The AKLT Model

Motivation, Model and Groundstate

Some backround on representation theory of su(2) is required, see for example the note on the Majumdar-
Ghosh model.
The AKLT Hamiltonian acts on a spin-1 chain. Two spin-1’s decompose as

1⊗1'0⊕1⊕2

and the projector on the spin-2 subspace is

1 1 ~ ~ 1 ~ ~ 2
P (2) = + (S1 · S2 ) + (S1 · S2 ) .
2 2 6
To see this, start form the Casimir operator
~1 + S
(S ~2 ) · (S
~1 + S
~1 ) = (S
~1 · S
~2 ) + 2(S
~1 · S
~ 2 ) + (S
~2 · S
~2 )

which has spectrum {0, 2, 6}. The operators S ~1 · S

~1 and S~2 · S
~2 are the total spin operators of the spin-1
subspaces, so they are each equal to 2 1. It follows that the spectrum of D := (S ~1 · S
~2 ) is {−2, −1, 1}. It is
then an easy matter to check that 1/2 + D/2 + D2 /6 is our desired projector.
The P (1) and P (0) projectors can, using the same method, also be written as a quadratic polynomial in
D. We are doing this work in anticipation of nearest neighbour Hamiltonians that can be written in terms
of these projectors. We see then that the most general candidate (up to a multiplicative factor) is
X 
H(θ) = ~i · S
cos(θ)S ~i+1 + sin(θ)(S
~i · S
~i+1 )2 .
i

The AKLT chain is obtained by choosing θ such that cos(θ) = 3 sin(θ), so that the Hamiltonian can be
written in terms of the P (2) : X (2)
HAKLT = Pi,i+1 .
i

This model is frustration free, as we now show.

The trick is to represent the Hilbert space of the spin-1 chain as a subspace of a spin-1/2 chain with twice
as many sites. i.e. , to each ‘physical site’ of the spin-1 chain we associate two ‘virtual stites’ of a spin-1/2
chain. The subspace of the spin-1/2 Hilbert space is the one that puts the two virtual spin-1/2 particles
beloning to a given physical site in a total spin-1 state, i.e. a ‘triplet’. So the basis for a single physical site
is represented by
1
| ↑↑i, √ (| ↑↓i + | ↓↑i) and | ↓↓i.
2
(i)
Lets denote these states by ψαβ where i labels the physical state amd α, β ∈ {↑, ↓}, where we take ψ↑↓ = ψ↓↑ ,
and choose normalization to comopensate for this redundancy.
Tensor products of these states over the physical sites provide a basis for the physical Hilbert space. Now,
for P (2) not to vanish, the relevant pair of spin-1 particles must be in a total spin-2 state. In our spin-1/2
representation, this is only possible if each pair among the four participating spin-1/2 degrees of freedom are
in a spin-1 triplet state. This is true for the spin-1/2’s belonging to the same physical site, but it need not
be true for a pair where the members of the pair belong to neighbouring physical sites. We can put those
pair in a ‘virtual singlet state’ to obtain a groundstate for the AKLT model:
(1) (L)
Ψ0 ∼ β1 α2 · · · βL α1 ψα1 β1 ⊗ · · · ⊗ ψαL βL .
PICTURE!

1
MPS representation
(shamelessly taken from “The density-matrix renormalization group in the age of matrix product states -
Ulrich Schollwock”)
First we describe the injection of the spin-1 states into the virtual spin-1/2 Hilbert space. Let σ = {σi }Li=1
label the product basis of the spin-1 Hilbert space, σi ∈ {+, 0, −}. Let a = {ai }L i=1 label the ‘left’ virtual
spin-1/2’s and b = {bi }L
i=1 the ‘right’ virtual spin-1/2’s, ai , bi ∈ {↑, ↓}.
We want to implement the following injection:

| ↑↑i 7→ |+i (1)

1
√ (| ↑↓i + | ↓↑i) 7→ |0i (2)
2
↓↓i 7→ |−i. (3)
σ
Do do this on a single site, we consider Mab |σihab| with
     
+ 1 0 0 1 0 1 − 0 0
M = , M =√ and M = .
0 0 2 1 0 0 1
The injection is then given by XX
Maσ11b1 · · · MaσlLbL |σihab|.
σ ab

Now, to get to the actual point. The state we want to inject from the virtual spin-1/2 Hilbert space into
the physical Hilbert space is the one where the bi and ai+1 spin form a singlet. Let’s start by forming one
singlet: X
|Σ(i) i = Σba |bi i ⊗ |ai+1 i
bi ai+1

where  
1 0 1
Σ= √ .
2 −1 0
It is then straightforward to apply this procedure to all pairs:
X
|ΨΣ i = Σb1 a2 · · · ΣbL a1 |abi.
ab

Finally, injecting this state in the physical Hilbert space, we obtain the AKLT groundstate:
X
|Ψ0 i ∼ Tr {M σ1 Σ · · · M σL Σ} |σi.
σ

We can multiply the M ’s with the Σ’s to obtain

X n o
|Ψ0 i ∼ Tr Aeσ1 · · · A
eσL |σi
σ

where      
e+ = √1 0
A
1
, e0 = 1 −1
A
0 e− = √1
and A
0 0
.
2 0 0 2 0 1 2 −1 0
√
To normalize the state in the thermodynamic limit, these matrices should be multiplied by a factor 2/ 3,
thus X
|Ψ0 i = Tr {Aσ1 · · · AσL } |σi
σ
where r     r  
+ 2 0 1 1 −1
0 0 2 0 − 0
A = , A =√ and A = .
3 0 0 3 0 1 3 −1 0