Sie sind auf Seite 1von 3

Löwdin calculus applied to half-filled Hubbard model

Agnieszka Werpachowska∗
February 1, 2011

We consider a Hubbard Hamiltonian of N sites indexed from 0 to N − 1, on each of which


there can be zero or one electron with spin σ = ↑ and zero or one electron with spin σ = ↓:

X NX
−1   N
X −1
H = −t c†i,σ ci−1,σ + c†i−1,σ ci,σ + U ni,↑ ni,↓ , (1)
σ i=1 i=0

where ci,σ , c†i,σ are annihilation and creation operator for spin σ on site i, and

ni,σ = c†i,σ ci,σ

is the occupation number operator. The real parameter t represents the hopping rate and U > 0
describes the repulsive Coulomb interaction between electrons.
The N -site basis is a tensor product:
N
O −1
Ψ= si
i=0

where si is a single-site state, si = ∅, ↑, ↓ or l, where the symbols correspond to no electrons,


single electron with spin up, single electron with spin down or two electrons on site, respectively.
For the Löwdin calculus we choose the basis set A as all states where si are either ↑ or ↓.
The set B contains all other (‘excited’) states. The effective Hamiltonian matrix element between
states ψ1 , ψ2 ∈ A is given by
X Hψ θ Hθψ
Hψeff1 ψ2 = Hψ1 ψ2 + 1 2
,
E − Hθθ
θ∈B

where E is the expected energy of the state we want to model by the effective Hamiltonian that we
find in a variational way (or estimate as it was possible in the case of kinetic Zeeman exchange in
my paper). Hamiltonian H contains only single spin transition terms, so the sum over θ is limited
to states containing 0 electrons on site j, 2 electrons on site j − 1 or j + 1, and 1 electron (with
spin up or down) on any other site. We redefine the set B accordingly.
Firstly, we evaluate Hθθ : the interaction terms do not contribute anything to the diagonal
matrix element, but the Coulomb term does: the site with 2 electrons has the term nj,↑ nj,↓ equal
to 1, all others being 0 (electrons do not interact electrostatically with themselves in the Hubbard
model). Hence, for θ ∈ B,
Hθθ = U .
For the same reasons we have Hψ1 ψ2 = 0.
Secondly, for a given θ, we want to find all pairs of ψ1 , ψ2 ∈ A for which Hψ1 θ Hθψ2 6= 0. Since
H contains all possible spin transitions in the interaction term, such pairs contain those ψ1 and
ψ2 for which one can create θ from ψ2 by transferring a single spin to a neighbouring site and
∗ I would like to thank Professor Carlos Araujo for the idea of applying the Löwdin variation-perturbational

calculus to the half-filled Hubbard model.

1
then create ψ1 from θ by transferring also a single spin to a neighbouring site. As both ψ1 and ψ2
are perfectly half-filled (all sites have 1 electron each), this can happen only if ψ1 = ψ2 or ψ1 is
created from ψ2 by swapping a pair of neighbouring spins. An additional requirement is that we
must be able to transfer a spin at all: that is, the neighbouring site (where the spin goes) must
not contain an electron with this spin already.
The above spin transfers are carried out with the help of creation-annihilation terms from the
interaction part of the Hubbard Hamiltonian (1). To write their action on states from A and B, we
order the electron modes first according to i and then according to σ: (0, ↓), (0, ↑), (1, ↓), (1, ↑), . . . .
We are interested in the actions which transform a state from A (red) into a state from B (blue),
or vice versa. Anticommutation rules lead to the following results (leading and trailing half-filled
parts are denoted by commas):

c†i−1,↓ ci,↓ | . . . , 0, 1, 1, 0, . . . i = −| . . . , 1, 1, 0, 0, . . . i
c†i+1,↑ ci,↑ | . . . , 0, 1, 1, 0, . . . i = −| . . . , 0, 0, 1, 1, . . . i
c†i−1,↑ ci,↑ | . . . , 1, 0, 0, 1, . . . i = | . . . , 1, 1, 0, 0, . . . i
c†i+1,↓ ci,↓ | . . . , 1, 0, 0, 1, . . . i = | . . . , 0, 0, 1, 1, . . . i
(2)
c†i−1,↑ ci,↑ | . . . , 0, 0, 1, 1, . . . i = −| . . . , 0, 1, 0, 1, . . . i
c†i+1,↓ ci,↓ | . . . , 1, 1, 0, 0, . . . i = −| . . . , 0, 1, 1, 0, . . . i
c†i+1,↑ ci,↑ | . . . , 1, 1, 0, 0, . . . i = | . . . , 1, 0, 0, 1, . . . i
c†i−1,↓ ci,↓ | . . . , 0, 0, 1, 1, . . . i = | . . . , 1, 0, 0, 1, . . . i

Let us consider diagonal terms first, ψ1 = ψ2 . States from A can be considered as spin chains.
Each pair of neighbouring spins in the chain with opposite alignment (either ↑ ⊗ ↓ or ↓ ⊗ ↑)
allows us to create two states θ, one by moving the left spin to the right and another by moving
the right spin to the left. Hence, we have
X Hψ θ Hθψ 2t2
1 1
= × number of pairs of opposite neighbouring spins in chain ψ1 .
E − Hθθ E−U
θ∈B

If ψ1 6= ψ2 , it must be that there is one, and only one, pair of neighbouring spins which are
swapped in ψ1 compared to ψ2 , otherwise the contribution to Hψeff1 ψ2 is zero. When it is not zero,
it is equal to −2t2 /(E −U ). The factor 2 comes from the fact that you can transform a pair σ1 ⊗σ2
(where σ1 , σ2 =↑, ↓ and σ1 6= σ2 ) either to l ⊗∅ or ∅⊗ l. The change of sign is a consequence of
the anticommutation rules for fermions, per Eqn. (2).
To summarize, we have
(
2
2t number of pairs of opposite neighbouring spins in chain ψ1 for ψ1 = ψ2 ,
Hψeff1 ψ2 =
E − U −1 when ψ1 is created from ψ2 by swapping a pair of neighbouring opposite spins .

We can express H eff with the use of Pauli spin-1/2 matrices. The non-diagonals term are given
by the action of an operator
N −1
2t2 X
Xi−1,i
E − U i=1
where

Xi−1,i ↑ ⊗ ↓ = − ↓ ⊗ ↑
Xi−1,i ↓ ⊗ ↑ = − ↑ ⊗ ↓
Xi−1,i ↑ ⊗ ↑ = 0
Xi−1,i ↓ ⊗ ↓ = 0

2
Some manipulation of Pauli matrices gives us the result
1 i−1 i
σ σx + σyi−1 σyi .

Xi−1,i = −
2 x
The diagonal term is best derived by considering what happens when we flip the spin at the end
of the chain. If by this action we create another pair of opposite neighbouring spins, the diagonal
matrix element goes up by 2t2 /(E − U ). If we destroy it, it goes down by this amount. This is
equivalent to the action of an operator
N −1
t2 X i−1 i
− σ σz + constant energy.
E − U i=1 z

Therefore, we have
N −1
eff t2 X (i−1) (i)
H =− ~σ · ~σ ,
E − U i=1
meaning that in the Löwding calculus approximation, the half-filled Hubbard model reduces to
the Heisenberg spin chain. Depending on the sign of E − U it is either ferromagnetic (if E > U )
or antiferromagnetic (if E < U ). If |t| ≪ U , we can assume that E < U , obtaining the well-known
result that in this limit the half-filled Hubbard model is antiferromagnetic.

Das könnte Ihnen auch gefallen