Quantum Matter / Matéria Quântica

Solutions to Problem set 5 — Hubbard model

Notation recap

The problems below use the notation and conventions of the lectures, which is standard for the
Hubbard and related models of quantum magnetism. The (repulsive) Hubbard Hamiltonian for
electrons in a single band is given by

T̂ ĤU
z }| { z
}| {
XX X
Ĥ = T̂ + ĤU = −t c†is cjs +U ni↑ ni↓ , U, t > 0, (1)
hi,ji s i

where hi, ji means that the summation is restricted to nearest-neighbor (NN) sites; t is the hopping
amplitude that characterizes the electronic band of interest; c†is is the creation operator for an
electron at a Wannier orbital located at site i with spin s ∈ {↑, ↓}; nis ≡ c†is cis is the number
operator for a given site i and spin s. It is convenient to define the additional number operators:
X X X X X
ni ≡ nis = ni↑ + ni↓ , N̂s ≡ nis , N̂ = N̂s = nis = (ni↑ + ni↓ ) = N̂↑ + N̂↓ ,
s i s i,s i

where N̂ is the total particle number operator and N̂s the total number operator for particles with
spin s.

Spin operators:

• To simplify the expressions, spin observables used below are in units of ~.

• Use Ŝiα to designate the Cartesian component α of the spin operator at site i (e.g.: Ŝiz ).
P
• Use Ŝtot ≡ i Ŝi to designate the total spin of a system of particles.

• The operator for the spin at site i can be written as (implicit sum over repeated indices)

1
Ŝi ≡ c†iα σαβ ciβ , α, β ∈ {↑, ↓}, (2)
2
where σ ≡ σ x x̂ + σ y ŷ + σ z ẑ is the vector of Pauli matrices.

1
Problem 1 (some general properties and results)

a) We can express any observable that is bilinear in the particle operators using a “spinor”-type
representation:
X † h i
 = ci Aij cj = ψ † A ψ, ψ † ≡ c†1 c†2 · · · c†M ,
ij

where A is an Hermitian matrix (Aij = A∗ji ), ψ † /ψ is a row/column of particle operators and

M is the dimension of the 1-particle Hilbert space in question. Show that, given two such
observables Â, B̂ and the respective matrices A, B, that define their spinor representation,
their commutator is given by
   
Â, B̂ = ψ † A, B ψ,
 
where A, B is the commutator of the two matrices.

Remark — This property applies equally to bosons and fermions.

b) Recall the representation of the spin operator given in eq. (2) of the “Notation recap” above.
Using the fermionic commutation relations, explicitly show that the angular momentum rela-
tions are indeed reproduced by the definition above, i.e.,
h i
Ŝjm , Ŝjn = i εlmn Ŝjl .

Suggestion — The result of the previous question is helpful and the following property of Pauli matrices
may be useful: (a · σ)(b · σ) = a · b + iσ · (a × b).

c) Show that Ŝi2 = 41 ni − 23 ni↑ ni↓ and, consequently, we may write the Hubbard interaction in a
manifestly rotation-invariant form:
X 2U X 2 U
ĤU ≡ U ni↑ ni↓ = − Ŝi + N̂ .
3 6
i i
P m m
Suggestion — You may use the following property of Pauli matrices: m=x,y,z σαβ σµν = 2δαν δβµ −δαβ δµν .

d) Verify that the total spin of the system is a conserved quantity in the Hubbard model:
  X
Ŝtot , Ĥ = 0, Ŝtot ≡ Ŝi .
i

e) Show that the Hubbard interaction can be alternatively expressed as

!
X U X UX 2 UX 2
ĤU ≡ U ni↑ ni↓ = n2i − N̂ or also as ĤU = ni − mi ,
2 4 4
i i i i

where mi ≡ ni↑ − ni↓ is the operator for the spin polarization at site i.

Solution

2
a) Let us use the notation
 
Â, B̂ ∓ ≡ ÂB̂ ∓ B̂ Â

to simultaneously refer to the commutator (−) and anticommutator (−), which allows us to demonstrate
the desired result for bosons and fermions in the same calculation. The commutator is given by
X X  XX h i
 
Â, B̂ − = c†i Aij cj , c†r Brs cs = Aij Brs c†i cj , c†r cs ,
− −
ij rs ij rs

and
h i
c†i cj , c†r cs = c†i cj c†r cs − c†r cs c†i cj
−
 
= c†i cj c†r cs − c†r [cs , c†i ]∓ ± c†i cs cj
 
= c†i cj c†r cs − c†r δsi ± c†i cs cj

= c†i cj c†r cs − c†r cj δsi ∓ c†r c†i cs cj

= c†i cj c†r cs − c†r cj δsi ∓ c†i c†r cj cs


= c†i cj c†r cs − c†r cj δsi ∓ c†i [c†r , cj ]∓ ± cj c†r cs
= c†i cj c†r cs − c†r cj δsi ∓ (∓δrj ) c†i cs − c†i cj c†r cs
= −c†r cj δsi + δrj c†i cs .

Hence,
  XX   X X
Â, B̂ −
= Aij Brs δrj c†i cs − c†r cj δsi = Aij Bjs c†i cs − Aij Bri c†r cj
ij rs ijs ijr
! ! !
X X X X X X X
= Aik Bkj c†i cj − Bik Akj c†i cj = Aik Bkj − Bik Akj c†i cj
ij k ij k ij k k
X
= [A, B]ij c†i cj ,
ij

or, in the spinorial representation,

   
Â, B̂ −
= ψ † A, B ψ.

b) Using the result demonstrated in the previous question (we momentarily drop the indices j, since the
commutator refers to the same spin and implicitly sum over repeated indices),

1 
[Ŝ m , Ŝ n ] = c†α σαβm n †
cβ , c†µ σµν cν
 4

y use generic result of previous question
1
= c†α [σ m , σ n ]αβ cβ
 4

y commutator of Pauli matrices: [σ m , σ n ] = 2iεlmn σ l
1 l
= iεlmn c†α σαβ cβ
2
= iεlmn Ŝ l .

Reinstating the particle index,

[Ŝjm , Ŝjn ] = iεlmn Ŝjl .

3
c) Replacing the fermionic representation (2) of the spin operator,
   
1 † 1 †
Ŝi2 = c σαβ ciβ · ciµ σµν ciν
2 iα 2
1
= (σαβ · σµν ) c†iα ciβ c†iµ ciν
 4 X m m

y use the property σαβ σµν = 2δαν δβµ − δαβ δµν
m
1
= (2δαν δβµ − δαβ δµν ) c†iα ciβ c†iµ ciν
4
1 1
= c†iα ciβ c†iβ ciα − c†iα ciα c†iµ ciµ
2 4
1   1
= c†iα ciβ {c†iβ , ciα } − ciα c†iβ − ni ni
2 4
1 †   1
†
= ciα ciβ δαβ − ciα ciβ − n2i
2 4
1 1 † † 1 2
= ni − ciα ciβ ciα ciβ − ni
2 2 4
1 1 † † 1 2
= ni + ciα ciα ciβ ciβ − ni
2 2 4
1 1 † 1
= ni + ciα ciα (1 − ni ) − n2i
2 2 4
1 1 1 2 1 2
= ni + ni − ni − ni
2 2 2 4
3 2
= ni − ni .
4
Noting that

n2i = (ni↑ + ni↓ )2 = n2i↓ + n2i↓ + 2ni↓ ni↓ = ni↓ + ni↓ + 2ni↓ ni↓ = ni + 2ni↓ ni↓ ,

where we used the property n2iα = niα applicable to fermions, we get

3 3 1 3
Ŝi2 = ni − n2i = ni − (ni + 2ni↓ ni↓ ) = ni − ni↓ ni↓ .
4 4 4 2
The Hubbard interaction becomes
X 1 X 2 X 2 U 2 X 2
ĤU ≡ U ni↑ ni↓ = U ni − U Ŝi = N̂ − U Ŝi .
i
6 i
3 i
6 3 i

P
d) Since Ŝtot ≡ i Ŝi , we have
  X  X 
Ŝtot , Ĥ = Ŝi , T̂ + Ŝi , ĤU .
i i

From the previous question,

  U  2 X  ✘2✘  0 U
✿ 
Ŝi , ĤU = Ŝi , N̂ − U ✘ i , Ŝj
Ŝ✘ = Ŝi , N̂ ,
6 3 j
6

and note that

1X † X XX    X † 
Ŝi = ciα σαβ ciβ , N̂ = c†jµ cjµ = c†jµ δµν cjν , Ŝi , N̂ = Ŝi , ciµ δµν ciν ,
2 j,µ j µν µν
αβ

4
 because the commutator with the other particles j 6= i is identically zero. We may then use the property
established in question 1:
  1X †
Ŝi , N̂ = ciα [σ, 1]αβ ciβ = 0,
2
αβ

where [σ, 1] represents the commutator of the Pauli vector with the 2 × 2 identity matrix. (The result
 
Ŝi , N̂ = 0 is expected because Ŝi does not change the total number of particles.) The remaining
commutator is
X  1 XX XX
Ŝk , T̂ = − tij σαβ [c†kα ckβ , c†is cjs ]
2 ij s αβ
k k


y from question 1: c†i cj , c†r cs = −c†r cj δsi + δrj c†i cs
 

1 XX XX  
=− tij σαβ c†kα cjs δki δβs − c†is ckβ δkj δαs
2 ij s αβ
k
 
1 X X X X X
=− tkj σαβ c†kα cjβ − tik σαβ c†iα ckβ 
2 j i
k αβ αβ
1 X X  
=− σαβ tki c†kα ciβ − tik c†iα ckβ
2
αβ ki


y interchange the dummy indices i, k in the second term
1X X 
=− σαβ tki c†kα ciβ − tki c†kα ciβ
2
αβ ki

= 0.

In conclusion,
0
  X ✯
✟ 0 X ✟✟✯

Ŝtot , Ĥ = Ŝ
✟i ,
✟ T̂ + Ŝ ,
✟
✟i U Ĥ = 0.
✟i ✟ ✟i ✟

e) This comes down to expressing the product ni↑ ni↓ in different forms. The first case is a direct application
of an intermediate result obtained in the solution to question (c) above:

n2i = (ni↑ + ni↓ )2 = n2i↓ + n2i↓ + 2ni↓ ni↓ = ni↓ + ni↓ + 2ni↓ ni↓ = ni + 2ni↓ ni↓ .

Consequently, !
X U X 2
U X
ĤU = U ni↑ ni↓ = ni − ni = n2i − N̂ .
i
2 i 2 i

For the second expression, note that

(
mi = ni↑ − ni↓ n i + mi n i − mi
−→ ni↑ = , ni↓ = ,
ni = ni↓ + ni↓ 2 2

5
 and therefore,
X X  n i + mi   n i − mi 
ĤU = U ni↑ ni↓ = U
i i
2 2
U X

= n2i − m2i + mi ni − ni mi
4 i
UX 2

= ni − m2i .
4 i

Problem 2 (two-site Hubbard problem)

The Hubbard problem can be addressed exactly “by hand” when the number of lattice sites, Nc , is
very small. In the questions below, we will do precisely that for the case of two sites (Nc = 2), in
which case the Hubbard Hamiltonian (1) reduces to
X  †  X
Ĥ = −t c1s c2s + c†2s c1s + U ni↑ ni↓ .
s=↑,↓ i=1,2

With 2 sites, we can accommodate from zero up to 4 electrons in this system. We can also adopt
the following notation to keep track of the various Fock (occupation number) states:

| 0 , 0 i≡| 0 i, | ↑, 0i ≡ c†1↑ |0i, | ↑, ↑i ≡ c†1↑ c†2↑ |0i, | ↑↓, 0i ≡ c†1↑ c†1↓ |0i, etc.
↑ ↑ ↑
site 1 site 2 vacuum

In the notation | · , · i, the first/left slot refers to the occupation of site 1 and the second/right slot
to the occupation of site 2. We should be mindful of the anti-commutation property of fermions:
for singly occupied states, let us use the convention where the operator for site 1 is always to the
left of that of site 2, for example,

| ↑, ↓i ≡ c†1↑ c†2↓ |0i, −→ c†2↓ c†1↑ |0i = −c†1↑ c†2↓ |0i = −| ↑, ↓i;

for doubly occupied states, the operator order similarly reflects the relative position of the arrows,
for example

| ↑↓, 0i ≡ c†1↑ c†1↓ |0i −→ | ↓↑, 0i = c†1↓ c†1↑ |0i = −c†1↑ c†1↓ |0i = −| ↑↓, 0i.

Keep this anti-symmetry in mind (the minus signs!) when evaluating matrix elements below.

a) Let Ne be the number of electrons. Since [Ĥ, N̂ ] = 0, we may organize the state space in sub-
spaces with fixed Ne and consider each subspace independently. Enumerate the Fock states
that define a basis for each of those subspaces, for all Ne ∈ {0, 1, 2, 3, 4}. Hence, find the
dimension of each subspace and, from there, verify that the dimension of the full Fock space
of this problem is 16.

6
 Remark — More generally, in the Hubbard model with Nc sites the full Fock space has dimension 4Nc .
Can you see how we obtain this value? (Also, note that it grows exponentially with the lattice size...)

b) Here and in all the remaining questions below consider only the subspace with Ne = 2 (half-
filling). In the previous question, you should have found that its dimension is 6. A possible
basis could be
basis I: {| ↑↓, 0i, | ↑, ↑i, | ↑, ↓i, | ↓, ↑i, | ↓, ↓i, |0, ↑↓i}.

Verify that, in this basis, the matrix representation of Ĥ is

 
U 0 −t t 0 0
 
0 0 0 0 0 0
 
−t 0 0 0 0 −t
Ĥ 7→ 
t
, (basis I).
 0 0 0 0 t 
 
0 0 0 0 0 0
0 0 −t t 0 U

c) With the above matrix representation, we instantly identify that | ↑, ↑i and | ↓, ↓i are eigen-
states of Ĥ, both with eigenvalue zero [not surprisingly, right?]. To obtain the rest of the
spectrum, we would still need to diagonalize the remaining 4 × 4 block of the matrix above.
But we can do better by choosing the basis in a smarter way that takes advantage of the sym-
metries of Ĥ, in particular the rotational symmetry of the Hubbard Hamiltonian (Problem 1),
which suggests we should use a basis of eigenstates of the total spin. Since we are adding two
spins 1/2, the possible magnitudes of Stot are 0 and 1. A basis adapted for each respective
subspace is thus

Stot =1 Stot =0
z }| { z }| {
n | ↑, ↓i + | ↓, ↑i | ↑, ↓i − | ↓, ↑i o
basis II: | ↑, ↑i, | ↓, ↓i, √ , | ↑↓, 0i, |0, ↑↓i, √
2 2

2 ] = 0, we expect Ĥ to be block-diagonal in this basis. By explicitly computing

Since [Ĥ, Ŝtot
the matrix elements in this new basis, verify that Ĥ has such block-diagonal structure, namely,
 
 
 0 0 
 
 
Ĥ 7→  √  , (basis II).
 U 0 − 2t 

 √  
 0 0 U − 2t 
√ √
− 2t − 2t 0

d) Verify that, in addition to the 3-fold degenerate energy eigenvalue E = 0 (for the subspace
√

Stot = 1), the remaining energy eigenvalues are U and 12 U ± U 2 + 16t2 .

e) Consider the limit U ≫ t and obtain the leading correction in t/U to the eigenenergy eigen-
values in that limit. Make a schematic of the relative position of the 6 energy levels and verify

7
 that: (i) the 6 energy levels cluster into two groups separated by ∼ U ; (ii) within each cluster,
non-degenerate levels are separated by ∼ t2 /U .

Comment — In the thermodynamic limit with Nc = Ne ≫ 1, these two clusters become continuous bands,
known as the upper and lower Hubbard bands, or as the high and low energy bands.

f) Obtain the value of the ground state energy, Egs , in the two opposite limits t/U → 0 and
U/t → 0. Thinking on the underlying physics, qualitatively explain why, in each limit, the
value of Egs makes sense for this system (2 sites and 2 electrons).

g) Obtain the exact expression for the ground state, |Ψgs i, and verify that it is a linear combi-
nation of singly and doubly occupied states.

Suggestion — Feel free to use a symbolic calculation tool (Mathematica, WolframAlpha, etc.) to expedite
the computation of the eigenvectors of the nonzero 3 × 3 block of Ĥ.

h) Verify that, when U/t ≫ 1,

| ↑, ↓i − | ↓, ↑i t
|Ψgs i = √ +O ,
2 U

which explicitly demonstrates that double occupancy is suppressed for large U .

i) Consider the large-U limit, where the low-energy sector contains 4 states and the ground state
energy is Egs ≃ −4t2 /U . Show that these 4 eigenstates and the corresponding eigenenergies
map exactly to the spectrum and eigenstates of two spins (one on each site) coupled by
antiferromagnetic Heisenberg exchange, J Ŝ1 · Ŝ2 , with exchange coupling J = 4t2 /U .

Comment — Note that this establishes, in this specific case with only 2 sites, what we discussed in the
U≫t
lectures: if P̂ designates the projector onto the low-energy subspace, then P̂ Ĥ P̂ −−−−→ J Ŝ1 · Ŝ2 with the
effective exchange coupling being J ≡ 4t2 /U > 0. In plain words, this means that the effective electron-
electron interaction at low energies of the half-filled Hubbard model is an antiferromagnetic exchange
interaction between the localized spins. Electrons are largely frozen on their respective sites since hopping
costs a double-occupancy energy ∼ U ; the system is thus an AFM insulator (a strongly correlated insulator).

Solution

a) Using the notation described in the text of the question,

Ne = 0 : |0, 0i. (dim = 1)

Ne = 1 : | ↑, 0i, |0, ↑i, | ↓, 0i, |0, ↓i. (dim = 4)
Ne = 2 : | ↑, ↑i, | ↑, ↓i, | ↓, ↑i, | ↓, ↓i, | ↑↓, 0i, |0, ↑↓i. (dim = 6)
Ne = 3 : | ↑↓, ↑i, | ↑↓, ↓i, | ↑, ↑↓i, | ↓, ↑↓i. (dim = 4)
Ne = 4 : | ↑↓, ↑↓i. (dim = 1)

The dimension of the Fock space with 2 sites is thus 1 + 4 + 6 + 4 + 1 = 16.

8
b) Let us compute the action of Ĥ,
X   X
Ĥ = −t c†1s c2s + c†2s c1s + U ni↑ ni↓ ,
s=↑,↓ i=1,2

on each of the states of the given basis I. In detail, for the doubly occupied states, we have:

X X † ✘✘✘
✿0
✘
Ĥ| ↑↓, 0i = U | ↑↓, 0i − t c†2s c1s | ↑↓, 0i − t c c
✘
✘1s 2s | ↑↓, 0i
s s
 
= U | ↑↓, 0i − t c†2↑ c1↑ + c†2↓ c1↓ | ↑↓, 0i

= U | ↑↓, 0i − tc†2↑ c1↑ c†1↑ c†1↓ |0i − tc†2↓ c1↓ c†1↑ c†1↓ |0i
 
= U | ↑↓, 0i − tc†2↑ 1 − c†1↑ c1↑ c†1↓ |0i + tc†2↓ c†1↑ c1↓ c†1↓ |0i
 
= U | ↑↓, 0i − tc†2↑ c†1↓ |0i + tc†2↑ c†1↑ c1↑ c†1↓ |0i + tc†2↓ c†1↑ 1 − c†1↓ c1↓ |0i
  0
✯
✟
✟  ✟ 0
✯
✟
= U | ↑↓, 0i + tc†1↓ c†2↑ |0i − tc†2↑ c†1↑ c†1↓ c✟ ✟ + tc† c† |0i − tc† c† c† c✟|0i
|0i ✟
✟ 1↑ 2↓ 1↑ 2↓ 1↑ 1↓
✟ 1↓
† † † †
= U | ↑↓, 0i + tc1↓ c2↑ |0i − tc1↑ c2↓ |0i
= U | ↑↓, 0i + t| ↓, ↑i − t| ↑, ↓i,

and, proceeding similarly, we obtain

Ĥ|0, ↑↓i = U |0, ↑↓i + t| ↓, ↑i − t| ↑, ↓i.

For the singly occupied states, the Hubbard term yields zero (it’s only nonzero when acting on a doubly
occupied state). Hence,
Ĥ| ↑, ↑i = Ĥ| ↓, ↓i = 0,

because the kinetic term T̂ promotes hopping without spin flipping, which is not allowed for these two
states. The remaining cases are
X X
Ĥ| ↑, ↓i = −t c†2s c1s | ↑, ↓i − t c†1s c2s | ↑, ↓i
s s
† †
= −tc2↑ c1↑ |
↑, ↓i − tc1↓ c2↓ | ↑, ↓i
= −t|0, ↑↓i − tc†1↓ c2↓ c†1↑ c†2↓ |0i
= −t|0, ↑↓i + tc†1↓ c†1↑ c2↓ c†2↓ |0i
 
= −t|0, ↑↓i + tc†1↓ c†1↑ 1 − c†2↓ c2↓ |0i

= −t|0, ↑↓i + tc†1↓ c†1↑ |0i

= −t|0, ↑↓i − tc†1↑ c†1↓ |0i
= −t|0, ↑↓i − t| ↑↓, 0i

9
 and
X X
Ĥ| ↓, ↑i = −t c†2s c1s | ↓, ↑i − t c†1s c2s | ↓, ↑i
s s
† †
= −tc2↓ c1↓ | ↓, ↑i − tc1↑ c2↑ | ↓, ↑i
= −tc†2↓ c1↓ c†1↓ c†2↑ |0i − tc†1↑ c2↑ c†1↓ c†2↑ |0i
!
✟✯ 0
= −tc†2↓ 1 − ✟ c†1↓✟
c1↓ c†2↑ |0i + tc†1↑ c†1↓ c2↑ c†2↑ |0i
!
† † † † † ✯0
✟
= −tc2↓ c2↑ |0i + tc1↑ c1↓ 1 − ✟ c2↑ |0i
c2↑✟

= +tc†2↑ c†2↓ |0i + tc†1↑ c†1↓ |0i

= t|0, ↑↓i + t| ↑↓, 0i.

Since the basis

{| ↑↓, 0i, | ↑, ↑i, | ↑, ↓i, | ↓, ↑i, | ↓, ↓i, |0, ↑↓i} (basis I)

is orthonormal, we obtain the following matrix representation:

 
U 0 −t t 0 0
0 0 0 0 0 0
 
−t 0 0 0 0 −t
 
Ĥ 7→  , (basis I).
t 0 0 0 0 t
 
0 0 0 0 0 0
0 0 −t t 0 U

We can see that Ĥ has a block-diagonal structure,

 
U 0 −t t 0 0
 
 0 0 0 0 0 0 
 
 −t 0 0 0 0 −t 
Ĥ 7→  , (basis I),

 t 0 0 0 0 t 

 
 0 0 0 0 0 0 
0 0 −t t 0 U

with the two eigenvectors | ↑, ↑i and | ↓, ↓i belonging to eigenvalue zero (the second and fourth one-
dimensional block along the diagonal). This is not surprising because we can easily see that, since | ↑, ↑i
and | ↓, ↓i are both singly occupied states with spins parallel on each site, the action of Ĥ can only yield
zero. Therefore, these should be eigenstates associated with zero energy.

c) We must now consider basis II, given by

Stot =1 Stot =0
z }| { z }| {
n o
| ↑, ↑i, | ↓, ↓i, |T i, | ↑↓, 0i, |0, ↑↓i, |Si , (basis II),

where, for simplicity, we named (for “triplet” and “singlet” states)

| ↑, ↓i + | ↓, ↑i | ↑, ↓i − | ↓, ↑i
|T i ≡ √ , |Si ≡ √ .
2 2

10
 Using the results computed in the previous question for the action of Ĥ, we readily obtain:

−t|0, ↑↓i − t| ↑↓, 0i t|0, ↑↓i + t| ↑↓, 0i

Ĥ|T i = √ + √ = 0,
2 2
−t|0, ↑↓i − t| ↑↓, 0i t|0, ↑↓i + t| ↑↓, 0i √ √
Ĥ|Si = √ − √ = − 2t|0, ↑↓i − 2t| ↑↓, 0i,
2 2

and
√ √
Ĥ| ↑↓, 0i = U | ↑↓, 0i − 2t|Si, Ĥ|0, ↑↓i = U | ↑↓, 0i − 2t|Si.

Therefore, the matrix representation in this basis becomes

 
 
 0 0 
 
 
Ĥ 7→  √  , (basis II).
 U 0 − 2t 

 √  
 0 0 U − 2t 
√ √
− 2t − 2t 0

d) Since the first diagonal block is identically zero, we need only find the eigenvalues of the second block:
√
U −E 0 − 2t
√  
0 U −E − 2t = 0 ⇔ (U − E) (U − E)E − 2t2 − 2t2 (U − E) = 0
√ √
− 2t − 2t −E
⇔ E=U ∨ (U − E)E − 4t2 = 0
1 p

⇔ E=U ∨ E = U ± U 2 + 16t2 .
2
So, in order of increasing energy, the full spectrum En is
√
U− U 2 + 16t2
E1 = , E2 = E3 = E4 = 0,
2 √
U + U 2 + 16t2
E5 = U, E6 = ,
2
where we note that the ground state has energy E1 < 0.

e) In the limit U ≫ t, the spectrum is approximately given by (expanding the square roots to leading order)

4t2 4t2
E1 ≃ − , E2 = E3 = E4 = 0, E5 = U, E6 = U + .
| U {z } | {z U}
low energy sector high energy sector

Since t2 /U ≪ 1, there are two well separated clusters of energies: a low energy sector near zero energy
comprising the levels n = 1, 2, 3, 4, and a high energy sector comprising the levels n = 5, 6, with energies

11
 near U . The energy amplitude of each cluster (the bandwidth) is clearly 4t2 /U :

—————— E6
2
l 4tU
—————— E5
x


U
y

—————— E2,3,4
2
l 4tU
—————— E1

f) From the previous question,


√ − 4t2 , U/t→∞
U− U 2 + 16t2 U U ≫t −−−−−−−−−→ 0
Egs = E1 = ≃
U/t→0
2 −2t 1 − U
, U ≪t −−−−−−−−→ −2t
4t

In the limit U/t → ∞ we expect total suppression of double occupancy: the kinetic term T̂ becomes irrele-
vant because, although there is only one electron at each site, there can be no hopping to the neighboring
sites since that would generate a doubly occupied state with a energy cost of U → ∞. In this regime, the
electrons are effectively “frozen” (localized) at fixed sites and hopping is forbidden. They are in a strongly
correlated, localized state; their energy will be just the local orbital energy, which was set to zero in the
definition of the Hubbard Hamiltonian. Consequently, in the asymptotic limit U/t → ∞, the ground state
will have zero energy and degeneracy 4.

In the opposite limit, U/t → 0, the Hubbard term, ĤU , becomes irrelevant relative to T̂ . The elec-
trons become non-interacting, which effectively removes any restrictions (correlations) in the occupations
of any state due to the electron-electron interaction. The ground state in this case will be the ground
state of two free electrons that can hop among the two sites; we thus expect the ground state to be the
non-interacting bonding state of this pair, which we know has energy −t. Since the two electrons can now
occupy this anti-bonding state, the total energy is 2 × (−t) = −2t.

g) We found in the previous questions that, in basis II, Ĥ has the block structure

Ĥ = Ĥ (S=1) ⊕ Ĥ (S=0) ,
√
U− U 2 +16t2
and that the ground state, with energy Egs = 2 , belongs to the subspace/block with S = 0.
Solving for the corresponding eigenvector,
 √    
U 0 − 2t u u
√  
Ĥ (S=0) |ψgs i = Egs |ψgs i
  
⇔ 0
 √ U − 2t  v  = Egs  v 
√
− 2t − 2t 0 w w
( √ √ √
Uu − 2tw = Egs u u 2t v 2t
√ ⇔ = , = ,
U v − 2tw = Egs v w U − Egs w U − Egs

12
 √
U+ U 2 +16t2
or, since U − Egs = 2 ,
   
u x
t
v = N  √ x
   
, x≡ .
1+ 1+(4x) 2 U
w √
2 2

where N is the normalization constant,


 p 2 −1/2
1 + 1 + (4x)2
N = 2x2 +
 
 .
8

For general t/U , all u, v, w are finite and the ground state is clearly a linear superposition of singly and
doubly occupied states:
| ↑, ↓i − | ↓, ↑i
|ψgs i = u | ↑↓, 0i + v |0, ↑↓i + w √ .
2
√
In the limit x ≡ t/U ≪ 0, we get N = 2 + O(x2 ) and, hence,
    √   
u x 2x 0
√ √  x→0
  
v  = 2  √ x  2
 + O(x ) =  2x + O(x) −−−−→
 
0 .
1+ 1+(4x)2
w √ 1 1
2 2

This means that

t/U→0 | ↑, ↓i − | ↓, ↑i t
|ψgs i −−−−−→ √ +O ,
2 U
so the ground state consists of the singlet state of the two electrons on each site with opposite spin. Doubly
occupied states appear only as higher order corrections in t/U .

h) Combining the results of questions (e) and (h), when t/U ≪ 1 the low-energy sector comprises the following
energies and corresponding eigenstates:

4t2 | ↑, ↓i − | ↓, ↑i
E1 ≃ − , |ψ1 i = √ ,
U 2
E2 = 0, |ψ2 i = | ↑, ↑i,
E3 = 0, |ψ3 i = | ↓, ↓i,
| ↑, ↓i + | ↓, ↑i
E4 = 0, |ψ3 i = √ . (⋆)
2

On the other hand, consider the spectrum of the Heisenberg exchange Hamiltonian between two localized
spins,
ĤHeisenberg ≡ J Ŝ1 · Ŝ2 ,

where the magnitude of each spin is S1 = S2 = 1/2 (electrons). We may write it in terms of the total spin
Ŝtot ≡ Ŝ1 + Ŝ2 as
J 2
ĤHeisenberg = (Ŝtot − Ŝ1 − Ŝ2 ),
2
whose spectrum is given by

Jh i Jh 3i
Stot (Stot + 1) − S1 (S1 + 1) − S2 (S2 + 1) = Stot (Stot + 1) − ,
2 2 2
and, according to the theory of angular momentum addition, the total spin eigenvalue takes the values

13
 Stot ∈ {0, 1}. The spectrum and corresponding eigenstates |Stot , M i of ĤHeisenberg are then

3J | ↑, ↓i − | ↓, ↑i
E1 ≃ − , |ψ1 i = |Stot = 0, M = 0i = √ ,
4 2
J
E2 = , |ψ2 i = |Stot = 1, M = 1i = | ↑, ↑i,
4
J
E3 = , |ψ3 i = |Stot = 1, M = −1i = | ↓, ↓i,
4
J | ↑, ↓i + | ↓, ↑i
E4 = , |ψ3 i = |Stot = 1, M = 0i = √ . (†)
4 2

Note that, if we shift ĤHeisenberg by adding the constant −J/4,

 
′ J 1
ĤHeisenberg ≡ ĤHeisenberg − = J Ŝ1 · Ŝ2 − ,
4 4

we obtain exactly the spectrum and eigenstates (⋆) of the Hubbard Hamiltonian in the large-U limit, with
the identification J ≡ 4t2 /U . In other words, in the large-U limit, the low-energy subspace of the Hubbard
model maps exactly to that of the effective Heisenberg Hamiltonian
 
′ 1
ĤHeisenberg ≡ J Ŝ1 · Ŝ2 − ,
4

with exchange coupling given by J = 4t2 /U . This is telling us that, for large values of U/t (strong Coulomb
interaction), the low energy physics of the Hubbard model at half filling (Ne = Nc ) is essentially that of
localized spins (because hopping leads to double occupancy, which is suppressed in this limit) with an
effective antiferromagnetic exchange coupling. Furthermore, we explicitly find that J = 4t2 /U > 0. The
antiferromagnetic state allows lowering of the ground-state energy for any finite U/t in perturbation theory,
which is clear when we do second-order perturbation theory, treating T̂ as a perturbation to ĤU : having
neighboring spins anti-parallel to each other allows non-zero matrix elements of T̂ (the perturbation), which
inevitably lowers the ground state energy in second order.

Problem 3 (exchange operator)

When we discussed the hierarchy of terms arising from the Coulomb interaction between electrons
expressed in a basis of localized Wannier states, we saw that, right after the leading on-site Hubbard
term, ĤU , we have the so-called direct exchange term:
X (F )
X X (F )
X
Ĥex ≡ Jij c†is c†js′ cis′ cjs = − Jij P̂ij , P̂ij ≡ − c†is c†js′ cis′ cjs , (i 6= j).
i6=j s,s′ i6=j s,s′

a) Use the representation of spin operators given in eq. (2) to show that

1
P̂ij = 2Ŝi · Ŝj + ni nj .
2

b) Consider a N -particle state that contains a single electron with spin α at site i and a single

14
 electron with spin β at site j,

|ψi = c†1s1 c†2s2 · · · c†iα · · · c†jβ · · · c†N sN |0i ≡ |s1 , s2 , · · · , α, · · · , β, · · · , sN i.

Show that

P̂ij |s1 , s2 , · · · , α, · · · , β, · · · , sN i = |s1 , s2 , · · · , β, · · · , α, · · · , sN i,

which is the origin of the designation “exchange coupling” or “exchange interaction”.

Suggestion — It’s instructive to demonstrate this result in (b) using both representations of the operator
P̂ij : the first, in terms of the fermion operators; the second, in terms of spin operators given in (a). For
the second representation, expand Ŝi · Ŝj in angular momentum ladder operators and use their action on
angular momentum eigenstates, because |s1 , s2 , · · · , sN i are eigenstates common to all the Ŝi2 and Ŝiz .

Solution

a) Let us show that  

1 1 X
Ŝi · Ŝj = P̂ij − ni nj , P̂ij ≡ − c†is c†js′ cis′ cjs .
2 2
s,s′

Starting from the representation (2), we have (implicit summation):

1 †    1
Ŝi · Ŝj = ciα σαβ ciβ · c†jµ σµν cjν = (σαβ · σµν ) c†iα ciβ c†jµ cjν
 4 X m m
4

y use the property σαβ σµν = 2δαν δβµ − δαβ δµν
m
1
= (2δαν δβµ − δαβ δµν ) c†iα ciβ c†jµ cjν
4
1 † 1
= c ciβ c†jβ cjα − c†iα ciα c†jµ cjµ
2 iα 4
1 † † 1
= c ciβ cjβ cjα − ni nj
 2 iα 4

y anticommute to bring creation operators to the left (note that i 6= j)
1 1
= − c†iα c†jβ ciβ cjα − ni nj
 2 4

y rename dummy indices α, β; make summations explicit
1X † † 1
=− c c ′ cis′ cjs − ni nj
2 ′ is js 4
ss
1 1
= P̂ij − ni nj .
2 4

b) Approach 1 (fermion representation) — Note that

X X  
P̂ij ≡ − c†is c†js′ cis′ cjs |ψi = − c†is c†js′ cis′ cjs |ψi ,
s,s′ s,s′

and, since the question states there is (only) one electron at sites i and j, the action of the annihilation

15
operators leads to a state where both i and j are empty

(i) (j)
cis′ cjs |ψi = cis′ cjs |s1 , s2 , · · · , α , · · · , β , · · · , sN i,
(i) (j)
= δs′ α δsβ (−1)η |s1 , s2 , · · · , 2, · · · , 2 , · · · , sN i

where the factor (−1)η (η ∈ N0 ) accounts for the anticommutation with all the creation operators related
(i)
to the other particles in the system, and 2 means that site i is empty. The subsequent action of the two
creation operators in P̂ij yields
  (i) (j)
c†is c†js′ cis′ cjs |ψi = δs′ α δsβ (−1)η c†iβ c†jα |s1 , s2 , · · · , 2, · · · , 2 , · · · , sN i
(i) (j)
= δs′ α δsβ (−1)η (−1)η−1 |s1 , s2 , · · · , β , · · · , α , · · · , sN i
(i) (j)
= −δs′ α δsβ |s1 , s2 , · · · , β , · · · , α , · · · , sN i,

where the factor (−1)η (−1)η−1 = −1 takes into account that, since the particles are removed in the
sequence “j first, i second” and added back in the same order, the action of c†is c†js′ leaves an overall minus
sign (because when the first particle is created at j, there is no particle at i to anticommute, in contrast
to the earlier situation when cjs acted to annihilate it; so, there’s a net minus sign at the end). Therefore,
we conclude that

(i) (i) X (i) (i)

P̂ij | · · · , α , · · · , β , · · · i = − c†is c†js′ cis′ cjs | · · · , α , · · · , β , · · · i
s,s′
!
X (i) (j)
=− −δs′ α δsβ | · · · , β , · · · , α , · · · i
s,s′
(i) (j)
= | · · · , β , · · · , α , · · · i,

which means that the result of P̂ij is indeed an exchange of the two particles at i and j.

Approach 2 (spin representation) — In this case, we write

1
Ŝi · Ŝj = Ŝiz Ŝjz + (Ŝi+ Ŝj− + Ŝi− Ŝj+ ),
2
and recall from the theory of angular momentum that
p
Ŝi± |s1 · · · , mi , · · · sN i = s(s + 1) − mi (mi ± 1) |s1 · · · , mi ± 1, · · · sN i


y s = 1/2, mi = ±1/2
p
= 1/2 ∓ mi |s1 · · · , si ± 1, · · · sN i.

Hence, the action of the diagonal term is

Ŝiz Ŝjz | · · · , mi , · · · , mj , · · · i = mi mj | · · · , mi , · · · , mj , · · · i,

while
(i) (j) (i) (j)
Ŝi+ Ŝj− | · · · ↓ · · · ↑ · · · i = | · · · ↑ · · · ↓ · · · i,
(i) (j) (i) (j)
Ŝi− Ŝj+ | · · · ↑ · · · ↓ · · · i = | · · · ↓ · · · ↑ · · · i,

16
are the only non-zero terms involving the action of Ŝi± Ŝj∓ . Consequently,

(i) (j) 1 (i) (j)

Ŝi · Ŝj | · · · ↑ · · · ↑ · · · i = |··· ↑ ··· ↑ ···i
4
(i) (j) 1 (i) (j)
Ŝi · Ŝj | · · · ↓ · · · ↓ · · · i = | · · · ↓ · · · ↓ · · · i
4
(i) (j) 1 (i) (j) 1 (i) (j)
Ŝi · Ŝj | · · · ↑ · · · ↓ · · · i = | · · · ↓ · · · ↑ · · · i − |··· ↑ ··· ↓ ···i
2 4
(i) (j) 1 (i) (j) 1 (i) (j)
Ŝi · Ŝj | · · · ↓ · · · ↑ · · · i = | · · · ↑ · · · ↓ · · · i − | · · · ↓ · · · ↑ · · · i.
2 4
Therefore, if there is one electron at site i and another at site j, this result shows that

1 1
Ŝi · Ŝj = P̂ij − ,
2 4

where P̂ij is the pair exchange/permutation operator.

Important — Note that this relation between the pair of spin operators Ŝi · Ŝj and the permutation
operator P̂ij applies only to the case of spin 1/2 (it doesn’t work for higher spin magnitudes). Also, this
is only applicable when we have just one electron (fermion with spin 1/2) in site i and one electron in site
j; if site i, j or both are empty, the action of P̂ij clearly yields zero.

17