The Hubbard Model —Introduction and Selected Rigorous Results

arXiv:cond-mat/9512169v4 [cond-mat.str-el] 19 Dec 1997

Hal Tasaki1

Department of Physics, Gakushuin University, Mejiro, Toshima-ku, Tokyo 171, JAPAN

Abstract
The Hubbard model is a “highly oversimplified model” for electrons in a solid
which interact with each other through extremely short ranged repulsive (Coulomb)
interaction. The Hamiltonian of the Hubbard model consists of two pieces; Hhop
which describes quantum mechanical hopping of electrons, and Hint which describes
nonlinear repulsive interaction. Either Hhop or Hint alone is easy to analyze, and does
not favor any specific order. But their sum H = Hhop + Hint is believed to exhibit
various nontrivial phenomena including metal-insulator transition, antiferromagnetism,
ferrimagnetism, ferromagnetism, Tomonaga-Luttinger liquid, and superconductivity.
It is believed that we can find various interesting “universality classes” of strongly
interacting electron systems by studying the idealized Hubbard model.
In the present article we review some mathematically rigorous results on the Hub-
bard model which shed light on “physics” of this fascinating model. We mainly concen-
trate on magnetic properties of the model at its ground states. We discuss Lieb-Mattis
theorem on the absence of ferromagnetism in one dimension, Koma-Tasaki bounds on
decay of correlations at finite temperatures in two-dimensions, Yamanaka-Oshikawa-
Affleck theorem on low-lying excitations in one-dimension, Lieb’s important theorem
for half-filled model on a bipartite lattice, Kubo-Kishi bounds on the charge and su-
perconducting susceptibilities of half-filled models at finite temperatures, and three
rigorous examples of saturated ferromagnetism due to Nagaoka, Mielke, and Tasaki.
We have tried to make the article accessible to nonexperts by describing basic defini-
tions and elementary materials in detail.

Contents
1 Introduction 2

2 Hubbard Model 3
2.1 Definition of the Hubbard Model . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Some Physical Quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1
hal.tasaki@gakushuin.ac.jp, http://www.gakushuin.ac.jp/˜881791/

1
3 Basic Facts about the Model 7
3.1 Non-Interacting System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Non-Hopping System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Hubbard Model is Difficult, but it is Interesting . . . . . . . . . . . . . . . . 10

4 Results for Low Dimensional Models 11

4.1 Lieb-Mattis Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 Decay of Correlations at Finite Temperatures . . . . . . . . . . . . . . . . . 12
4.3 Yamanaka-Oshikawa-Affleck Theorem . . . . . . . . . . . . . . . . . . . . . . 12

5 Half-Filled Systems 14
5.1 Perturbation for U ≫ t . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
5.2 Lieb’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
5.3 Lieb’s Ferrimagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5.4 Kubo-Kishi Bounds on Susceptibilities at Finite Temperatures . . . . . . . . 19

6 Ferromagnetism 20
6.1 Instability of Ferromagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . 20
6.2 Toy Model with Two Electrons . . . . . . . . . . . . . . . . . . . . . . . . . 21
6.3 Nagaoka’s Ferromagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.4 Mielke’s Ferromagnetism and Flat-Band Ferromagnetism . . . . . . . . . . . 26
6.5 Ferromagnetism in a Non-Singular Hubbard model . . . . . . . . . . . . . . 29

1 Introduction
According to the textbook of Ashcroft and Mermin, the Hubbard model is “a highly over-
simplified model” for strongly interacting electrons in a solid. The Hubbard model is a kind
of minimum model which takes into account quantum mechanical motion of electrons in a
solid, and nonlinear repulsive interaction between electrons. There is little doubt that the
model is too simple to describe actual solids faithfully.
Nevertheless, the Hubbard model is one of the most important models in theoretical
physics. In spite of its simple definition, the Hubbard model is believed to exhibit various
interesting phenomena including metal-insulator transition, antiferromagnetism, ferrimag-
netism, ferromagnetism, Tomonaga-Luttinger liquid, and superconductivity. Serious theo-
retical studies have also revealed that to understand various properties of the Hubbard model
is a very difficult problem. We believe that in course of getting deeper understanding of the
Hubbard model, we will learn many new physical and mathematical techniques, concepts,
and ways of thinking. Perhaps a more important point comes from the idea of “universal-
ity.” We believe that nontrivial phenomena and mechanisms found in the idealized Hubbard

2
model can also be found in other systems in the same “universality class” as the idealized
model. The universality class is expected to be large and rich enough so that it contains
various realistic strongly interacting electron systems with complicated details which are
ignored in the idealized model.
The situation is very similar to that of the Ising model for classical spin systems. The
Ising model is too simple to be a realistic model of magnetic materials, but has turned out to
be extremely important and useful in developing various notions and techniques in statistical
physics of many degrees of freedom. Many important universality classes (of spin systems
and field theories) were discovered by studying the Ising model.
In the present article, we review some mathematically rigorous results2 known for the
Hubbard model. We shall concentrate ourselves mainly on magnetic properties of the model
at its ground state, i.e., at the zero temperature. We have also decided not to cover many
important rigorous and/or exact results in one-dimensional models based on the Bethe ansatz
solutions. Even with these restrictions, we do not try to cover all the existing rigorous
results. We recall that there is an excellent review article by Lieb [1] which covers wider
topics than we do here. As for the more restricted topics of Nagaoka’s ferromagnetism, flat-
band ferromagnetism, and some related topics, there is a separate review [2] which is more
detailed and elementary than the present one.

2 Hubbard Model
2.1 Definition of the Hubbard Model
We first give general definition of the Hubbard model3 . Let the lattice Λ be a collection
of sites x, y, . . .. Physically speaking, each lattice site corresponds to an atomic site in
a crystal. In the standard Hubbard model, one simplifies the situation considerably, and
assumes that each atom has only one electron orbit and the corresponding orbital state is
non-degenerate4 . Of course actual atoms can have more then one orbits (or bands) and
electrons in the corresponding states. The philosophy behind the model building is that
those electrons in other states do not play significant roles in low-energy physics that we are
interested in, and can be “forgotten” for the moment. See Figure 1.
By c†x,σ , we denote the operator which creates an electron with spin σ =↑, ↓ at site x ∈ Λ.
2
In order to reduce the number of references, we have decided not to include many important references
on the related topics which do not provide rigorous results.
3
The readers who are new to the filed are recommended to take a look at [2], which contains more careful
introduction to the Hubbard model.
4
Such a model is usually referred to as a single-band Hubbard model. This terminology is confusing since
such a model can possess more than one (single-electron) band depending on the lattice structure. Perhaps
“single-orbital Hubbard model” is a better terminology.

3
 a)

b)

c)

d)

Figure 1: A highly schematic figure which explain philosophy of tight-binding descriptions.

a) A single atom with multiple electrons in different orbits. b) When atoms get together
to form a solid, electrons in the black orbits become itinerant, while those in the light gray
orbits are still localized at the original atomic sites. Electrons in the gray orbits are mostly
localized around the atomic sites, but tunnel to nearby gray orbits with a non-negligible
probabilities. c) We only consider the electrons in the gray orbits, which are expected to
play essential roles in determining various low energy physics of the system. d) If the gray
orbit is non-degenerate, we get a lattice model in which electrons live on lattice sites and
hop from site to another.

4
The corresponding annihilation operator is cx,σ , and nx,σ = c†x,σ cx,σ is the number operator.
These fermion operators obey the canonical anticommutation relations
! "
c†x,σ , cy,τ = δx,y δσ,τ , (2.1)

and ! "
c†x,σ , c†y,τ = {cx,σ , cy,τ } = 0, (2.2)
where {A, B} = AB + BA.
By Φvac we denote the state without any electrons. We have cx,σ Φvac = 0 for any x ∈
Λ and σ =↑, ↓. The Hilbert space of the model is generated by the states obtained by
successively operating the creation operator c†x,σ with various x and σ onto the state Φvac .
Since the anticommutation relation (2.2) implies (c†x,σ )2 = 0, each lattice site can either be
vacant, occupied by an ↑ or ↓ electron, or occupied by both ↑ and ↓ electrons. The total
dimension of the Hilbert space is thus5 4|Λ| .
The Hamiltonian of the Hubbard model is most naturally represented as the sum of two
terms as
H = Hhop + Hint . (2.3)
The most general form of the hopping Hamiltonian Hhop is6
# #
Hhop = tx,y c†x,σ cy,σ . (2.4)
x,y∈Λ σ=↑,↓

The hopping amplitude tx,y = ty,x , which is assumed to be real, represents the quantum
mechanical amplitude that an electron hops from site x to y (or from y to x). When x = y,
the summand in (2.4) becomes tx,x c†x,σ cx,σ = tx,x nx,σ , which is nothing but a single-body
potential.
The interaction Hamiltonian Hint is written as
#
Hint = Ux nx,↑ nx,↓ , (2.5)
x∈Λ

where Ux > 0 is a constant. The Hamiltonian represents a nonlinear interaction which

raises the energy by Ux when two electrons occupy a single orbital state at x. Although
the original Coulomb interaction is long ranged, we have “oversimplified” the situation and
5
Throughout the present article we denote by |S| the number of elements in a set S.
6
The standard convention is to put a minus sign in front of the summation in (2.4), and to assume
tx,y ≥ 0. However it seems that there is no simple reason that the hopping amplitude should have such
signs. If the system is bipartite (See Definition 5.1), one can change the sign of all tx,y (x ̸= y) by performing
a gauge transformation c†x,σ → −c†x,σ for all x ∈ A.

5
took into account the strongest part out of the interaction7 . Another interpretation is that
the Coulomb interaction is screened by the electrons in different orbital states which we had
decided to forget.

2.2 Some Physical Quantities

We shall define some basic conserved quantities. The total number operator
#
N̂e = (nx,↑ + nx,↓ ) (2.6)
x∈Λ

commutes with the Hamiltonian H. Although there are some conserved quantities other than
N̂e , one usually discusses stationary states or equilibrium states of the system by keeping
the eigenvalue or the expectation value of N̂e constant8 . In the present article, we mostly
consider9 the Hilbert space in which the number operator N̂e has a fixed eigenvalue Ne . Since
each lattice site can have at most two electrons, we have 0 ≤ Ne ≤ 2|Λ|. The total electron
number Ne is the most fundamental parameter in the Hubbard model.
The spin operator Ŝx = (Ŝx(1) , Ŝx(2) , Ŝx(3) ) at site x is defined as
1 # †
Ŝx(α) = c (p(α) )σ,τ cx,σ , (2.7)
2 σ,τ =↑,↓ x,σ

for α = 1, 2, and 3, where p(α) are the Pauli matrices. The operators for the total spin of
the system are defined as # (α)
(α)
Ŝtot = Ŝx , (2.8)
x∈Λ
(α)
for α = 1, 2, and 3. The operator Ŝtot commute with both the hopping Hamiltonian Hhop
(2.4) and with the interaction Hamiltonian Hint (2.5). In other words, these Hamiltonians
are invariant under any global rotation in the spin space.
(α)
As the operators Ŝtot with α = 1, 2, 3 do not commute with each other, we follow the
convention in the theory of angular momenta, and simultaneously diagonalize the total spin
(3) $ (α) (3)
operators Ŝtot , (Ŝtot )2 = 3α=1 (Ŝtot )2 , and the Hamiltonian H. We denote by Stot and
7
There are many important works on various extended Hubbard models in which one takes into account
other short range interactions which arise from the original Coulomb interaction. See [3, 4, 5, 6] and many
references therein.
8
For example the total spin is also a conserved quantity. But we do not fix its eigenvalue or expectation
value, since the total spin is not definitely conserved in reality because there is an LS coupling and the actual
solids are not rotation invariant. The situation for N̂e is essentially different since the charge conservation
is an exact law. See Section 2.2 of [2].
9
Section 5.4 is the only exception.

6
 (3)
Stot (Stot + 1) the eigenvalues of Ŝtot and (Ŝtot )2 , respectively. For a given electron number
Ne , we let %
Ne /2 when 0 ≤ Ne ≤ |Λ|;
Smax = (2.9)
|Λ| − (Ne /2) when |Λ| ≤ Ne ≤ 2|Λ|.
Then the possible values of Stot are Stot = 0, 1, . . . , Smax (or Stot = 1/2, 3/2, . . . , Smax ).
When we discuss magnetism of the system, the most important issue is to determine the
value of Stot in the ground state(s). If the total spin of the ground state grows proportionally
to the number of sites |Λ| as we increase the size of Λ, we say that the system exhibits
ferromagnetism in a broad sense. This roughly means that the system behaves as a “magnet.”
If the total spin of the ground state(s) coincides with the maximum possible value Smax , we
say that the system exhibits saturated ferromagnetism.
The following quantity will be useful in the later analysis.

Definition 2.1 (The lowest energy for each Stot ) Fix the electron number Ne . For S =
0, 1, . . . , Smax (or S = 1/2, 3/2, . . . , Smax ), we denote by Emin (S) the lowest possible energy
among the states which satisfy N̂e Φ = Ne Φ and (Ŝtot )2 Φ = S(S + 1)Φ (i.e., Stot = S).

The appearance of saturated ferromagnetism is equivalent to have Emin (S) > Emin (Smax )
for any S such that S < Smax .

3 Basic Facts about the Model

In order to understand the meaning of the Hamiltonian of the Hubbard model, we discuss
physics we encounter in two limiting situations.

3.1 Non-Interacting System

Let us assume that the Coulomb interaction in Hint (2.5) satisfies Ux = 0 for any x ∈ Λ. Since
the remaining Hamiltonian H = Hhop (2.4) is a quadratic form in fermion operators, it can
be diagonalized easily (in principle). The single-electron Schrödinger equation corresponding
to the hopping Hamiltonian Hhop (2.4) is
#
tx,y ϕy = εϕx , (3.1)
y∈Λ

where ϕ = (ϕx )x∈Λ is a single-electron wave function, and ε is the single-electron energy
eigenvalue. We shall denote the eigenvalues and the eigenstates of (3.1) as εj and ϕ(j) =
(ϕ(j)
x )x∈Λ , respectively, where the index takes values j = 1, 2, . . . , |Λ|. We count the energy
levels with taking degeneracies into account, and order them as εj ≤ εj+1.

7
 Let us discuss a simple and standard example. Take a one dimensional lattice Λ =
{1, 2, . . . , N}, and impose a periodic boundary condition which identifies the site 1 with the
site N + 1. As for the hopping matrix elements, we set tx,x+1 = tx+1,x = −t, and tx,y = 0
otherwise. The corresponding Schrödinger equation (3.1) can be solved easily. By using the
wave number k = 2πn/N (with n = 0, ±1, ±2, . . . , ± {(N/2) − 1} , N/2), the eigenstates and
the eigenvalues can be written as N −1/2 exp[ikx] and ε(k) = −2t cos k, respectively. If one
makes a suitable correspondence between n and j = 1, 2, . . . , N, we get the desired energy
level εj .
We return to the general setting, and define fermion operators corresponding to the
eigenstates ϕ(j) = (ϕ(j) x )x∈Λ by # (j) †
a†j,σ = ϕx cx,σ . (3.2)
x∈Λ

By using the orthonormality of the set of eigenstates (ϕ(j) )j=1,2,...,|Λ| (we redefine the eigen-
states if they do not form an orthonormal set), one finds that the inverse transformation of
$|Λ|
(3.2) is cx,σ = j=1 ϕ(j)
x aj,σ . Substituting this into (2.4), and by using (3.1), we find that
Hhop can be diagonalized as
|Λ| |Λ|
# # # #
Hhop = εj a†j,σ aj,σ = εj ñj,σ . (3.3)
σ=↑,↓ j=1 σ=↑,↓ j=1

Here ñj,σ = a†j,σ aj,σ can be interpreted as the electron number operator for the j-th single-
electron eigenstate.
Let A, B be two arbitrary subsets of {1, 2, . . . , |Λ|} which satisfy |A| + |B| = Ne . By
using (3.3), we find that the state
⎛ ⎞⎛ ⎞
( (
ΦA,B = ⎝ a†j,↑ ⎠ ⎝ a†j,↓ ⎠ Φvac (3.4)
j∈A j∈B

is an eigenstate of H = Hhop and its energy eigenvalue is

# #
EA,B = εj + εj . (3.5)
j∈A j∈B

By choosing subsets A, B which minimize EA,B , we get ground state(s) of the non-interacting
model.
In particular if the corresponding single-electron energy eigenvalues are nondegenerate,
i.e., εj < εj+1, and Ne is even, the ground state of H = Hhop is unique and written as
⎛ ⎞
Ne /2
(
ΦGS = ⎝ a†j,↑ a†j,↓ ⎠ Φvac . (3.6)
j=1

8
Figure 2: Schematic picture of the ground state of a non-interacting many electron system.
The lowest Ne /2 single-electron energy levels are “filled” by both up spin and down spin
electrons. The state naturally exhibits paramagnetism known as Pauli paramagnetism.

This is nothing but the state obtained by “filling up” the low energy levels by up and down
spin electrons, as one learns in elementary quantum mechanics (Figure 2). It is easily verified
that the above state has a definite total spin Stot = 0. The ground state (3.6) exhibits no long
range order. A system with no magnetic ordering is usually said to exhibit paramagnetism10 .
In the simple example in one dimension, all the energy levels except ε(0) = −2t and
ε(π) = 2t are two-fold degenerate. In this case the ground state of H = Hhop may not
be unique for some values of Ne . However the degeneracy of the ground states is at most
four-fold, and the total spin of the ground states can take values in Stot = 0, 1/2, or 1. We
can conclude that the property of the ground state(s) is essentially the same as the models
without degeneracy. In general we can draw the same conclusion unless the single-electron
spectrum has a bulk degeneracy.
In a single-electron eigenstate of the example in one dimension, the electron is in a plane
wave state with a definite wave number k. The fact that the Hamiltonian H = Hhop is
diagonalized as (3.3) implies that the electrons behave as “waves” in this non-interacting
(Hubbard) model. The same comment applies to any translation invariant (Hubbard) model
with Ux = 0.

3.2 Non-Hopping System

Let us next assume that the hopping matrix elements in Hhop (2.4) satisfy tx,y = 0 for any
x, y ∈ Λ. Then the remaining Hamiltonian H = Hint (2.5) is already in a diagonal form. A
10
More precisely, this is true when one talks only about magnetism carried by electron spins. If one takes
into account magnetism induced by orbital motion of electrons, the system may exhibit diamagnetism.

9
general eigenstate can be written as
+ ,+ ,
( (
ΨX,Y = c†x,↑ c†x,↓ Φvac . (3.7)
x∈X x∈Y

Here X and Y are arbitrary subsets of Λ, and represent lattice sites which are occupied by
up-spin electrons and down-spin electrons, respectively. The total electron number in this
state is Ne = |X| + |Y |, and the energy eigenvalue is given by
#
EX,Y = Ux . (3.8)
x∈X∩Y

The ground state for a given electron number Ne can be constructed by choosing subsets
X, Y that minimize the energy EX,Y . When one has Ne ≤ |Λ|, one can always choose X and
Y such that X ∩ Y = ∅. Thus the ground state has energy equal to 0.
The ground states of the non-hopping Hubbard model has no magnetic long range order.
Again the system is paramagnetic. It is also clear (from the beginning) that the electrons
behave as “particles” in non-hopping models.

3.3 Hubbard Model is Difficult, but it is Interesting

We have investigated the properties of the two pieces Hhop and Hint in the Hubbard Hamilto-
nian. It turned out that both Hhop and Hint are easy to analyze. We also found that neither
of them favor any magnetic ordering.
We also observed, however, that electrons behave as “waves” in Hhop , while they behave
as “particles” in Hint . How do they behave in a system with the Hamiltonian which is a sum
of these totally different Hamiltonians? This is indeed a fascinating problem which is deeply
related to the wave-particle dualism in quantum physics. We might say that many of the
important models in many-body problems, including the ϕ4 quantum field theory and the
Kondo problem, are minimum models which take into account at the same time the wave-like
nature and the particle-like nature (through point-like nonlinear interaction) of matter.
From a technical point of view, the wave-particle dualism implies that the Hamiltonians
Hint are Hhop do not commute with each other. Even when each Hamiltonian is diagonal-
ized, it is still highly nontrivial (or impossible) to find the properties of their sum. Of course
mathematical difficulty does not automatically guarantee that the model is worth studying.
A really exciting thing about the Hubbard model is that, though the Hamiltonians Hhop and
Hint do not favor any nontrivial order, their sum H = Hhop + Hint is believed to generate
various nontrivial order including antiferromagnetism, ferromagnetism, and superconductiv-
ity. When we sum up the two innocent Hamiltonians Hhop and Hint , competition between
wave-like character and particle-like character (or between linearity and nonlinearity) takes
place, and one gets various interesting “physics.” To confirm this fascinating scenario is a
very challenging problem to theoretical and mathematical physicists.

10
4 Results for Low Dimensional Models
We discuss some theorems which are proved by using special natures of low-dimensional
systems.

4.1 Lieb-Mattis Theorem

Theorems discussed in the present and the next sections state that the Hubbard model
does not exhibit interesting long range order under some conditions. The main purpose of
studying an idealized model like the Hubbard model should be to show that some interesting
physics do take place. Results which say something does not take place may be regarded
as less exciting. But to have a definite knowledge that something does not happen under
some conditions is very useful and important even if our final goal is to show something does
happen.
The classical Lieb-Mattis theorem [7] states (among other things) that one can never have
ferromagnetism in one-dimensional Hubbard model with only nearest neighbor hoppings11 .
Theorem 4.1 (Lieb-Mattis theorem) Consider a Hubbard model on a one-dimensional
lattice Λ = {1, 2, . . . , N} with open boundary conditions. We assume that the hopping matrix
elements satisfy |tx,y | < ∞ when x = y, 0 < |tx,y | < ∞ when |x − y| = 1, and are vanishing
otherwise. We also assume |Ux | < ∞ for any x ∈ Λ. Then the quantity Emin (S) (see
Definition 2.1) satisfies the inequality
Emin (S) < Emin (S + 1), (4.1)
for any S = 0, 1, . . . , Smax − 1 (or S = 1/2, 3/2, . . . , Smax − 1).
One of the most important consequences of the Lieb-Mattis theorem is that any one-
dimensional Hubbard model in the above class has the total spin Stot = 0 (or Stot = 1/2)
in its ground state. One cannot conclude from this fact alone that the system exhibits
paramagnetism, but can conclude that there is no ferromagnetism.
Theorem 4.1 does not apply to models with periodic boundary conditions. But it seems
reasonable that boundary conditions do not change the essential physics provided that the
system is sufficiently large. If there exist hoppings to sites further than the nearest neighbor,
on the other hand, the story is totally different. We do not only find that the proof of
Theorem 4.1 fails, but we also find essentially new physics. See Section 6.5.
Theorem 4.1 is proved by noting that, in a suitable basis, the Hamiltonian is written as
a matrix whose nondiagonal elements are nonpositive, and by using the standard Perron-
Frobenius argument. Similar argument was used by Lieb and Mattis in their study of the
Heisenberg quantum spin system [8].
11
The present theorem appears in the Appendix of [7]. The main body of [7] treats interacting electron
systems in continuous spaces.

11
4.2 Decay of Correlations at Finite Temperatures
Among other rigorous results which show the absence of order is the extension by Ghosh [9]
of the wellknown theorem of Mermin and Wagner. Ghosh proved that the Hubbard model in
one- or two-dimensions does not exhibit symmetry breaking related to magnetic long range
order at any finite temperatures. By using the same method, one can also prove the absence
of superconducting U(1) symmetry breaking.
Koma and Tasaki proved essentially the same facts in terms of explicit upper bounds for
various correlation functions [10]. Among the results of [10] is the following.

Theorem 4.2 (Koma-Tasaki bounds for correlations) Consider an arbitrary Hubbard

model in one- or two-dimensions with finite range hoppings. Then there are constants α, γ,
and we have
-. 0
/ --
- † †
- c c cy,↑ cy,↓ |x − y|−αf (β) in d = 2;
- x,↑ x,↓ + H.c. -- ≤ (4.2)
β exp[−γf (β)|x − y|] in d = 1,

and 0
- -
- - |x − y|−αf (β) in d = 2;
-⟨Sx · Sy ⟩β - ≤ (4.3)
exp[−γf (β)|x − y|] in d = 1,
for sufficiently large |x−y|, where ⟨· · ·⟩β denotes the canonical average in the thermodynamic
limit at the inverse temperature β. Here f (β) is a decreasing function of β and behaves as
f (β) ≈ 1/β for β ≫ δ and f (β) ≈ (2/δ)| ln β| for β ≪ δ, where δ is a constant.

The bounds (4.2) and (4.3) establishes the widely accepted fact that there can be
no superconducting12 or magnetic long-range order at finite temperatures in one- or two-
dimensions. The method employed in [10], i.e., a combination of the McBryant-Spencer
method and the quantum mechanical global U(1) gauge invariance, is rather interesting. It
is amusing that only by using the U(1) symmetry which exists in any quantum mechanical
system, one gets upper bounds for correlations which are almost optimal at low temperatures
(especially in one-dimension).

4.3 Yamanaka-Oshikawa-Affleck Theorem

We discuss a recent important theorem by Yamanaka, Oshikawa, and Affleck [11, 12] about
low-lying excitations in general electron systems on a one-dimensional lattice. The theorem
is an extension of Lieb-Schultz-Mattis theorem for quantum spin chains. It can also be in-
terpreted as a nonperturbative version of Luttinger’s “theorem” restricted to one-dimension.
12
One can easily extend (4.2) to rule out the condensation of other types of electron pairs.

12
 We consider the Hubbard model13 on the one-dimensional lattice Λ = {1, 2, . . . , N} with
periodic boundary conditions. The model is characterized by positive integers R and P ,
which are the range of hopping and the period of the system, respectively. We assume
tx,y = 0 whenever |x − y| > R, and tx+P,y+P = tx,y , Ux+P = Ux for any x, y ∈ Λ. Under this
general assumption, we have the following.

Theorem 4.3 (Yamanaka-Oshikawa-Affleck theorem) Consider the infinite volume limit

N → ∞ with a fixed electron density ν = Ne /N. If P ν/2 is not an integer, then we have
one of the following two possibilities.
i) There is a symmetry breaking, and the infinite volume ground states are not unique.
ii) There is a gapless excitation above the infinite volume ground state.

In other words, the theorem rules out the third possibility;

iii) The infinite volume ground state is unique, and there is a finite gap above it.
Note that iii) does take place if P ν/2 is an integer and the system describes an innocent
insulator14 . In a non-interacting system, it is evident that iii) is impossible when P ν/2 is
not an integer, since there is a partially filled band. The above theorem guarantees that we
cannot change the situation by introducing strong interaction. This is of course far from
obvious.
The proof of Theorem 4.3 is based on the following explicit construction of a (trial) low-
lying excitation15 . Consider a system on a periodic chain of length N, and assume that the
$
ground state ΦGS is unique. We define the “twist” operator by U = exp[2πi N x=1 (x/N)nx,↑ ]
which introduces a gradual twist in the U(1) phase of the up-spin electron field, and consider
the trial excited state Ψ = UΦGS . It is not hard to show that ⟨Ψ, HΨ⟩ − EGS = O(1/L).
Thus Ψ contains a low-lying excited state provided it is orthogonal to the ground state. To
see the orthogonality, let TP be the translation by P , and assume that the ground state is
chosen as TP ΦGS = ΦGS . It is easily found that TP Ψ = eiπP ν Ψ, and hence the trial state Ψ
is orthogonal to ΦGS whenever P ν/2 is not an integer.
The above result also implies that, in the case ii), the gapless excitation Ψ has a crystal
momentum kΨ = πν. If we interpret this excitation as obtained from the ground state by
moving an electron at a “fermi point16 ” to the other fermi point, we find that the fermi
momentum kF satisfies kΨ = 2kF , and hence kF = πν/2. This is nothing but the fermi
momentum of the free system. Therefore the fermi momentum is not “renormalized” by
strong interaction among electrons. As far as we know, this is the only general rigorous
13
The theorem applies to a much larger class of lattice electron systems. It is especially meaningful when
applied to the Kondo lattice model.
14
Consider, for example, a two-band system with a band gap which has P = 2. Then, in a free system,
ν = 1 (the half-filling) corresponds to an insulator with a charge gap.
15
An extra care is needed to discuss infinite volume limits [12].
16
In a one-dimensional model, fermi surface (if any) becomes two “fermi points.”

13
result which gives precise meaning to fermi momentum in truly interacting many electron
systems.

5 Half-Filled Systems
A system in which the electron number Ne is identical to the number of sites |Λ| is said to
be half-filled, since the maximum possible value of Ne is 2|Λ|. The system becomes half-
filled if each atom provides one electron to the system. Thus the half-filled models represent
physically natural situations. Half-filled models have nice properties17 from mathematical
point of view as well, and there are some very nice rigorous results.

5.1 Perturbation for U ≫ t

Let us first look at the ground states of the non-hopping model with Hhop = 0. We here
assume that Ux > 0 for any x ∈ Λ. As we found in Section 3.2, one can choose X ∩ Y = ∅
in the state ΨX,Y to get a ground state with EX,Y = 0, provided that Ne ≤ |Λ|. Since we
have Ne = |Λ|, the assumption X ∩ Y = ∅ automatically implies X ∪ Y = Λ. Therefore the
ground state ΨX,Y (3.7) with X ∩ Y = ∅ can be written also as
+ ,
(
Ψσ = c†x,σ(x) Φvac , (5.1)
x∈Λ

where σ = (σ(x))x∈Λ is a collection of spin indices σ(x) =↑, ↓. By using the terminology of
spin systems, one can call σ a spin configuration. One can say that the degeneracy of the
ground states (5.1) precisely corresponds to all the possible spin configurations.
Let us take into account effects of nonvanishing Hhop by using a simple-minded perturba-
$ $
tion theory. As the diagonal elements of Hhop , i.e., x,σ tx,x c†x,σ cx,σ = x tx,x (nx,↑ +nx,↓ ), only
shift the energy of the states Ψσ by a constant amount (independent of σ), they can be omit-
1 $ †
ted in the lowest order perturbation calculation. Let us denote by H hop = x̸=y,σ tx,y cx,σ cy,σ
the off-diagonal part of Hhop . By operating H 1
hop once onto Ψσ , an electron moves, and we
get a state with one vacant site and one doubly occupied site. The resulting state is not a
ground state of Hint . We thus find that the lowest order contribution from this perturbation
theory comes from the second order.
Figure 3 shows a process that is taken into account in the second order perturbation
theory. The electron at site x hops to site y with the transition amplitude tx,y , and generates
a new state with extra energy Uy . Then one of the two electrons at site y will hop back to
17
Some half-filled models can be mapped onto a Hubbard model with attractive interaction via a partial
hole-particle transformation. This fact plays crucial role in the proof of Lieb’s theorem.

14
 x y x y x y
Figure 3: When electrons hop twice, spins on sites x and y may be exchanged. This is the
ultimate origin of antiferromagnetic nature in the half-filled Hubbard model.

site x, and we recover one of the ground states. In this process, spins at the sites x and y
may be exchanged as Figure 3 shows. The hopping between sites x and y is inhibited by the
Pauli principle if the electronic spins on these two sites are pointing in the same direction.
We find that this second order perturbation process lowers the energy of states in which the
spins at sites x and y are not pointing in the same direction (or more precisely, the states in
which the total spin is vanishing).
Let us rederive this result in a more formal manner. Let P0 be the projection operator
onto the subspace spanned by the states Ψσ (5.1) with all the possible σ. That the first
order perturbation has no contribution can be read off from the fact that P0 H 1 P = 0.
hop 0
To find out how the degeneracy in the (unperturbed) ground states (5.1) is lifted can be
determined by the effective Hamiltonian

1 1 1
Heff = −P0 H hop Hhop P0
⎧
Hint ⎫
⎨ # 5 6⎬
1
= P0 Jx,y Ŝx · Ŝy − P0 . (5.2)
⎩ 4 ⎭
x,y∈Λ

Here the exchange interaction parameter is given by Jx,y = {(tx,y )2 /Ux } + {(tx,y )2 /Uy }.
Note that (5.2) is nothing but the Hamiltonian of the S = 1/2 antiferromagnetic quantum
Heisenberg spin system. This suggests that the low-energy behavior of the half-filled Hubbard
model is well described by the Heisenberg antiferromagnets when Ux are much larger than
tx,y .

5.2 Lieb’s Theorem

In 1989, Lieb proved an important and fundamental theorem for the half-filled Hubbard
model. The theorem provides, among other things, partial support to the conjecture about
the similarity of the half-filled Hubbard model and the Heisenberg antiferromagnets. Let us
first introduce the notion of bipartiteness.

Definition 5.1 (Bipartiteness) Consider a Hubbard model (or other tight-binding electron
model) on a lattice Λ with hopping matrix elements (tx,y )x,y∈Λ . The system is said to be

15
bipartite if the lattice Λ can be decomposed into a disjoint union of two sublattices as Λ =
A ∪ B (with A ∩ B = ∅), and it holds that tx,y = 0 whenever x, y ∈ A or x, y ∈ B. In other
words, only hoppings between different sublattices are allowed.

Then Lieb’s theorem [13] for the repulsive Hubbard model is as follows.

Theorem 5.2 (Lieb’s theorem) Consider a bipartite Hubbard model. We assume |Λ| is
even, and the whole Λ is connected18 through nonvanishing tx,y . We also assume Ux = U > 0
for any x ∈ Λ. Then the ground states of the model are nondegenerate apart from the trivial
spin degeneracy19 , and have total spin Stot = ||A| − |B||/2.

The total spin Stot of the ground state determined in the theorem is exactly the same
as that of the ground state(s) of the corresponding Heisenberg antiferromagnet on the same
lattice. In fact the conclusion of the theorem is quite similar to that of the Lieb-Mattis
theorem [8] for Heisenberg antiferromagnets. However the straightforward Perron-Frobenius
argument used in the proof of the latter theorem does not apply to the Hubbard model
except in one dimension. (See 4.) This is not only a technical difficulty, but is a consequence
of the important fact that quantum mechanical processes allowed in the Hubbard model
are in general much richer and more complex than those in the Heisenberg model. Lieb’s
proof is compactly presented in a Letter, but is deep and elegant. The proof again makes
use of a kind of Perron-Frobenius argument, but is based on an interesting technique called
spin-space reflection positivity.
Lieb’s theorem is valid for any value of Coulomb repulsion U, only provided that it is
positive. It is quite likely that physical properties of the Hubbard model are drastically
different in the weak coupling region with small U and in the strong coupling region with
large U. It is very surprising and interesting that a single proof of Lieb’s covers the whole
range with U > 0 and clarifies the basic properties of the ground states.
It should be noted, however, that the knowledge of the total spin of the ground states
in finite volume does not necessarily determine the properties of the ground states in the
corresponding infinite system. When two sublattices have the same number of sites as
|A| = |B|, for example, one knows that finite volume ground state is unique and has Stot = 0.
Although one might well conclude that the system has no long range order in its ground
states, this is not true. It is certainly possible that infinite volume ground states exhibit
long range order and symmetry breaking (such as antiferromagnetism of superconductivity)
even when finite volume ground state is unique and symmetric. (See, for example, [14].)
18
More precisely, for any x, y ∈ Λ, one can find a sequence of sites x0 , x1 , . . ., xN with x0 = x, xN = y,
and txi ,xi+1 ̸= 0 for i = 0, 1, . . . , N − 1.
19
In any quantum mechanical system with a rotation invariant Hamiltonian, an eigenstate of the Hamil-
tonian with the angular momentum J is always (2J + 1)-fold degenerate.

16
By knowing any finite volume ground state has Stot = 0, however, one can rule out the
possibility of ferromagnetism.
By using Lieb’s results in [13], one gets some information about excited states. For
example, by combining Theorem 1 in [13] with the method of [8], one can easily prove the
inequality20
Emin (S) < Emin (S + 1) (5.3)
for any ||A| − |B||/2 ≤ S ≤ (|Λ|/2) − 1.
Another theorem which suggests the similarity between the half-filled Hubbard model and
the Heisenberg antiferromagnets is the following, proved by Shen, Qiu, and Tian [15, 16] by
extending Lieb’s method.

Theorem 5.3 (Explicit signs of correlations) Assume the conditions for Theorem 5.2.
If we denote by ΦGS the ground state of the model, we have the inequalities
. /%>
0 when x, y ∈ A, or x, y ∈ B;
ΦGS , Ŝx · Ŝy ΦGS (5.4)
<0 when x ∈ A, y ∈ B, or x ∈ B, y ∈ A,
were ⟨·, ·⟩ denotes the quantum mechanical inner product.

We see that spins on different sublattices have negative correlations, indicating antifer-
romagnetic tendency.
The S = 1/2 Heisenberg antiferromagnet on the cubic lattice, for example, is proved
to exhibit an antiferromagnetic long range order at sufficiently low temperatures or in the
ground states [17, 18]. It is likely that the same statements hold for the half-filled Hubbard
model with sufficiently large U. But, for the moment, there are no methods or ideas which
are useful in proving the conjecture. To extend the powerful (but not very natural) method
of [17, 18] based on the (spatial) reflection positivity seems hopeless.
In the Hubbard model and related models at half-filling, there have been proved several
interesting general results. Among the recent examples are the uniform density theorem [19],
the solution of the flux phase problem [20, 21], and the stability of the Peierls instability
[22].

5.3 Lieb’s Ferrimagnetism

A very important corollary of Lieb’s theorem 5.2 is that the half-filled Hubbard models
on asymmetric bipartite lattices universally exhibit a kind of ferromagnetism (in the broad
sense), or more precisely, ferrimagnetism [13].
Take, for example, the so called CuO lattice in Figure 4. The lattice can be decomposed
into two sublattices distinguished by black sites and white sites. When the black sites form
20
I wish to thank Shun-Qing Shen and Elliott Lieb for discussions related to this corollary.

17
Figure 4: An example (the so called CuO lattice) of a bipartite lattice in which the number
of sites in two sublattices are different. Lieb’s theorem implies that the half-filled Hubbard
model defined on this lattice exhibits ferrimagnetism.

a square lattice with side L, there are L2 black sites and 2L2 white sites. We define the
Hubbard model on this lattice, and put nonvanishing hopping tx,y on each bond in the
lattice, and put Coulomb interaction U > 0 on each site. Then Lieb’s theorem implies that
the ground state of this Hubbard model has total spin Stot = ||A| − |B||/2 = L2 /2. Since
the total spin magnetic moment of the system is proportional to the number of lattice sites
3L2 , we conclude that the model exhibits ferromagnetism in the broad sense.
Of course the present ferromagnetism is not a saturated ferromagnetism in which all the
spins in the system completely align with each other. As the inequality (5.4) suggests, spins
on neighboring sites have tendency to point in the opposite direction. But the big difference
in the numbers of sites in the sublattices cause the system to possess bulk magnetic moment.
Such a magnetic ordering is usually called ferrimagnetism21 .
One can similarly construct Hubbard models which exhibit ferrimagnetism on any bipar-
tite lattice in which the difference in the number of sites in two sublattices is proportional
to the system size. The value of U > 0 is again arbitrary, so Lieb’s ferrimagnetism covers
surprisingly general class of models including weakly coupled ones as well as strongly coupled
ones.
If one recalls the conclusion of Section 3.1 that systems with U = 0 exhibit paramag-
netism, one might feel it somehow contradicting that the above ferrimagnetism appears for
arbitrarily small U > 0. This is one of the special features of Lieb’s ferrimagnetism. In
the single-electron Schrödinger equation corresponding the Hubbard model on Figure 4, for
example, the eigenstates for the eigenvalue ε = 0 are L2 -fold degenerate. (The eigenvalue
21
It is also possible to consider order parameters to see the order is indeed ferrimagnetic [15].

18
ε = 0 is at the middle of the single-electron spectrum.) Consequently the ground states of
the half-filled (Ne = |Λ| = 3L2 ) system with U = 0 are highly degenerate, and the total spin
can take values in Stot = 0, 1, . . . , L2 /2. The role of Coulomb interaction U is to lift this
degeneracy, and select states with the largest magnetic moment as ground states.

5.4 Kubo-Kishi Bounds on Susceptibilities at Finite Tempera-

tures
A theorem which can be regarded as a finite temperature version of Lieb’s theorem was
proved by Kubo and Kishi [23]. It deals with the charge susceptibility and the on-site
pairing (superconducting) susceptibility in a half-filled system at finite temperatures.
We define the thermodynamic function22 J corresponding to the grand canonical ensemble
by

J(β, µ, (γx)x∈Λ , (ηx )x∈Λ ) =

⎡ ⎛ ⎞⎤
1 ⎢ ⎜ # # ⎟⎥
= − log Tr exp ⎢
⎣−β ⎜
⎝ H − µN̂e − γx nx,σ − ηx (c†x,↑ c†x,↓ + cx,↓ cx,↑)⎟ ⎥
⎠⎦ , (5.5)
β x∈Λ x∈Λ
σ=↑,↓

where β and µ are the inverse temperature and the chemical potential, respectively, and the
trace is taken over the Hilbert spaces with all the possible electron numbers. We added to
the Hamiltonian two fictitious external fields (γx )x∈Λ and (ηx )x∈Λ to test for the possible
charge ordering and superconducting ordering, respectively.
We define the charge susceptibility χc and the on-site pairing susceptibility χp by
-
∂ ∂ -
χcq (β, µ) =− J(β, µ, (γx), (ηx ))-- ≥ 0, (5.6)
∂γ̃q ∂γ̃−q -
(γx )=(ηx )=0

and -
∂ ∂ -
-
χpq (β, µ) =− J(β, µ, (γx), (ηx ))- ≥ 0. (5.7)
∂ η̃q ∂ η̃−q -
(γx )=(ηx )=0

The Fourier transformation of the external fields are

# #
γ̃q = |Λ|−1/2 γx eiq·x , η̃q = |Λ|−1/2 ηx eiq·x , (5.8)
x∈Λ x∈Λ

where q is a wave vector corresponding to the lattice Λ (which we assume to have a periodic
structure).
Then the Kubo-Kishi theorem states that
22
We have J = −pV = F − G.

19
Theorem 5.4 Consider any bipartite (see Definition 5.1) Hubbard model with Ux = U > 0
for any x ∈ Λ. Then for any β > 0 and for any wave vector q, we have
1 2
χcq (β, U/2) ≤ , and χpq (β, U/2) ≤ . (5.9)
U U
Note that the choice µ = U/2 corresponds to the half-filling. The theorem states that
the charge and the on-site paring susceptibilities for any wave vector q are finite in a half-
filled model at finite temperatures. This means that the model does not exhibit any CDW
ordering or superconducting ordering.

6 Ferromagnetism
Ferromagnetism, where almost all of the spins in the system align in the same direction, is
a remarkable phenomenon. Standard theories about the origin of ferromagnetism have been
the Heisenberg’s exchange interaction picture, and the Stoner criterion derived from the
Hartree-Fock approximation for band electrons. But there have been serious doubts if these
theories really explain the appearance of ferromagnetism in a system of electrons interacting
via spin-independent Coulomb interaction. One of the motivations to study the Hubbard
model was to understand the origin of ferromagnetism in an idealized situation.
As we have seen in the previous section, half-filled models have tendency towards antifer-
romagnetism. In this section we shall concentrate on systems in which the electron numbers
deviate from half-filling.

6.1 Instability of Ferromagnetism

To see that ferromagnetism is indeed a delicate phenomenon, we discuss two elementary
results which show that the Hubbard model with certain conditions does not exhibit ferro-
magnetism23 .
The following theorem states that there can be no saturated ferromagnetism if the
Coulomb interaction U is too small in a system with a “healthy” single-electron spectrum.

Theorem 6.1 (Impossibility of ferromagnetism for small U) Let {εj }j=1,...,N denote
the single-electron energy eigenvalues with εj ≤ εj+1 as in Section 3.1. If 0 ≤ U < εNe − ε1 ,
we have24
Emin (Smax − 1) < Emin (Smax ). (6.1)
Thus the ground state of the model does not have Stot = Smax .
23
Detailed proofs of the results in the present section can be found in [2].
24
Note that the fermi energy εNe − ε1 is an intensive quantity.

20
 When the density of electrons is very low, it is expected that the chance of electrons to
collide with each other becomes very small. It is likely that the model is close to an ideal
gas no matter how strong the interaction is, and there is no ferromagnetism.
This naive guess is easily justified for “healthy” models in dimensions three (or higher).
The dimensionality of the lattice is taken into account by assuming that there are positive
constants c, ν0 , and d, and the single electron energy levels satisfy
+ ,2/d
n−1
εn − ε1 ≥ c , (6.2)
|Λ|

for any n such that n/|Λ| ≤ ν0 . Note that the right-hand side represents the n dependence
of energy levels in an usual d-dimensional quantum mechanical system. Then we have the
following theorem due to Pieri, Daul, Baeriswyl, Dzierzawa, and Fazekas [24].

Theorem 6.2 (Impossibility of ferromagnetism at low densities) Suppose we have Hhop

satisfying (6.2) with positive c, ν0 , and d > 2. Then there exists a constant ν1 > 0, and the
same conclusion as Theorem 6.1 holds for any U ≥ 0 if Ne /|Λ| ≤ ν1 holds.

That we have a restriction on dimensionality in Theorem 6.2 is not merely technical. In

a one-dimensional system, moving electrons must eventually collide with each other for an
obvious geometric reason. Thus a one-dimensional model cannot be regarded as close to
ideal no matter how low the electron density is. We do not know whether the inapplicability
of the theorem to d = 2 systems is physically meaningful or not.

6.2 Toy Model with Two Electrons

As a starting point of our study of ferromagnetism, we consider a toy model with two
electrons on a small lattice. Interestingly enough, some essential features of ferromagnetism
found in many-electron systems (that we will discuss later in this section) are already present
in the toy model.
The smallest possible model in which we can discuss electron interaction and which is
away from half-filling is that with two electrons on a lattice with three sites. Consider the
lattice Λ = {1, 2, 3}, and put one electron with σ =↑ and one with σ =↓. The hopping
matrix is defined by t1,2 = t2,3 = t′ , and t1,3 = t. Note that there are two kinds of hoppings
t and t′ . Since the sign of t′ can be changed by the gauge transformation c2,σ → −c2,σ , we
shall fix t′ > 0. Figure 5 shows the lattice and the hopping. For simplicity, we assume there
is only one kind of interaction, and set U1 = U2 = U3 = U ≥ 0. We have Smax = 1 because
Ne = 2. Therefore we can say that there appears saturated ferromagnetism if the ground
state has Stot = 1, i.e., if it is a part of a spin-triplet.

21
 t
t' t'
1 2 3

Figure 5: The lattice and the hopping of the toy model. By considering the system with two
electrons on this lattice, we can observe some very important aspects of ferromagnetism in
the Hubbard model.

Φ13

t' t'
Φ12 Φ23
t t
Φ32 Φ21

t' t'
Φ31
Figure 6: Allowed states and transition amplitudes in the toy model with U = ∞. The total
spin of the ground states can be easily read off from this diagram.

22
 Let us take the limit U → ∞, in which the effect of interaction becomes most drastic,
and consider only those states with finite energies. This is equivalent to consider only states
in which two electrons never occupy a same site. There are six states which satisfy the
constraint, and they can be written as Φx,y = c†x,↑ c†y,↓ Φvac where x, y = 1, 2, 3, and x ̸= y.
Transition amplitudes between these states are shown in Figure 6. We find that the problem
is equivalent to that of a quantum mechanical particle hopping around on a ring consisting
of six sites. The basic structure of the ground state can be determined by the standard
Perron-Frobenius sign convention25 . The ground state for t < 0 is written as
(t<0)
ΦGS = Φ1,2 + Φ3,2 − α(t, t′ )Φ3,1 + Φ2,1 + Φ2,3 − α(t, t′ )Φ1,3 , (6.3)

and that for t > 0 as

(t>0)
ΦGS = Φ1,2 − Φ3,2 + β(t, t′ )Φ3,1 − Φ2,1 + Φ2,3 − β(t, t′ )Φ1,3 , (6.4)

where α(t, t′ ) and β(t, t′ ) are positive functions of t and t′ .

To find the total spin of these states, it suffices to concentrate on two lattice sites, say
(t<0) (t>0)
sites 1 and 2, and note that ΦGS = Φ1,2 + Φ2,1 + · · ·, and ΦGS = Φ1,2 − Φ2,1 + · · ·. It
(t<0) (t>0)
immediately follows that26 ΦGS has Stot = 0, and ΦGS has Stot = 1. A ferromagnetic
coupling is generated when t > 0!
Let us look at the mechanism which generates the ferromagnetism. The states Φ1,2 and
Φ2,1 can be found in the upper left and and the lower right, respectively, in the diagram of
Figure 6. By starting from Φ1,2 and following the possible transitions, one reaches the state
Φ2,1 . In other words, electrons hop around in the lattice, and the spins on sites 1 and 2 are
“exchanged.” When t > 0, the quantum mechanical amplitude associated with the exchange
process generates the superposition of the two states which precisely yields ferromagnetism.
Let us briefly look at the cases with finite U. In Figure 7, we plotted Emin (0) and Emin (1)
for the toy model with t = t′ /2 as functions of U. (See Definition 2.1.) As is suggested by
the result in the U → ∞ limit, we have ferromagnetism in the sense that Emin (0) > Emin (1)
when U is sufficiently large. A level crossing takes place at finite U, and the system is
no longer ferromagnetic for small U. Even in the simplest toy model, ferromagnetism is a
“nonperturbative” phenomenon which takes place only when U is sufficiently large.
The only exception is the case with t = t′ . See Figure 8. For this parameter value, the
ground states are degenerate in spin when U = 0. Ferromagnetic state is the only ground
state for U > 0.
25
If the transition amplitude between two states is negative (resp., positive), one superposes the two states
with the same (resp., opposite) signs.
26
A quick way to find the total spin of the state Φ1,2 + Φ2,1 is to rewrite the state in the “spin language”
as Φ1,2 + Φ2,1 = c†1,↑ c†2,↓ Φvac + c†2,↑ c†1,↓ Φvac = c†1,↑ c†2,↓ Φvac − c†1,↓ c†2,↑ Φvac = |↑⟩1 |↓⟩2 − |↓⟩1 |↑⟩2 , and use the
standard knowledge about the addition of angular momenta. One can easily convince oneself that the state
has Stot = 0.

23
 E / t'
–1.6

2 4 U / t'
–2.2

Figure 7: The U dependence of Emin (0) (gray curve) and Emin (1) (black line) in the toy
model with t = t′ /2. We have ferromagnetism in the sense that Emin (0) > Emin (1) when U
is sufficiently large. We find that ferromagnetism is a “nonperturbative” phenomenon.

E / t'
–1.4

–1.8

2 4 U / t'

Figure 8: The U dependence of Emin (0) (gray curve) and Emin (1) (black line) in the toy model
with t = t′ . Only for this special parameter, we have ferromagnetism Emin (0) > Emin (1) for
any value of U > 0. One can regard this case as the simplest example of the flat-band
ferromagnetism that we will discuss in Section 6.4.

24
Figure 9: Schematic picture about the origin of Nagaoka’s ferromagnetism. When the hole
hops around the lattice, the spin configuration is changed. For a model with tx,y ≥ 0, the
hole motion produces a precise linear combination of various spin configurations which leads
to ferromagnetic state.

From Figures 7 and 8, we find that the energy Emin (1) of ferromagnetic states is indepen-
dent of U. As we see in the following, this is a general property of ferromagnetic eigenstates
in the Hubbard model. An arbitrary state Ψ which has total spin Stot = Smax can be written
as a superposition of states which are obtained by rotating the state Ψ B which consists only
B
of ↑ spin electrons. If one operates the interaction Hamiltonian Hint (2.5) onto the state Ψ,
one has Hint ΨB = 0 because n Ψ B = 0. Since the interaction Hamiltonian (2.5) is invariant
x,↓
under rotation in spin space, we have shown that Hint Ψ = 0. One might say that states with
saturated magnetization do not feel Hubbard type interaction at all. This is one of conve-
nient (but “oversimplified”) features in discussing saturated ferromagnetism in the Hubbard
model.

6.3 Nagaoka’s Ferromagnetism

The transitions between the states in Figure 6 are generated by hoppings of electrons. One
can also regard that the transitions are caused by hoppings of a single hole27 , which is
the site without electrons. At least in the limit U → ∞, one can say that the origin of
ferromagnetism in the toy model is the motion of a single hole, which mix up various spin
configurations with proper signs.
As Nagaoka [25] demonstrated rigorously, there is a class of many electron models in
which saturated ferromagnetism is generated by exactly the same mechanism. See Figure 9.
27
This is different from the notion of hole in the usual band theory.

25
Nagaoka’s theorem (in the extended form of [26] whose complete proof can be found in [2])
is as follows.

Theorem 6.3 (Nagaoka’s ferromagnetism) Take an arbitrary finite lattice Λ, and as-
sume that28 tx,y ≥ 0 for any x ̸= y, and Ux = ∞ for any x ∈ Λ. We fix the electron number
as Ne = |Λ| − 1. Then among the ground states of the model, there exist states with total
spin Stot = Smax (= Ne /2). If the system further satisfies the connectivity condition, then the
ground states have Stot = Smax (= Ne /2) and are nondegenerate apart from the trivial spin
degeneracy.

The connectivity condition is a simple condition which holds on most of the lattices in
two or higher dimensions, including the square lattice, the triangular lattice, or the cubic
lattice. To be precise the condition requires that “by starting from any electron configuration
on the lattice and by moving around the hole along nonvanishing tx,y , one can get any other
electron configuration.”
Thouless also reached a similar conclusion [27], but Nagaoka’s treatment covers a larger
class of models including non-bipartite systems. The proof of Nagaoka’s theorem (especially
the recent proof in [26]) is surprisingly simple. It essentially uses the Perron-Frobenius
argument exactly the same as that we used in Section 6.2 to determine the total spin of the
ground state of the toy model.
The requirements that U should be infinitely large and there should be exactly one hole
are indeed rather pathological. Nevertheless, the theorem is very important since it showed
for the first time in a rigorous manner that quantum mechanical motion of electrons and
strong Coulomb repulsion can generate ferromagnetism. The conclusion that the system
which has one less electron than the half-filled model exhibits ferromagnetism is indeed
surprising. This is a very nice example which demonstrates that strongly interacting electron
systems produce very rich physics.
It is desirable to extend Nagaoka’s ferromagnetism to systems with a finite U and with
a finite density of holes. Although more than thirty years have passed since Nagaoka’s and
Thouless’s papers, it is still not known if such extensions are possible. There are, however,
considerable amount of rigorous works which establish that saturated ferromagnetism does
not take place in certain situations. See, for example, [28, 29, 30, 31, 32, 33].

6.4 Mielke’s Ferromagnetism and Flat-Band Ferromagnetism

Let us once again look at the toy model of section 6.2. As is shown in Figure 8, the
ground state of the model exhibits ferromagnetism for any U > 0 for the special choice of
28
As we noted in Section 2.1, this sign of tx,y is opposite from the “standard” choice. In bipartite systems
(such as those on the square lattice or the cubic lattice with nearest neighbor hoppings), one can change the
sign of tx,y by a gauge transformation.

26
Figure 10: The Hubbard model on the kagomé lattice is a typical example which exhibits
flat-band ferromagnetism. “Kagomé” is a Japanese word for a pattern of woven bamboo of
baskets.

the parameters t = t′ > 0. For this choice of parameters, the energy eigenvalues of the
corresponding Schrödinger equation are ε1 = ε2 = −t′ , and ε3 = 2t′ . The single-electron
ground states are doubly degenerate. As a consequence, the ground states of the two electron
system with U = 0 are also degenerate, and can have Stot = 0 or Stot = 1. The degeneracy
is lifted for U > 0, and the ferromagnetic state is “selected” as the true and unique ground
state. It is crucial here that the dimension of the degeneracy in the single-electron ground
states (which is two) is the same as the electron number Ne = 2.
Mielke [34] showed that there is a class of Hubbard models with many electrons which
show saturated ferromagnetism through a somewhat similar mechanism. Take, for example,
the kagomé lattice of Figure 10, and define a Hubbard model on it by setting tx,y = t > 0 for
neighboring sites x, y, tx,y = 0 for other situations, and Ux = U ≥ 0 for any x ∈ Λ. It is worth
mentioning that the kagomé lattice of Figure 10 can be regarded as constructed by putting
together many copies of the lattice used in the toy model (Figure 5). The energy eigenvalues
of the corresponding Schrödinger equation can be shown to satisfy ε1 = ε2 = · · · = εM = −2t,
and εj > −2t for j > M. Here the dimension M of the degeneracy of the single-electron
ground states is given by M = (|Λ|/3) + 1, and is proportional to the lattice size.
We shall fix the electron number as Ne = M, the same as the dimension of the degeneracy.
Let us consider the case with U = 0 first. Let A and B be arbitrary subsets of
{1, 2, . . . , Ne } which satisfy |A| + |B| = Ne , and consider the state ΦA,B (3.4) obtained by
creating the corresponding single-electron eigenstates. In the present model on the kagomé
lattice, the fermion operator a†j,σ (3.2) creates one of the single-electron ground states with
the energy ε = −2t. This means that we have Hhop ΦA,B = −2tNe ΦA,B for arbitrary choice

27
of A and B, and hence ΦA,B is a ground state of H = Hhop . We find that the ground states
are highly degenerate, and can have Stot = 0, 1, . . . , Smax = M/2 (or Stot = 1/2, . . . , Smax ).
What is the effect of nonvanishing Coulomb interaction U in such a situation? Let us
denote by Φ↑ the state obtained by setting A = {1, 2, . . . , Ne } and B = ∅ in ΦA,B . Of course
Φ↑ is one of the ground states of Hhop . As we discussed at the end of Section 6.2, the state Φ↑
which consists only of ↑ spin electrons “does not feel” Hubbard type Coulomb interaction.
This means that we have Hint Φ↑ = 0. Since 0 is the minimum possible eigenvalue of Hint ,
we find that the state Φ↑ is a ground state of the total HamiltonianH = Hhop + Hint for any
U > 0.
These are all simple observations. A really interesting problem is whether there can be
ground states other than Φ↑ when U > 0. The following theorem due to Mielke shows that
the ferromagnetic state is indeed “selected” as the true ground state exactly as in the toy
model.

Theorem 6.4 (Mielke’s flat-band ferromagnetism) Consider the Hubbard model on the
kagomé lattice described above. For any U > 0, the ground states have Stot = Smax (= M/2)
and are nondegenerate apart from the trivial spin degeneracy.

Mielke [35] also extended his results to the situation where the electron density29 Ne /|Λ|
is less than 1/3 but close to 1/3.
In Mielke’s work, it was proved for the first time that the Hubbard model with finite
U can exhibit saturated ferromagnetism. The model is very simple, and the result is very
important. As far as the author knows, there had been no discussions about the possibility
of ferromagnetism in the Hubbard model on the kagomé lattice. Mielke’s work is not only
mathematically rigorous, but important from physicists’ point of view as it opened a new
way of approaching itinerant electron ferromagnetism.
Mielke’s proof of his main theorem is an elegant induction which makes use of a graph
theoretic language. The proof is not at all trivial since the problem is intrinsically a many-
body one. However, there is a very special feature of the model that any ground state of
the total Hamiltonian H = Hhop + Hint is at the same time a ground state of each of Hhop
and Hint . Because of this property, one does not have to face the very difficult problem
in many-body problems called the “competition between Hhop and Hint .” That one has
ferromagnetism in this model for any U(> 0) is closely related to this fact.
Mielke’s theorem applies not only to the Hubbard model on the kagomé lattice but to
those on a wide class of lattices called line graphs. In all of these models, the ground states
in the corresponding single-electron Schrödinger equation are highly degenerate. There have
been constructed [36, 37] other examples of Hubbard models in which the corresponding
29
There is a minor error in the derivation of critical electron density in Mielke’s paper. One should modify
this part by using the method of [37].

28
single-electron ground states are highly degenerate, and exhibit saturated ferromagnetism
for any U > 0. Ferromagnetism in the examples of Mielke and in [36, 37] are now called
flat-band ferromagnetism30 . Mielke [38] obtained a necessary and sufficient condition for a
Hubbard model with highly degenerate single-electron ground states to exhibit saturated
ferromagnetism. It is interesting that Lieb’s ferrimagnetism discussed in Section 5.3 resem-
bles flat-band ferromagnetism in that the corresponding single-electron spectrum has a bulk
degeneracy.
It is needless to say that the models in which single-electron ground states are highly
degenerate are rather singular. By adding a generic small perturbation to the hopping
Hamiltonian, the degeneracy is lifted in general, and one gets a nearly flat lowest band
rather than a completely flat one. A very interesting and important problem is whether
ferromagnetism remains stable after such a perturbation is added. Of course one does not
have ferromagnetism for small enough U if the bulk degeneracy in the single-electron ground
states is lifted. What one expects is that ferromagnetism remains stable when U is sufficiently
large. (Recall that in the toy model of Section 6.2, we had ferromagnetism for all U > 0 only
for the special choice of parameters t = t′ .) There are some indications (from numerical or
variational calculations) that ferromagnetism is stable under perturbation. As for rigorous
results, stability of ferromagnetism under single-spin flip is proved in [39, 40] for the model
obtained by adding arbitrary small perturbation to the Hubbard model of [36, 37]. As for
a special class of perturbations, the problem of stability of ferromagnetism is completely
solved as we shall see in the next section.

6.5 Ferromagnetism in a Non-Singular Hubbard model

We have seen two theorems which show that certain Hubbard models exhibit saturated fer-
romagnetism. In Nagaoka’s theorem, it is assumed that the system has exactly one hole,
and has infinitely large Coulomb interaction. In Mielke’s theorem and other flat-band ferro-
magnetism, it is essential that the single-electron ground states have a bulk degeneracy. Is it
possible to prove the existence of saturated ferromagnetism in a non-singular Hubbard model
which have finite U and in which the single-electron spectrum is not singular? Recently such
examples were constructed [41].
For simplicity, we concentrate on the simplest models in one dimension31 . Take the one
dimensional lattice Λ = {1, 2, . . . , N} with N sites (where N is an even integer), and impose
a periodic boundary condition by identifying the site N + 1 with the site 1. The hopping
30
From the view point of band structure in the single-electron problem, the bulk degeneracy in the single-
electron ground states correspond to the lowest band being completely dispersionless (or flat).
31
There are models in higher dimensions√[42]. In the original paper [41], the model contains an additional
parameter λ > 0. Here we have set λ = 2 to simplify the discussion. The proof of the main theorem in
[41] is considerably improved in [42]. The condition λ ≥ λc in [41] is replaced by λ > 0.

29
 t t t
t' t' t' t' t' t'

–s –s
Figure 11: An example of non-singular Hubbard model which exhibits saturated ferromag-
netism [41]. If we look at three adjacent sites, the lattice structure and the hopping resemble
that of the toy model of Figure 5.

matrix is defined by setting tx,x+1 = tx+1,x = t′ for any x ∈ Λ, tx,x+2 = tx+2,x = t for even x,
tx,x+2 = tx+2,x = −s for odd x, and tx,y = 0 otherwise. √Here t > 0 and s > 0 are independent
parameters, but the parameter t′ is determined as t′ = 2(t+s). As can be seen from Figure
11, the model32 has two kinds of next nearest neighbor hoppings t and −s, as well as the
nearest neighbor hopping t′ . If we look at an odd site and the two neighboring even sites,
the model is exactly the same as the toy model we treated in Section 6.2. Roughly speaking,
this resemblance is the basic origin of ferromagnetism in the present model. We also note
that because there are next nearest neighbor hoppings, Lieb-Mattis theorem (Theorem 4.1)
does not apply to the present model.
The single electron energy eigenvalue in this model can be expressed by using the wave
number k = 2πn/N (n = 0, ±1, . . . , ± {(N/4) − 1} , N/4) as ε1 (k) = −2t−2s(1+cos 2k) and
ε2 (k) = 2s + 2t(1 + cos 2k). There are two bands, and each of them has healthy dispersion.
As for the Coulomb interaction, we set Ux = U > 0 for any x ∈ Λ. We fix the electron
number as Ne = N/2. In terms of filling factor, this corresponds to the quarter filling. The
maximum possible value of total spin is Smax = N/4.
Unlike in flat-band ferromagnetism, there is no saturated ferromagnetism when U is suf-
ficiently small. Theorem 6.1 ensures that the ground state have Stot < Smax if U < 4s. If
the present system were to show saturated ferromagnetism, it should be in the “nonper-
turbative” region with sufficiently large U. The following theorem of [41] provides such a
nonperturbative result.

Theorem 6.5 (Ferromagnetism in a non-singular Hubbard model) Suppose that the

two dimensionless parameters t/s and U/s are sufficiently large. Then the ground states have
Stot = Smax (= N/4) and are nondegenerate apart from the trivial spin degeneracy.
32
Solvable Hubbard models with U = ∞ which have similar structure as the present models were found
by Brandt and Giesekus [43], and were extended in [44, 45, 46, 47]. The conjectured uniqueness of the
ground state was proved in [48, 47]. The ground state correlation functions in one dimensional models were
calculated exactly in [48, 49], and insulating behavior was found. ([46] contains an error which is corrected
in the footnote 6 of [47]. Although I discussed the possibility of superconductivity in these model in [46], I
am not very optimistic about this conjecture at present.)

30
 The theorem is valid, for example, when t/s ≥ 4.5 if U/s = 50, and t/s ≥ 2.6 if
U/s = 100. The ferromagnetic ground state can be constructed in exactly the same manner
as Φ↑ in the previous section.
Although the model is rather artificial, this is the first rigorous example of saturated
ferromagnetism in a non-singular Hubbard model on which we have to overcome the com-
petition between Hint and Hhop . In a class of similar models, it is also proved that low-lying
excitation above the ground state has a normal dispersion relation of a spin-wave excitation
[41, 40]. Starting from a Hubbard model model of itinerant electrons, the existence of a
“healthy” ferromagnetism is established rigorously.
If we set s = 0 in the present model, the ground states of the single-electron Schrödinger
equation become N/2-fold degenerate. In this case, the model exhibits saturated ferromag-
netism (flat-band ferromagnetism) for any U > 0. Theorem 6.5 for s ̸= 0 can be regarded as
a solution to the problem about stability of flat-band ferromagnetism against perturbation
to the hopping Hamiltonian.
The basic strategy in the proof of Theorem 6.5 is first to establish the existence of
saturated ferromagnetism in a Hubbard model on a chain with five sites, and then “connect”
together these local ferromagnetism to get ferromagnetism in the whole system. Generally
speaking, this is a crazy idea. In a quantum mechanical system, especially in a system with
healthy dispersion (like the present one), electrons have strong tendency to extend in a large
region and reduce kinetic energy. To confine electrons in a finite region usually costs extra
energy. To obtain exact information about large system from smaller system seems to be
impossible. The reasons that such a strategy works in the present model are twofold. One is
the special construction of the model. The other is that we described electron states using a
language which takes into account both the particle-like character of electrons and the band
structure of the model. The latter is a natural strategy to deal with the Hubbard model in
which wave-particle dualism generate interesting physics.
It is believed that the ground states of the present Hubbard model describes an insulator.
When the electron number is less than N/2, we expect that the present model exhibits
metallic ferromagnetism in which the same set of electrons participate in conduction as well
as magnetism33 . For the moment, we still do not know of any useful ideas in proving this
fascinating conjecture.

I wish to thank Kenn Kubo and Balint Tóth for useful comments on the early version of
the present article.

33
By using a heuristic perturbation theory based on the Wannier functions, the low energy effective theory
of the present Hubbard model is shown to be the ferromagnetic t-J model. Moreover, by considering models
close to the flat-band model, one can make |t/J| arbitrarily small. This observation gives a strong support
to the above conjecture.

31
References
[1] E. H. Lieb, in Proceedings of 1993 conference in honor of G.F. Dell’Antonio, Advances
in Dynamical Systems and Quantum Physics, pp. 173-193, World Scientific (1995),
Proccedings of 1993 NATO ASW The Physics and Mathematical Physics of the Hub-
bard Model, Plenum (in press), and Proceedings of the XIth International Congress of
Mathematical Physics, Paris, 1994, D. Iagolnitzer ed., pp. 392-412, International Press
(1995). Archived as cond-mat/9311033.

[2] H. Tasaki, “From Nagaoka’s ferromagnetism to flat-band ferromagnetism and beyond:

An introduction to ferromagnetism in the Hubbard model”, preprint (1997), cond-
mat/9712219.

[3] R. Strack and D. Vollhardt, J. Low Temp. Phys. 99, 385 (1995).

[4] J. de Boer, V. E. Korepin and A. Schadschneider, Phys. Rev. Lett. 74, 789 (1995).

[5] J. de Boer and A. Schadschneider. Phy. Rev. Lett. 75, 4298 (1995).

[6] D. Vollhardt, N. Blümer, K. Held, M. Kollar, J. Schlipf, and M. Ulmke, Z. Phys. B 103,
283 (1997).

[7] E. H. Lieb and D. Mattis, Phy. Rev. 125, 164 (1962).

[8] E. H. Lieb and D. Mattis, J. Math. Phys. 3, 749 (1962).

[9] D. Ghosh, Phy. Rev. Lett. 27, 1584 (1971).

[10] T. Koma and H. Tasaki, Phy. Rev. Lett. 68, 3248 (1992).

[11] M. Yamanaka, M. Oshikawa, and I. Affleck, Phys. Rev. lett. 79, 1110, (1997).

[12] M. Oshikawa and M. Yamanaka, in preparation.

[13] E. H. Lieb, Phy. Rev. Lett. 62, 1201 (1989), errata Phy. Rev. Lett. 62, 1927 (1989)

[14] T. Koma and H. Tasaki, J. Stat. Phys. 76, 745 (1994).

[15] S. Q. Shen, Z. M. Qiu, and G. S. Tian, Phy. Rev. Lett. 72, 1280 (1994).

[16] S. Q. Shen, Phys. Rev. B 53, 14252 (1996).

[17] F. Dyson, E. H. Lieb, and B. Simon, J. Stat. Phys. 18, 335 (1978).

[18] T. Kennedy, E. H. Lieb, and B. S. Shastry, J. Stat. Phys. 53, 1019 (1988).

32
[19] E. H. Lieb, M. Loss, and R. McCann, J. Math. Phys. 34, 891 (1993).

[20] E. H. Lieb, Phy. Rev. Lett. 73, 2158 (1994).

[21] N. Macris and B. Nachtergaele, J. Stat. Phys. 85, 754 (1996)

[22] B. Nachtergaele and E. H. Lieb, Phys. Rev. B51, 4777 (1995).

[23] K. Kubo and K. Kishi, Phys. Rev. B41, 4866 (1990).

[24] P. Pieri, S. Daul, D. Baeriswyl, M. Dzierzawa, and P. Fazekas, Phys. Rev. B45, 9250
(1996).

[25] Y. Nagaoka, Phy. Rev. 147, 392 (1966).

[26] H. Tasaki, Phy. Rev. B40, 9192 (1989).

[27] D. J. Thouless, Proc. Phys. Soc. London 86, 893 (1965).

[28] B. Douçot and X. G. Wen, Phy. Rev. B40, 2719 (1989).

[29] B. S. Shastry, H. R. Krishnamurthy, and P. W. Anderson, Phy. Rev. B41, 2375 (1990).

[30] B. Tóth, Lett. Math. Phys. 22, 321 (1991).

[31] A. Sütő, Commun. Math Phys. 40, 43 (1991).

[32] T. Hanisch and E. Müller-Hartmann, Ann. Physik 2, 381 (1993).

[33] T. Hanisch, G. S. Uhrig, and E. Mueller-Hartmann, preprint (1997), cond-mat/9707286.

[34] A. Mielke, J. Phys. A24, 3311 (1991).

[35] A. Mielke, J. Phys. A25, 4335 (1992).

[36] H. Tasaki, Phy. Rev. Lett. 69, 1608 (1992).

[37] A. Mielke and H. Tasaki, Commun. Math Phys. 158, 341 (1993).

[38] A. Mielke, Phys. Lett. A174, 443 (1993).

[39] H. Tasaki, Phy. Rev. Lett. 73, 1158 (1994).

[40] H. Tasaki, J. Stat. Phys. 84, 535 (1996).

[41] H. Tasaki, Phy. Rev. Lett., 75, 4678 (1995).

33
[42] H. Tasaki, “Ferromagnetism in the Hubbard model: A constructive approach”, in prepa-
ration.

[43] U. Brandt and A. Giesekus. Phy. Rev. Lett. 68, 2648 (1992).

[44] A. Mielke, J. Phys. A25, 6507 (1992)

[45] R. Strack, Phys. Rev. Lett. 70, 833 (1993).

[46] H. Tasaki, Phys. Rev. Lett. 70, 3303 (1993).

[47] H. Tasaki, Phys. Rev. B49, 7763 (1993).

[48] P.-A. Bares and P. A. Lee, Phys. Rev. B49, 8882 (1993).

[49] M. Yamanaka, S. Honjou, Y. Hatsugai and M. Kohmoto, J. Stat. Phys. 84, 1133 (1996).

34