Mathematical Foundations of Quantum Information

John Watrous Department of Computer Science University of Calgary

Overview
So far, we have been using a simple mathematical framework for discussing quantum information: quantum state evolution measurement unit vector in a Hilbert space unitary operators projections

In many situations that arise when studying quantum information, this framework is either inconvenient or inadequate

Overview
We extend this formalism by considering a different way of representing quantum states:

quantum state

a matrix (or operator) acting on a Hilbert space

This extension has various advantages over the simpler formalism (in many situations) as we will see

Ket vectors
Suppose we have a superposition on n qubits: known as a ket

y = a x x
x =0

2n -1

a0 a1 M a n 2 -1

Let

be a space corresponding to n qubits

y is a unit vector in H.
Terminology: y is a pure state.

Bra Vectors
The corresponding bra:

y = a x x
x =0

2n -1

( a

a1 L

a 2n -1 )

The names come from the fact that a bra plus a ket form a bracket: if

j = bx x
x =0

2n -1

2n -1

then

y j = a x b x
x =0

Density Matrices
The density matrix corresponding to y is:

y y

a 0a 0 a 0 a1 aa a1a1 1 0 M M a n a a n a 2 -1 1 2 -1 0

L L O L

a 0 a 2n -1 a1a 2n -1 M a 2n -1a 2n -1
2 x

Tr ( y y

) = a a
x x

a
x

=1

Density Matrices
Now suppose we have a collection of pure states:

{y

, y 2 ,K , y k

and we imagine randomly choosing a state; choose yj with probability pj for each j=1,,k.
k

p
j =1

yj

doesnt make sense the pj values are probabilities not amplitudes.

Density Matrices
Now suppose we have a collection of pure states:

{y

, y 2 ,K , y k

and we imagine randomly choosing a state; choose yj with probability pj for each j=1,,k. For density matrices it works: this is called a mixture (or ensemble)

p
j =1

yj yj

Examples

1 1 + = 0 + 1 2 2
+ +

1 1 2 1 1 1 1 2 1 1 1 1 2 -1 1 1 - 1 2 - 1 1

1 1 - = 0 1 2 2
- -

Examples
Suppose we randomly choose one of + and each with probability 1/2. Resulting density matrix:

1 1 + + + - 2 2
Same thing with states 0 and 1 :

1 1 0 2 0 1
equal ?

1 1 0 0 + 1 1 2 2

1 1 0 2 0 1

Mixtures vs. density matrices

It is not an accident that different mixtures can give the same density matrix two mixtures can be distinguished if and only if they yield different density matrices. Density matrices describe mixed states. For instance,

1 1 0 0 + 1 1 2 2

describes a mixed state. It is equal to

1 1 + + + - 2 2

Facts about density matrices

Every density matrix has trace equal to 1:

k k Tr p j y j y j = p j Tr y j y j j =1 j =1

)= p
j =1

=1

Every density matrix is positive semidefinite (Hermitian, with all eigenvalues nonnegative.) Implies that every density matrix r comes from a mixture of orthogonal pure states:

r = qj j j j j
j =1

where

{j

, K , jm

} is an orthonormal set.

Quantum Transformations
The class of physically realizable transformations is easily characterized:

T :r a
provided

A rA
j j =1

A A
j j =1

=I .

Equivalently, T is completely positive and trace preserving.

Measurements
Any collection

{E1, K , Ek } of matrices satisfying

E
j =1

Ej = I

defines a measurement. If r is measured, the outcome j results with probability

p j = Tr ( j r E j ) E

and the state becomes

1 Ejr Ej pj

corrected (incorrect during talk)

Relation to simpler model

Note: from an algorithmic point of view, there is nothing to be gained from these more general transformations and measurements can simulate general transformations and measurements with unitary gates and projective measurements.

Fidelity and Trace-Distance

Natural notions of closeness between mixed states exist: Fidelity:

F ( r , x ) = Tr

rx

Trace distance:

r -x

tr

= Tr r - x

Bipartite Systems
Suppose Alice and Bob share some state y

Alice

Bob

y
but Bob decides to leave town. What is Alice left with? Combined state: Answer: a mixed state.

y AB

Bipartite Systems
Suppose Alice and Bob share some state y

Alice

Bob

y
but Bob decides to leave town. What is Alice left with? Answer: a mixed state.

Alices state (after Bob leaves town): Partial trace

TrB y y

Partial Trace

Alice

Bob

TrB y y = (I j )y y (I j
j

or

TrB A B = (Tr B ) A

(and extend to all matrices by linearity)

Example

Alice

Bob

1 1 = 0 0 + 11 2 2

TrB j

j = (I 0
+

(I 0 ) + (I 1 )j j (I 1 )
+

)j

1 1 = 0 0+ 1 1 2 2

Example

Alice

Bob

1 1 = 0 0 11 2 2

TrB j

1 1 j = 0 0+ 1 1 2 2
-

Bell Basis
j
+

1 1 = 0 0 + 11 2 2

1 1 = 0 0 11 2 2

1 1 = 0 1 + 1 0 2 2

1 1 = 0 1 1 0 2 2

They all look the same to Alice:

TrB j

j = TrB j
+

j = TrB y
-

= TrB y

1 1 = 0 0+ 1 1 2 2

Schmidt Decomposition

Alice

Bob

y
Suppose we have orthonormal bases for

A and B:

A:

{g

,K , g n

}
n m

B:

{d

, K , dm

It is possible to write

y = a j ,k g j d k
j =1 k =1

for some choice of complex numbers a j ,k

{ }.

Schmidt Decomposition

Alice

Bob

y
The Schmidt decomposition says that there exist particular choices of orthonormal bases

A:

{g

,K , g n

B:

{d

, K , dm

(depending on y ) such that min (n ,m )

eigenvectors of

y =
for some choice of

j =1
j

pj g j d j

TrB y y

{p } .

No cross terms!

Schmidt Decomposition

Alice

Bob

y
The Schmidt decomposition says that there exist particular choices of orthonormal bases

A:

{g

,K , g n

B:

{d

, K , dm

(depending on y ) such that min (n ,m )

eigenvectors of

y =
for some choice of

j =1
j

pj g j d j

TrA y y

{p } .

No cross terms!

Schmidt Decomposition
Several interesting facts follow. For instance The (nonzero) eigenvalues of the reduced states

TrA y y
are the same.

and

TrB y y

min (n ,m )

y =

j =1

pj g j d j
TrA y y = p j d j d j
j

TrB y y = p j g j g j
j

Purifications
The previous fact is often used in conjunction with the fact that every mixed state has a purification: Given a mixed state r, there is an orthonormal basis

{g

,K , g n

} such that
n j =1

r = pj g j g j
Let

y = pj g j g j AB
j =1

Then

TrB y y = p j d j d j = r
j

Schmidt Decomposition
Another interesting consequence of the Schmidt decomposition Suppose y and j are bipartite quantum states

y , j AB
that look the same to Alice:

T rB y y = TrB j j
Then there exists a unitary operator U acting only on B such that

( I U ) y = j

Schmidt Decomposition

Alice

Bob

y a j
Suppose now that Bob doesnt leave town, but instead decides he wants to change the state he shares with Alice to some other (pure) state. What are his choices? He can change the state to any state j for which

TrB y y = TrB j j

Superdense Coding

Alice

Bob

1 1 = 0 0 + 11 2 2

In superdense coding Alice and Bob share an entangled state suppose Bob wants to communicate 2 classical bits to Alice by sending only one qubit.

Superdense Coding

Alice

Bob

j
Encoding: 00 01 10 11

1 1 = 0 0 + 11 2 2
j+ =
1 2 1 = 2 1 = 2 1 = 2 1 11 2 1 0 0 11 2 1 0 1 + 1 0 2 1 0 1 1 0 2 0 0 +

jy+

y-

Superdense Coding

Alice

Bob

1 1 = 0 0 + 11 2 2

All Bell states look the same to Alice

TrB j

j = TrB j
+

j = TrB y
-

= TrB y

1 1 = 0 0+ 1 1 2 2
so Bob can convert between them as he chooses.

Bit Commitment
The same principle can be used to show that an interesting taskbit commitmentis impossible. Bit commitment works as follows: Alice has a bit b {0, 1 }and she wants to commit to this bit but she doesnt want Bob to know the bit until later when she decides to reveal it. Two requirements: binding and concealing.

Bit Commitment
We can imagine implementing bit commitment in the following way: 1. When Alice wants to commit her bit a, she writes a on a piece of paper, locks it in a safe, and sends the safe to Bob. (Alice keeps the key.) Alice Bob

Bit Commitment
We can imagine implementing bit commitment in the following way: 1. When Alice wants to commit her bit a, she writes a on a piece of paper, locks it in a safe, and sends the safe to Bob. (Alice keeps the key.) Alice Bob

Bit Commitment
2. When Alice wants to reveal her bit, she sends Bob the key.

Alice

Bob

Bit Commitment
2. When Alice wants to reveal her bit, she sends Bob the key.

Alice

Bob

Bit Commitment
Information-theoretically secure bit commitment is impossible classically. Quantum bit commitment schemes were proposed in the early 1990s they were originally thought to be secure. But it turns out that they were not secure after all moreover, we now know that quantum bit commitment is impossible using any scheme.

Impossibility of Bit Commitment

Suppose we have a scheme where Alice sends Bob half of some entangled state: Alice Bob

b = 0 state = y 0

b = 1 state = y 1
If the scheme is perfectly concealing, Bob cannot distinguish the two states:

TrA y 0 y 0 = TrA y 1 y 1

Impossibility of Bit Commitment

Suppose we have a scheme where Alice sends Bob half of some entangled state: Alice Bob

b = 0 state = y 0

b = 1 state = y 1
This gives Alice the freedom to change her mind:

(U I ) y 0

= y1

(for some U)

so the scheme cannot be binding.

Entanglement
The notion of entanglement has been mentioned several times so far this week. Archetypal example of an entangled quantum state:

1 1 0 0 + 11 2 2

Entanglement is useful for various tasks: teleportation superdense coding quantum communication protocols quantum computation?

Entanglement is (arguably) not well understood

Entanglement
What is entanglement? Given a pure state of a bipartite system:

y AB
We say that

is a product state if

y g d
for If

g A

and

d B .

is not a product state, then it is entangled.

Entanglement
Mixed state case: r is separable if
k

r = p jx j s j
j =1

for x1 ,K , x k and s 1 ,K , s k mixed states of the first and second system, respectively. If r is not separable, then it is entangled. (Given a density matrix r, it is a very difficult computational problem to test whether it is entangled.)

Entanglement
For example, the following state is not entangled:

11
while this state is entangled:

Which one is more entangled?

1 1 0 0 + 11 2 2
So is this state:

10

-9

0 0 + 1 - 10 1 1

-9

Measures of Entanglement
There are many ways to measure entanglement. how much does it cost to create?

Two natural measures: Entanglement cost.

Distillable entanglement.

how much can you get out of it?

Local quantum operations + classical communication

Alice Bob

r
Alice and Bob share some entangled state r. Any transformation they can perform on r that does not require them to send quantum information is said to be an LOCC transformation.

Entanglement Cost
Suppose Alice and Bob want to share N copies of r (where N is very large), but they only share + copies of j . It is always possible for them to convert kN copies of j + into N copies of r (approximately) via some LOCC transformation for some k. The entanglement cost of r is the infimum over all values of k for which this is possible.

EC (r ) = entanglement cost of r

Distillable Entanglement
Distillable entanglement is essentially the opposite Suppose Alice and Bob share N copies of r (where N is very large), and they want copies j+ . of The distillable entanglement of r is the supremum over all values of k for which they + can extract kN copies of j from N copies of r.

ED (r ) = distillable entanglement of r

The von Neumann Entropy

In the case of pure states, these quantities are always equal:

EC ( y

)= E ( y ) = E ( y )
def

and this quantity is given by the von Neumann entropy of Alices (or Bobs) reduced state:

E( y
where

)= S (Tr

y y )= S ( B y y Tr

S (r ) = - Tr (r log r )

The von Neumann Entropy

Proof starts by looking at the Schmidt decomposition of y :

y = pj g j d j
j =1

A large number of copies N of this state behaves in a very similar way to N independent samples from a random source with respect to the bases

{g

,K , g m

and

{d

, K , dm

Distillation and formation are very similar in spirit to compression and decompression

Mixed state entanglement

Things become much more complicated (and more interesting) for mixed states for instance: The task of testing whether a given density matrix is entangled or separable is NP-hard (with respect to Cook reductions). There exist states r for which

0 < ED ( r ) < EC ( r )
There exist entangled states r for which

ED ( r ) = 0
(bound entangled states).

Diagram of bipartite states

all states PPT NPT

distillable anything here? entangled separable

Example
Is this state distillable?

1 2 r= 0(2 j j k k - j k k j 15 j ,k =
(It is an NPT state, and is conjectured to be undistillable.)

Conclusion
The purpose of this talk has been to give an introduction to the mathematical foundations of quantum information. There are many other interesting topics in quantum information theory. For example: many other aspects of entanglement (such as multiparty entanglement) quantum channel capacities, additivity questions. quantum error correction