A state ρ, is measured in terms of a set of kets, {|ψν ⟩}.

The expectation value of for

a projective measurement is: sν = ⟨ψν |ρ̂|ψν ⟩. Experimentally, given detector efficiencies,
exposure times, etc. we will be able to measure

nν = N sν = N ⟨ψν |ρ̂|ψν ⟩ . (1)

An n-partite qubit state can be completely characterized by 4n real values. Thus for
bipartite qubit system we require 42 = 16 real values. A set of these values may be found
using 16 linearly independent, 4x4 matrices, with the properties:
Tr(Γ̂ν Γ̂µ ) = δν,µ (2)
∑
16
 = Γ̂µ Tr(Γ̂µ Â)∀ (3)
µ=1

I.e. the matrices are orthonormal and conveniently span the space.
Using the second property (Eq 3), we may write
∑
16
ρ̂ = Γ̂µ rµ , (4)
µ=1

letting rµ ≡ Tr(Γ̂µ ρ̂). These 16 rµ represent the real values needed to characterize ρ̂. Sub-
stituting the above expression for ρ̂ back into Equation 1,
nν = N ⟨ψν |ρ̂|ψν ⟩ (5)
∑
16
= N ⟨ψν |( Γ̂µ rµ )|ψν ⟩ (6)
µ=1

∑
16
=N rµ ⟨ψν |Γ̂µ |ψν ⟩ (7)
µ=1

Let nν and rµ be 16 × 1 column vectors and allow B̂ to be a matrix with elements

B̂ν,µ = ⟨ψν |Γ̂µ |ψν ⟩. Following rules of matrix multiplication, we can rewrite Equation 7 as

⃗n = N B̂⃗r (8)

And, so long as B̂ is non-singular, we can recover ⃗r,

1 −1
⃗r = B̂ ⃗n (9)
N
Equivalently,

1 ∑ −1
16
rµ = (B̂ )µ,ν nν (10)
N ν=1
∑
16
= (B̂ −1 )µ,ν sν (11)
ν=1

1
 Recalling Eq. 4

∑
16
ρ̂ = Γ̂µ rµ (12)
µ=1

∑
16 ∑
16
= Γ̂µ ( (B̂ −1 )µ,ν sν ) (13)
µ=1 ν=1

∑
16
= Γ̂µ (B̂ −1 )µ,ν sν (14)
µ,ν=1

∑
16 ∑
16
= sν Γ̂µ (B̂ −1 )µ,ν (15)
ν=1 µ=1

∑
16
= sν M̂ν (16)
ν=1

∑ −1
for M̂ν ≡ 16µ=1 Γ̂µ (B̂ )µ,ν .
We are nearly finished: we have described ρ̂ in terms of a set of matrices, {Mν }, and
expectation values, sν . The {Mν } are obtained from the pre-defined {Γν } and our choice of
projective measurements, |ψν ⟩ ⟨ψν |. The sν are the expectation values of the aforementioned
projective measurements. Experimentally, we have access to nν , not sν . We’ll finish by
noting that

∑
16
N ρ̂ = nν M̂ν (17)
ν=1

and Tr(N ρ̂) = N Tr(ρ̂) = N . If we let ρ̂′ ≡ N ρ̂, then

ρ̂′ ρ̂′
ρ̂ = = . (18)
N Tr(ρ̂′ )

As ρ̂′ can be obtained from the measured nν , we are finished.

2
Summary
Experimentally:
nν = N sν = N ⟨ψν |ρ̂|ψν ⟩ . (19)
Find a set of matrices with the properties

Tr(Γ̂ν Γ̂µ ) = δν,µ (20)

∑
16
 = Γ̂µ Tr(Γ̂µ Â)∀ (21)
µ=1

We can write

ρ̂′ ρ̂′
ρ̂ = = . (22)
N Tr(ρ̂′ )

where
∑
16
ρ̂′ = N ρ̂ = nν M̂ν (23)
ν=1

∑ −1
for M̂ν ≡ 16µ=1 Γ̂µ (B̂ )µ,ν .
And B̂ν,µ = ⟨ψν |Γ̂µ |ψν ⟩

3
Probably don’t bother:
The set of {M̂ν } have a number of interesting properties. Among them it can be shown:
∑
Tr(M̂ν ) |ψν ⟩ ⟨ψν | = I.
ˆ (24)
ν

Now, we can also claim

∑
Tr(M̂ν ) |ψν ⟩ ⟨ψν | ρ̂ = ρ̂ (25)
ν
(∑ )
Tr Tr(M̂ν ) |ψν ⟩ ⟨ψν | ρ̂ = Tr(ρ̂) = 1 (26)
ν
∑ ( )
Tr(M̂ν ) Tr |ψν ⟩ ⟨ψν | ρ̂ = 1 (27)
ν
∑
Tr(M̂ν ) ⟨ψν |ρ̂|ψν ⟩ = 1 (28)
ν
∑
Tr(M̂ν )sν = 1 (29)
ν
∑
Tr(M̂ν )nν = N. (30)
ν

Another useful property of the {M̂ν }, is that Tr(Mν ) = 1 for ν = 1, 2, 3, 4 and Tr(Mν ) = 0
otherwise. Now, we can fin