The Hubbard Model for Dummies

Introduction Some math

2nd quantization The partition function

Limiting cases
t=0 U=0 Mean field theory

Matrix notation and the t-J model Extensions of the Hubbard model
Bruce Patterson 13.6.07

The Hubbard Model: Introduction

J. Hubbard, Proc. Roy. Soc. London, A266, 238 (1963). Simplest model of interacting particles in a lattice: extension of the tightbinding model to include short-range el-el interactions. One-band Hubbard Hamiltonian (one orbital per site):

H = t c j c i + U n in i
ij i

<ij> implies adjacent sites

hopping term

interaction term

Half-filling: on average one electron per site.

Electrons may hop from site to site, but for large U/t, we get a Mott insulator.
Note: One of the first correlated electron calculations: Heitler+London treatment of H2-molecule (1927).

Second Quantization

Boson annihilation and creation operators for the harmonic oscillator:

a a
From

m 1 x+i p 2h 2mh m 1 x i p 2h 2mh

[ x, p] = ih

, we verify that

[a,a ] = 1

and that the oscillator Hamiltonian is:

H osc

1 2 m 2 2 = p + x 2m 2 1 1 = h a a + = h n + 2 2
0

n=aa = number operator.

From the ground state

, we build up the excited states:

a n = n + 1 n + 1

1 H osc n = h n + n 2

Electron states for the (N=8) Hubbard model in terms of fermionic operators:

c1 0 = ,,,,,,,

c 0 = ,,,,,,, 5
c c1 0 = ,,,,,,, 5

c1c1 ,,,,,,, n1 ,,,,,,, = 1,,,,,,,

c c 5 ,,,,,,, n 5 ,,,,,,, = 0,,,,,,, 5

c c 0 = ,,,,,,, 4 4
c1 = 0 0

c1c1 0 = 0

{c i ,c j ' } c i c j ' + c j 'c i {c ,c j ' } = 0 i {c i ,c j ' } = ij '

=0
{} = anticommutator

Partition Function
Grand Canonical Ensemble: particles can be exchanged with a resevoir.

partition function : Z Tr e ( H n )

"chemical potential" 1 A = Tr Ae ( H n ) Thermal expectation value of operator A: Z

1 kT

Consider a single state with 0 or 1 electrons with energy :

H = c c = n
Average occupation:

states :

{0 ,1}

Z = 0 e ( H n ) 0 + 1 e ( H n) 1 = 1+ e( )

1 1 e ( ) ( H n ) ( H n ) ( ) n = 0 ne 0 + 1 ne 1 = 0+e = Z Z 1+ e ( ) 1 = ( ) = Fermi - Dirac distribution e +1

Hubbard Model with t=0

No hopping implies independent sites. We thus consider a single site and calculate its average occupation.

possible states = { 0 , , , }

H = Unn
Z = e ( H n ) = 1+ e + e + e U +2

1 1 ( H n ) n = ( n + n )e = [0 + e + e + 2e U +2 ] Z Z = 2(e + e U +2 ) 1+ 2e + e U +2

2 1,8 1,6 "Hubbard gap" (=U) U=1

T=0.05 T=0.1 T=0.5

<n>

1,4 1,2 1 0,8 0,6 0 0,5 1 1,5 2 "half filling"

2,5

Note: Since =

E , at half-filling, the energy to add another particle to the n

site jumps by U. This is the Hubbard gap.

Hubbard Model with U=0

Non-interacting limit: we go into reciprocal space:

1 = N

e
l

ik l l

where k takes the values kn=2n/N. (1-d, periodic boundary conditions)

t H = t c c l = e ikj eik' l c c k' k N k,k' jl jl

j

t ikj ik' ( j +1) ikj ik' ( j1) ik' ik' 1 = +e e = t c k c k' [e + e ] e i(kk' ) j c k c k' e e N k,k', N j j k,k',

= t c c k' [2cos k']k,k' = k n k k

k,k', k

k = 2t cos k

For U=0, the energy-eigenvalues of H are:

k = 2t cos k
3 2

kn =

2n N

n = 0,1,2,...N 1

[t]
k

1 0 -1 -2 -3

4t

-5

k
The bandwidth is equal to 4t.

[2/N]

 The Mott Transition:

U>>t: diagonalize H in real space - at half-filling we obtain a static (insulating) system. t>>U: diagonalize H in reciprocal space - obtain itinerant electron band, width 4t.

At some intermediate value of U/t, there will be a metal-to-insulaltor transition: the Mott transition.

Mean Field Theory

We wish to treat the full Hubbard model (in 1-d):

H = t c j c + U n ln l
jl
Problem: the second term is proportional to c4. Mean field approximation: each electron feels an average interaction from all others. This ignores possibly important electron correlations. Write:

n l = n + n l n n + l

and assume is small, and analogously for the down spins. Expand:

n ln l n n + l n + l n n l n + n l n n n

In reciprocal space:

c = l
H is now diagonal:

1 N

e
k

ikl k

H = 2t cos k ( n k + n k ) + U n k n + n k n

= [kn k + kn k ] UN n n
k

)] UN n

k 2t cos k + U n

We define spin-dependent Fermi wavevectors:

N = k

kF

dk
0

kF =

n
N

After integrating, we obtain for the total energy:

E tot

2Nt = sin n + sin n + UN n n

 For half-filling:

n + n = 1

write Etot as a function of

E tot n
4Nt
0,3 0,2 0,1

( ) = sin n

U + n 1 n 4t

5 4.75 4.5 4.25

small U a paramagnet large U a ferromagnet

E/4Nt

This is completely wrong!

The 1-dimensional Hubbard model has been exactly solved by Lieb and Wu, PRL 20, 1445 (1998). At infinitesimal U/t an antiferromagnetic state forms.

0 -0,1 -0,2 U/4t = 3 -0,3 0 0,2 0,4

4 3.75 3.5 3.25

0,6
up

0,8

<n >

Hubbard Model in Matrix Notation

Toy Model (simplified H2 molecule)
Two orthonormal orbitals separated by some distance: 1 0 1 = , 2 = schematically: 1 0 1

= , , 2 = ,

Add a single electron, which moves with an amplitude -t. Tight-binding" Hamiltonian:

0 t H = t 0

1 1 1 (1 2 ) = Eigenvectors: = 2 2 1
Energy eigenvalues:

= mt

Note: + is the ground state.

 We now consider two electrons:

, = c c 0
, = c c1 0 2 , = c1c1 0

2 1

covalent states

ionic states

, = c c 0 2 2
Hamiltonian

0 0 t t 0 0 +t +t H = t +t U 0 t +t 0 U

, , , ,

Note: +t due to fermion transposition

 Diagonalize H:

U U 2 + 16t 2 = 2 2

= cov = ion

, , [ , + ] 2t
2 2 + /(2t 2 )

cov = 0 ion = U
Energy levels:
8

1 ( , + , ) 2 1 = ( , , ) 2

6 4 2

/t

ion

cov

-2 0 1 2 3 4 5 6

U/t

 Consider the limit of large U:

cov = 0
4t 2 U
- is now largely covalent, with a small ionic admixture.

With downfolding, we will project out the interesting, lowenergy, covalent-like part of Hilbert space.

 A note on matrix inversion by partitioning:

A0 A = A2
1

A1 A3

We divide a matrix A into the blocks A0-A3.

B0 B1 A = B2 B3

The inverse can also be written in block form:

B0 = ( A0 A1 A3 A2)
1

B1 = B0 A1 A31 B2 = A31 A2 B0 B3 = A31 B2 A1 A31

Ref: Ch. 2.7, Numerical Recipes, 2nd Ed. (1992)

We begin downfolding by partitioning H into blocks:

0 0 t t 0 0 +t +t H 00 H = t +t U 0 T10 t +t 0 U

T01 H11

H00 = covalent Hamiltonian H11= ionic Hamiltonian Tij=covalent-ionic transitions

The Greens function can also be partitioned:

H 00 T01 1 G00 G01 1 G() ( H ) = T10 H11 G10 G11

and the (energy-dependent) covalent part G00 can be written:
1

G00 () = H 00 + T01 ( H11 ) T10

( [
1

])

This looks like the Greens function for the effective Hamiltonian:

H eff () = H 00 + T01 ( H11 ) T10 H eff (0 )

0 is a typical covalent energy.

We evaluate the effective Hamiltonian between the covalent states:

1 0 0 t t 0 U 0 t 1 H eff (0 ) = H 00 + T01 (0 H11 ) T10 = + 0 0 t t 0 0 U t 2t 2 1 1 , H eff (0) = Note that we choose 0=0. U 1 1 ,

t t

which diagonalizes to:

4t 2 1 s = s ( , , ) U 2 1 t = 0 t = ( , + , ) 2
The other triplet states,

Recall that s includes a small ionic admixture.

These states correspond to singlet and triplet states of the H2 molecule.

, , ,

, also have zero energy. (There is no

Coulomb energy, and the electrons cannot hop, due to the Pauli principle.)

 Kinetic Exchange

Anti-parallel alignment of the spins is favored, since it allows the electrons to hop to the neighboring site.

 We can write Heff in terms of the fermion operators:

H eff

2t 2 = (c1c1c 2c 2 c1c1c 2c 2 c1c1c 2c 2 + c1c1c 2c 2 ) U

Consider the properties of the Pauli spin matrices:

1 0 + 0 1 r 1 ) ) ) S = ( x x + y y + z z ) 2 0 1 0 i 1 0 x = y = z = 1 0 i 0 0 1
x = (cc + cc ) = m y = i(cc cc ) = i m z = (cc cc ) = ( n n ) =

 We can therefore rewrite Heff :

H eff

4t 2 r r n1n 2 = S1 S2 Heisenberg Hamiltonian U 4

n i n i + n i

The Hubbard Hamiltonian, for large U/t, has led us to an

antiferromagnetic exchange coupling J=4t2/U.

This is the basis of the t-J model.

We have solved the 4x4 Hubbard Hamiltonian by exact diagonalization. For larger systems, matrix expressions for H grow quickly in size. For N sites:

N N dimension of Hilbert space = n n

N N! n n!( N n )!

2 2 = 4 1 1

For 30 sites and 15 spins of each orientation, this dimension is 2.4 x 1016.

This is why we use 2nd quantization.

No exact solution the Hubble model in >1 dimension is known.

 (mutually inconsistent) approximate solutions of the 2-D Hubbard Model

n 1 = 0 half - filling

Results of 6 groups, compiled by M.P. March, Condensed Matter Physics, Wiley, 2000.

Extensions of the Hubbard Model

Three-Band Hubbard Model
To describe the 2-d CuO2 layers in superconducting cuprates, this model includes: Cu(3dx2-y2) states (Ud) O(2px,y) states (Up) their nn interaction (Upd)

For large Upd, this Hamiltonian reduces to the t-J model.

 2-d hole-pairing:
AF-coupling causes holes to form pairs.

unpaired: 8 bonds broken

paired: 7 bonds broken

 Charge-Spin separation: In 1-d, Fermi liquid theory breaks down, due to the destruction of the Fermi surface (Fermi points) by an arbitrarily small perturbation. In this case, separate gapless charge (holon) and spin (spinon) excitations arise. This state is a Luttinger liquid.

Creation in photoemission of a holon and a spinon.

 The Hubbard model with an infinite number of dimensions Each site of the Hubbard model can then be treated as an isolated impurity interacting with an electron resevoir. This problem can be attacked with the dynamical mean-field theory. Ref: G. Kotliar and D. Vollhardt, Phys. Today March, 53 (2004).

At intermediate U/W (=U/t), we now see a sharp DOS peak at the Fermi level (Kondo resonance).

 The Hubbard-Holstein model (with electron-phonon interaction)

H H H = t c j c i + U n in i g (bi + bi ) n i + 0 bibi
ij i i i

bi = creation operator of the local phonon mode at site i

g = electron - phonon coupling constant

In the limit of large 0, the H-H model maps onto the Hubbard model with Heff=U-2g2/ 0.

Ref: W. Koller, et al, cond-mat/0312367v2 (2004).

 Introductory References to the Hubbard Model

E. Koch, Electronic Structure of Matter: Electron Correlations, in Probing the Nanoworld, lecture manuscripts of the 38th Spring School, Jlich (2006). R.T. Scalettar, Elementary Introduction to the Hubbard Model, lecture notes, UC Davis, http://leopard.physics.ucdavis.edu/rts/p210/hubbard7.pdf M.P. March, Condensed Matter Physics, Wiley, 2000.

 Some examples of 2-electron calculations:

c1c 2 , = c1 c 2 c c1 0 = c1c c 2 c1 0 = +c1c c1 c 2 0 = 0 2 2 2 c1c 2 , = c1 c 2c c 0 = c1 c 0 c1 c c 2 c 0 = c c1 0 = , 2 2 2 2 2 2 c c1 , = c c1c c1 0 = c c c1c1 0 = c c 0 + c c c1c1 0 = c c 0 = , 2 2 2 2 2 2 2 2 2 2 2