

– (Â ⊗ B̂) · (Ĉ ⊗ D̂) = ÂĈ ⊗ B̂ D̂.

– When  is referred to within Ĥ1 ⊗ Ĥ2 , it usually means  ⊗ 1. With this

convention, Â⊗ B̂ is usually written as ÂB̂, which means (Â⊗ 1)·(1 ⊗ B̂) = Â⊗ B̂.

– Tr1⊗2 (Â ⊗ B̂) = Tr1 (Â) · Tr2 (B̂), where the three different traces are taken in
Hilbert spaces H1 ⊗ H2 , H1 , H2 , respectively.

Advanced Quantum Mechanics, Fall 2018 8/15

 II QUANTUM MECHANICAL OPERATORS

– With complete orthonormal basis, |ei i ∈ H1 (i = 1, . . . , n) and |e0j i ∈ H2 (j =

1, . . . , m), the matrix representation of  ⊗ B̂ is a (n × m)-row (n × m)-column
matrix, (Â ⊗ B̂)(i,j)(i0 ,j 0 ) ≡ hei |he0j |Â ⊗ B̂|ei0 i|e0j 0 i = hei |Â|ei0 i · he0j |B̂|e0j 0 i = Aii0 Bjj 0 .
The combination (i, j)[(i0 , j 0 )] is the row[column] index (i, i0 = 1, . . . , n and j, j 0 =
1, . . . , m).

C. Back to Wavefunction

• The coordinate operator x̂: φ(x) 7→ x · φ(x). It is obviously Hermitian.

– Worry #1: x · φ(x) may not be normalizable!

– Worry #2: What are the eigenstate wavefunctions of x̂? Are they normalizable?

• Despite the above worries, denote the eigenstates of x̂ by |xi, i.e. x̂|xi = x|xi.

– ‘Normalization’: hx0 |xi = δ(x0 − x), where δ is the Dirac-δ ‘function’.

– Resolution of identity: 1 = |xihx| dx.

R

– The wavefunction ψ(x): expansion coefficients of state ψ in the basis |xi.

R
ψ(x) = hx|ψi. And |ψi = ψ(x)|xi dx.

• The momentum operator p̂: φ(x) 7→ −ih̄ ∂x

∂
φ(x). It is not-so-obviously Hermitian.

– Canonical commutation relation [x̂, p̂] = ih̄.

– Similar worries as for the coordinate operator.

– Nonetheless, denote the eigenstate of p̂ as |pi, p̂|pi = p|pi.

– ‘Normalization’: hp0 |pi = δ(p0 − p).

– Resolution of identity: 1 = |pihp| dp.

R

ipx/h̄
e√
– hx|pi = 2πh̄
for 1D infinite space.

• ? Construct examples of normalizable φ(x) so that x̂ φ(x) or p̂ φ(x) is not normalizable.

• IMPORTANT: h̄ will be frequently omitted hereafter.

Advanced Quantum Mechanics, Fall 2018 9/15

 III DENSITY MATRIX & ENTANGLEMENT

III. DENSITY MATRIX & ENTANGLEMENT

A. Density Matrix

• Density matrix of a normalized ‘pure state’ ψ: ρ̂ψ = |ψihψ| is a projection operator.

– Expectation value of Ô in ψ is hψ|Ô|ψi = Tr(ρ̂ψ Ô) = Tr(Ô ρ̂ψ ).

– ρ̂ is independent of the complex phase of |ψi, is a ‘better’ description of the state.

• Generic density matrix ρ: linear Hermitian non-negative operator of trace unity.

ρ̂† = ρ̂; hφ|ρ̂|φi ≥ 0, ∀φ; and Tr(ρ̂) = 1.

P P
– ρ= i λi |ei ihei |, with some orthonormal basis ei , and λi > 0, i λi = 1.

– Expectation value of Ô in generic ‘mixed state’ is Tr(ρ̂ Ô).

P
– If ρ̂j are density matrices, and cj > 0, and j cj = 1,
P
then j cj ρ̂j is also a density matrix.

• The density matrix of Hamiltonian H

b at finite temperature T :
b B T )/Z = P exp(−Ei /kB T ) |Ei ihEi |,
ρ = exp(−H/k Ei Z

where Ei are eigenvalues, |Ei i are corresponding eigenstates,

b B T )] = P exp(−Ei /kB T ).
Z = Tr[exp(−H/k Ei

B. ? Some Quantum Information Basics (not required)

P
• von Neumann entropy of a density matrix ρ̂: S ≡ −Tr(ρ̂ ln ρ̂) = − i λi ln λi .

– Pure states have zero entropy & ρ̂2 = ρ̂.

Mixed states have positive entropy & ρ̂2 < ρ̂.

– In n(finite)-dimensional Hilbert space, 0 ≤ S[ρ̂] ≤ ln(n).

ln[Tr(ρ̂n )]
– Rényi entropy: Sn ≡ 1−n
. Note: formally limn→1 Sn = S.

• ? ? ? Concavity of von Neumann entropy: mixing two systems increases entropy.

S[λ ρ̂1 + (1 − λ) ρ̂2 ] ≥ λ S[ρ̂1 ] + (1 − λ) S[ρ̂2 ],
for two density matrices ρ̂1,2 and 0 < λ < 1.

Advanced Quantum Mechanics, Fall 2018 10/15

 III DENSITY MATRIX & ENTANGLEMENT

• Reduced density matrix: given a density matrix ρ̂ on Ha ⊗ Hb , reduced density matrix

ρa on Ha is ρ̂a = Trb (ρ̂), obtained by taking partial trace over Hb .

P
– Meaning of partial trace: for any ψ1,2 ∈ Ha , hψ1 |ρ̂a |ψ2 i = i hψ1 ⊗ φi |ρ̂|ψ2 ⊗ φi i,
and the sum is over a complete orthonormal basis φi of Hb . The matrix elements
of ρ̂a under a orthonormal basis Ha can be computed by this relation.

– ? ? ? Subadditivity of entropy: information is ‘lost’ by separating two subsystems.

Sa⊗b [ρ̂] = Tra⊗b (−ρ̂ ln ρ̂) ≤ Sa [ρ̂a ] + Sb [ρ̂b ] = Tra (−ρ̂a ln ρ̂a ) + Trb (−ρ̂b ln ρ̂b ) =
Sa⊗b [ρ̂a ⊗ ρ̂b ].

• Special case: ρ̂ = |ψihψ| is for a normalized pure state ψ ∈ Ha ⊗ Hb .

– “Entanglement entropy”: von Neumann entropy of reduced density matrix:

Sa = −Tra (ρ̂a ln ρ̂a ), where ρ̂a = Trb (ρ̂).

– The degrees of freedom of a and b are ‘entangled’ in this state ψ if Sa > 0.

– The state ψ is a disentangled product state ψa ⊗ φb if and only if Sa = 0.

P
– Schmidt decomposition (just SVD): |ψi = i λi |φi i ⊗ |ϕi i,
where (λi )2 are eigenvalues of ρ̂a , φi (ϕi ) are orthonormal eigenstates of ρ̂a (ρ̂b ).

– ? Reciprocity: define reduced density matrix ρ̂b = Tra (ρ̂) on Hb .

Sb = Trb (−ρ̂b ln ρ̂b ) equals to Sa above for a pure state in Ha ⊗ Hb .

• Example: Bell state.

Ha (Hb ) is 2-dimensional, with orthonormal basis |0i, |1i (|0̃i, |1̃i).
One of the Bell states is √1 (|0i|1̃i − |1i|0̃i).
2
Exercise: write down the reduced density matrices ρ̂a and ρ̂b . Compute the entangle-
ment entropy.

Advanced Quantum Mechanics, Fall 2018 11/15

 IV MEASUREMENT & THE UNCERTAINTY PRINCIPLE

IV. MEASUREMENT & THE UNCERTAINTY PRINCIPLE

A. Measurement

• Measurement can be done for a Hermitian operator  on pure or mixed states ρ̂.

• The outcome of the measurement will be eigenvalues of Â.

• The probability of outcome λ is Pλ = Tr(P̂λ ρ̂), P̂λ is the projection to eigenvalue-λ

P
subspace. P̂λ = | = λih = λ|, summing over orthonormal eigenstates of  with
eigenvalue λ.

– If all eigenvalues λ0 of  are known, then P̂λ can be formally obtained by the
Â−λ0 1
Q  
“Lagrange interpolating polynomial”, P̂λ = λ0 , λ0 6=λ λ−λ0 .

– The statistical average of outcome is the expectation value of  in state ρ̂,

P P P
Tr(Â ρ̂) = Tr[( λ λ P̂λ ) ρ̂] = λ λ Tr(P̂λ ρ̂) = λ λ Pλ .
For pure state ρ̂ = |ψihψ|, this is hψ|Â|ψi.

• The “collapse” postulate:

P̂λ ρ̂ P̂λ
if the measurement outcome is λ, the quantum state will “collapse” to Tr(P̂λ ρ̂)
.

– If eigenvalue-λ eigenstate is unique, this is the familiar statement

that the system collapses to eigenstate |Â = λi.

• ? ? ? Information is gained by measurement: ‘entropy’ decreases.

P
S[ρ̂] ≥ λ Pλ S[ρ̂λ ], Pλ is the probability of outcome λ, ρ̂λ is the collapsed state.

• An example:

– DefinePauli
 matrices
     
1 0 0 1 0 −i 1 0
σ0 =  , σ1 =  , σ2 =  , σ 3 =  .
0 1 1 0 i 0 0 −1
– Consider a state described by the density matrix ρ̂, represented in some basis as
0 0 i
 
1
1 −i 0
 
0
ρ̂ = 41   = 4 [1 − σ1 ⊗ σ2 ]. Exercise: is this a pure state?
  1
0
 i 1 0 
−i 0 0 1

Advanced Quantum Mechanics, Fall 2018 12/15

 IV MEASUREMENT & THE UNCERTAINTY PRINCIPLE

– Measure a Hermitian operator Â, represented in the same basis as  = σ1 ⊗ σ3 .

– Eigenvalues of  are ±1.

(−1)−Â 1−Â .
Corresponding projection operators are P̂+1 = (−1)−1
and P̂−1 = 1−(−1)

– Outcome +1: probability Tr(ρ̂ P̂+1 ) = 1/2,

collapsed state is ρ̂+1 = (1/4)[1 + σ1 ⊗ σ3 ].
Outcome −1: probability Tr(ρ̂ P̂−1 ) = 1/2,
collapsed state is ρ̂−1 = (1/4)[1 − σ1 ⊗ σ3 ].

– ? Exercise: compute entropies S[ρ̂], S[ρ̂+1 ] and S[ρ̂−1 ], check if any informa-
tion can be gained by this measurement, namely whether S[ρ̂] > (1/2)S[ρ̂+1 ] +
(1/2)S[ρ̂−1 ] ?

B. The Uncertainty Principle

• For Hermitian  & B̂, (hÂ2 i − hÂi2 )(hB̂ 2 i − hB̂i2 ) ≥ 41 |h[Â, B̂]i|2 . − W. Heisenberg
h·i is the expectation value under a quantum state ρ̂.

– Rough description: product of variances of measurement outcomes for  & B̂ is

bounded below by the square of their commutator’s expectation value.

– Variances of measurement outcome: λ Pλ (λ − λ̄)2 = ( λ Pλ λ2 ) − λ̄2 , where λ

P P
P
are eigenvalues, Pλ is the probability of outcome λ, λ̄ = λ Pλ λ is the ‘average’.

– Proof:
Define the inner product of two operators Â, B̂ as (Â, B̂) = h† B̂i = Tr(† B̂ ρ̂).
Exercise: check that this indeed satisfies the “axioms” of inner product.
Define two new operators Â0 = Â − hÂi, B̂ 0 = B̂ − hB̂i. For Hermitian Â, B̂,
1
4
|h[Â, B̂]i|2 = 41 |h[Â0 , B̂ 0 ]i|2 = 12 (Â0 , B̂ 0 )(B̂ 0 , Â0 ) − 14 (Â0 , B̂ 0 )2 − 14 (B̂ 0 , Â0 )2
= [Im(Â0 , B̂ 0 )]2 ≤ |(Â0 , B̂ 0 )|2 ≤ (Â0 , Â0 )(B̂ 0 , B̂ 0 ) = (hÂ2 i − |hÂi|2 )(hB̂ 2 i − |hB̂i|2 ).
The last inequality used here is Cauchy-Schwarz.

– Exercise: what is the condition for the equality to be true?

• Familiar case: Â = x̂, B̂ = p̂. hx̂2 − x̄2 ihp̂2 − p̄2 i ≥ h̄2 /4.

Advanced Quantum Mechanics, Fall 2018 13/15

 A ? ABOUT THE STATEMENTS WITH ? (NOT REQUIRED)

Appendix A: ? About the Statements with ? (not required)

• ? In infinite dimensional Hilbert space, there can be cases with B̂ Â = 1 while ÂB̂ 6= 1,
B̂ shall not be called Â−1 .

– Example:
Assume |0i, |1i, . . . are complete orthonormal basis.
Define  = ∞
P
n=0 |n + 1ihn| = |1ih0| + |2ih1| + . . . .

Consider B̂ = ∞ n=0 |nihn| = 1.

P P∞
n=0 |nihn + 1|, then B̂ · Â =

However  · B̂ = n=0 |n + 1ihn + 1| = 1 − |n = 0ihn = 0| 6= 1.

P∞

• ? Construct examples of normalizable φ(x) so that x̂ φ(x) or p̂ φ(x) is not normalizable.

– Example:
sin(x3 )
φ(x) = x
, defined on real axis of x.
Check that φ is normalizable while both x̂ φ and p̂ φ are not.

• ? ? ? Concavity of von Neumann entropy: S[λ ρ̂1 + (1 − λ) ρ̂2 ] ≥ λ S[ρ̂1 ] + (1 − λ) S[ρ̂2 ],

for two density matrices ρ̂1,2 and 0 < λ < 1.

– Proof: see e.g., M.A. Nielsen, I.L. Chuang, Quantum Computation and Quantum
Information, section 11.3.5.

• ? ? ? Subadditivity of entropy: Sa⊕b [ρ̂] = Tra⊕b (−ρ̂ ln ρ̂) ≤ Sa [ρ̂a ] + Sb [ρ̂b ] =

Tra (−ρ̂a ln ρ̂a ) + Trb (−ρ̂b ln ρ̂b ) = Sa⊕b [ρ̂a ⊗ ρ̂b ].

– Proof: see H. Araki, E. H. Lieb, Commun. Math. Phys. 18, 160 (1970).

• ? Reciprocity: define reduced density matrices ρ̂a = Trb (ρ̂) & ρ̂b = Tra (ρ̂) on subspace
Ha & Hb respectively, where ρ = |ψihψ| is a pure state on Ha ⊗ Hb . Then Sb =
Trb (−ρ̂b ln ρ̂b ) = Sa = Tra (−ρ̂a ln ρ̂a ).

– Proof:
This a simple consequence of the Schmidt decomposition of a pure state.
P
|ψi = i λi |ei i⊗|ẽi i, with orthonormal basis ei for Ha and ẽi for Hb , and real pos-
itive singular values λi . Then the reduced density matrix on Ha is i λ2i |ei ihei |,
P

and on Hb is i λ2i |ẽi ihẽi |. So Sa = − i λ2i ln(λ2i ) = Sb .

P P

Advanced Quantum Mechanics, Fall 2018 14/15

 A ? ABOUT THE STATEMENTS WITH ? (NOT REQUIRED)

• ? ? ? Information is gained by measurement: ‘entropy’ decreases.

P
S[ρ̂] ≥ λ Pλ S[ρ̂λ ], Pλ is the probability of outcome λ, ρ̂λ is the collapsed state.

– Proof: see G. Lindblad, Commun. Math. Phys. 28, 245 (1972).

Advanced Quantum Mechanics, Fall 2018 15/15