Lecture 24:

Tensor Product States

Phy851 Fall 2009

Basis sets for a particle in 3D

Clearly the Hilbert space of a particle in three
dimensions is not the same as the Hilbert
space for a particle in one-dimension
In one dimension, X and P are incompatible
If you specify the wave function in coordinatespace, x|, its momentum-space state is
completely specified as well:
p| =dxp|xx|
You thus specify a state by assigning an
amplitude to every possible position OR by
assigning and amplitude to every possible
momentum

( x) = x

or

( p) = p

In three dimensions, X, Y, and Z, are

compatible.
Thus, to specify a state, you must assign an
position in three
amplitude to each possible
dimensions.
This requires three quantum numbers

( x, y , z ) = x, y , z

So apparently, one basis set is:

{ x, y,z }

x, y,z (x, y,z) R 3

Definition of Tensor product

Suppose you have a system with 10 possible
states
Now you want to enlarge your system by
adding ten more states to its Hilbert space.
The dimensionality of the Hilbert space increases
from 10 to 20
The system can now be found in one of 20
possible states
This is a sum of two Hilbert sub-spaces
One quantum number is required to specify
which state

Instead, suppose you want to combine your

system with a second system, which has ten
states of its own
The first system can be in 1 of its 10 states
The second system can be in 1 of its 10 states
The state of the second system is independent of
the state of the first system

So two independent quantum numbers are

required to specify the combined state

The dimensionality of the combined Hilbert

space thus goes from 10 to 10x10=100
This combined Hilbert space is a
(Tensor) Product of the two Hilbert sub-spaces

Formalism
Let

H1 and H2 be two Hilbert spaces

We will temporarily tag states with a label to

specify which space the state belongs to

(1)

H1

Let the Hilbert space

product of spaces

( 2)

H12

H2

be the tensor-

H1 and H2 .

H 12 = H 1 H 2 H12 is the tensor product

of H1 and H2

The Tensor product state |12=|1(1) |2(2)

belongs to H12.
The KEY POINT TO GET IS:
Bras and kets in the same Hilbert space
attach.

(1)

(1)

1 2

BUT, Bras and kets in different Hilbert

spaces do not attach

(1)

( 2)

( 2)

(1)

They just slide past

each other

Schmidt Basis
The easiest way to find a good basis for a
tensor product space is to use tensor products
of basis states from each sub-space
If:

H1
{|n2(2) } ; n2=1,2,,N2 is a basis in H2
{|n1(1) } ; n1=1,2,,N1 is a basis in

It follows that:
{|n1, n2(12) }; |n1, n2(12)=|n1(1)|n2(2) is a basis in
If System 1 is in state:

(1)

N1

= an1 n1

H12.

(1)

n1 =1

and System 2 is in state: 2

( 2)

N2

= bn2 n2

( 2)

n2 =1

Then the combined system is in state:

1, 2

(12)

= 1

(1)

(2)

N1

N2

= an1 bn 2 n1,n 2

(12)

n1 =1n 2 =1

Schmidt Decomposition Theorem:

All states in a tensor-product space can be

expressed as a linear combination of tensor
product states
N N

12

(12 )

= cn1n2 n1, n2
n1 =1 n2 =1

(12 )

Entangled States
The essence of quantum `weirdness lies in
the fact that there exist states in the tensorproduct space of physically distinct systems
that are not tensor product states
A tensor-product state is of the form

(12 )

= 1

(1)

( 2)

Tensor-product states are called factorizable

The most general state is

12

(12 )

N1

N2

= cn1n2 n1, n2

(12 )

n1 =1 n2 =1

This may or may-not be factorizable

Non-factorizable states are called entangled

For an `entangled state, each subsystem has no independent objective
reality

Configuration Space
The state of a quantum system of N
particles in 3 dimensions lives in
configuration space (We dont know about spin yet)
There are three quantum numbers
associated with each particle
It takes 3N quantum numbers to specify a
state of the full system

Coordinate Basis:

{ x1, y1,z1, x 2, y 2 ,z2,..., x N , y N ,zN }

or we could just write

r r
r
{ r1, r2,..., rN

Wavefunction:

This form is the

coordinate system
independent
representation

r r
r
r r
r

(r1 , r2 ,..., rN ) = r1 , r2 ,..., rN

To specify a state of N particles in d
dimensions requires dN quantum
numbers
So counting quantum numbers might be a good
way to check if you are using a valid basis

Tensor Products of Operators

THEOREM:
Let A (1) act in

H1 , and B(2) act in H2 ,

Then the tensor product operator C(12) =A(1) B(2)

acts in H12 .

PROOF:

A(1) = am am

(1)

am

(1)

B ( 2 ) = bm bm

( 2)

bm

( 2)

A(1) B (2) = am bn am

(1)

am

(1)

(2)

bn

bn

(2)

m,n

= am bn am
m,n

(1)

(12 )

C (12 ) = ambn am , bn

bn

(2)

)(

am , bn

am

(1)

bn

(2)

(12 )

m,n

The action of C(12) on a tensor-product state:

1 , 2
C

(12 )

1 , 2

(12 )

(12 )

:= 1

(1)

= ambn am , bn
m,n

(12 )

( 2)

am 1

(1)

bn 2

( 2)

General form of Operators in

Tensor-product spaces
The most general form of an operator in
is:

C (12 ) = cm ,n;m ',n ' m, n

(12 )

m' , n'

H12

(12 )

m,n

cm ,n;m ',n ' := m, n

(12 )

C (12 ) m, n

(12 )

Here |m,n may or may not be a tensor

product state. The important thing is that
it takes two quantum numbers to specify
a basis state in H12
A basis that is not formed from tensorproduct states is an entangled-state basis

In the beginning, you should always start

with a tensor-product basis as your physical
basis
Then all operators are well-defined
Just expand states and operators onto tensorproduct states
Then match up the bras and kets with their
proper partners when taking inner products

Upgrading Subspace Operators

Any operator in

H1

can be upgraded to an

H12 by taking the tensor product

with the identity operator in H2 :
operator in

A(12 ) = A(1) I ( 2 )
If A1 is an observable in H1, then it is also an
observable in H12 (since it remains Hermitian
when upgraded).
The spectrum of A1 remains the same after
upgrading

Proof:
Let

A(1) am

(1)

= am am

(1)

Then:

A(1) I (2) am

(1)

(2)

) (

(1)

= A(1) am
= am am

= am am

(1)

(1)

Note that |2 is completely

) (

arbitrary, but |a1|2 is an

eigenstate of A(12)

(2)

(2)

(2)

(2)

Product of two Upgraded Operators

Let A(1) and B(2) be observables in their
respective Hilbert spaces
Let A(12)= A(1) I(2) and B(12)= I(1) B(2) .
The product A(12)B(12) is given by A(1) B(2)
Proof:

( A1 I2 )( I1 B2 ) = A1I1 I2 B2
= A1 B2

Compatible observables
Let A(1) and B(2) be observables in their
respective Hilbert spaces
Let A= A(1) I(2) and B= I(1) B(2) .
Theorem: [A,B]=0
Proof:

[A, B]= AB BA
= (A1 I 2 )(I1 B2 ) (I1 B2 )(A1 I 2 )
= (A1 I1 ) (I 2 B2 ) (I1 A1 ) (B2 I 2 )
= A1 B2 A1 B2 = 0

Conclusion: any operator in

H1 , is compatible

with any operator in H2 ,.

I.e. simultaneous eigenstates exist.
Let A1 |a1= a1 |a1 and B2 |b2= b2 |b2.
Let a= a1 and b= b2
Let |ab= |a1 |b2.
Then AB|ab=ab|ab.

And versus Or
The tensor product correlates with a system
having property A and property B
Dimension of combined Hilbert space is product
of dimensions of subspaces associated with A
and B
Example: start with a system having 4 energy
levels. Let it interact with a 2 level system. The
Hilbert space of the combined system has 8
possible states.

Hilbert spaces are added when a system can

have either property A or property B
Dimension of combined Hilbert space is sum of
dimensions of subspaces associated with A and
B
Example: start with a system having 4 energy
levels. Add 2 more energy levels to your model,
and the dimension goes from 4 to 6

Example #1 : Particle in Three Dimensions

Let H1 be the Hilbert space of functions
in one dimension
The projector is: I1 = dx x x

So a basis is:

{x }

Then H3= (H1)3 would then be the

Hilbert space of square integrable

functions in three dimensions.

Proof: I3 = I1 I1 I1

dx x x dy y y dz z z
= dxdydz x x y y z z
= dxdydz( x y z )( x y z )
= dxdydz x, y,z x, y,z
=

x, y , z x y z

( x, y , z ) = x, y , z

= dxdydz x, y, z ( x, y, z )

Note: H3 is also the Hilbert space of

three particles in one-dimensional space

Three-dimensional Operators
We can define the vector operators:

r
R = X x + Y y + Z z
r
P = Px x + Py y + Pz z
r
R x, y, z = (x x + y y + z z ) x, y, z
rr rr
or R r = r r
Note that: X = X(1)I(2) I(3) and Py = I(1)P (2) I(3)
so that [X, Py]=0.
With

R1 = X

P1 = Px

R2 = Y

P2 = Py

R3 = Z

P3 = Pz

We can use:

[R ,R ] = [P ,P ] = 0
[R ,P ] = ih
j

jk

Example #2: Two particles in One

Dimension
For two particles in one-dimensional space,
the Hilbert space is (H1 )2.

x1 , x2 = x1

(1)

x2

( 2)

( x1 , x2 ) = x1 , x2
I=

dx dx
1

x1, x 2 x1, x 2 = 1

X1 = X1 I2
P2 = I1 P2

etc...

[ X , X ] = [P ,P ] = 0
[ X ,P ] = ih
j
j

jk

Hamiltonians
One particle in three dimensions:
Each component of momentum contributes
additively to the Kinetic Energy

1
H=
Px2 + Py2 + Pz2 + V ( X , Y , Z )
2m
v
1 r r
=
P P + V ( R)
2m

Two particles in one dimension:

P12
P22
H=
+
+ V ( X1, X 2 )
2m1 2m2

Conclusions
The take home messages are:
The combined Hilbert space of two
systems, dimensions d1 and d2, has
dimension d12=d1d2
A physical basis set for the combined
Hilbert space, H12 can be formed by
taking all possible products of one basis
state from space H1 with one basis state
from H2.

n1

(1)

H 1,

n1,n 2 := n1
{ n1,n 2

(1)

n2

n2

(2)

} H

(2)

} H 12 := H 1 H 2

In a tensor product space, a bra from one

subspace can only attach to a ket from

the same (1subspace:

)
(1)

(1)

( 2)

1 2

( 2)

(1)

For N particles (spin 0) in d dimensions,

dN quantum numbers are required to
specify
in any basis
r r a unit-vector
r
r1 , r2 , K rN

n1x , n1 y , n1z , n2 x , n2 y , n2 z , K , nNx , nNy , nNz