Many-Body Localization

Boris Altshuler
Physics Department, Columbia University

The 4th Windsor Summer School on Condensed Matter Theory

Quantum Transport and Dynamics in Nanostructures
Great Park, Windsor, UK, August 9 - 21, 2010
Outline:
1. Introduction to Anderson Localization
2. Phononless conductivity
3. Localization beyond real space
4. Spectral Statistics and Localization
5. Many – Body Localization
6. Disordered bosons in 1D
7. Metal - Insulator transition in electronic systems
1.Introduction
>50 years of Anderson Localization

One quantum
particle
Random potential
(e.g., impurities)
Elastic scattering
q.p.
 Einstein (1905): r 2 = Dt
Random walk
diffusion constant

always diffusion
as long as the system has no memory

Anderson(1958): It might be that

For quantum r 2 t
→ const
→∞
particles
D=0
not always!

Quantum interference memory

 Einstein Relation (1905)
dn
=σ e Dν 2
ν≡
dµ
Conductivity Density of states

Diffusion Constant

ρ
I ∂ρ t<
G =  ; − D∇ 2 ρ =0
 V  V =0 ∂t
I L t>
V σ =G
A x
 Einstein Relation (1905)
dn
=σ e Dν 2
ν≡
dµ
Conductivity Density of states

Diffusion Constant

No diffusion – no conductivity
Localized states – insulator
Extended states - metal
Metal – insulator transition
Localization of single-electron wave-functions:

z
Disorder
extended
I
V
extended
Conductance
 Lx Ly 
 σ L extended

 
localized =
z
localized 
∝ exp  − Lz  localized

   
 ς
 loc  
I was cited for work both. in the field of magnetism and in that of
disordered systems, and I would like to describe here one development
in each held which was specifically mentioned in that citation. The two
theories I will discuss differed sharply in some ways. The theory of local
moments in metals was, in a sense, easy: it was the condensation into a
simple mathematical model of ideas which. were very much in the air at
the time, and it had rapid and permanent acceptance because of its
timeliness and its relative simplicity. What mathematical difficulty it
contained has been almost fully- cleared up within the past few years.

Localization was a different matter: very few believed it at the time, and
even fewer saw its importance; among those who failed to fully
understand it at first was certainly its author. It has yet to receive
adequate mathematical treatment, and one has to resort to the indignity
of numerical simulations to settle even the simplest questions about it .
Spin Diffusion
Experiment
Feher, G., Phys. Rev. 114, 1219 (1959); Feher, G. & Gere, E. A., Phys. Rev. 114, 1245 (1959).
Light
Wiersma, D.S., Bartolini, P., Lagendijk, A. & Righini R. “Localization of light in a disordered
medium”, Nature 390, 671-673 (1997).
Scheffold, F., Lenke, R., Tweer, R. & Maret, G. “Localization or classical diffusion of light”,
Nature 398,206-270 (1999).
Schwartz, T., Bartal, G., Fishman, S. & Segev, M. “Transport and Anderson localization in
disordered two dimensional photonic lattices”. Nature 446, 52-55 (2007).
C.M. Aegerter, M.Störzer, S.Fiebig, W. Bührer, and G. Maret : JOSA A, 24, #10, A23, (2007)
Microwave
Dalichaouch, R., Armstrong, J.P., Schultz, S.,Platzman, P.M. & McCall, S.L. “Microwave
localization by 2-dimensional random scattering”. Nature 354, 53, (1991).
Chabanov, A.A., Stoytchev, M. & Genack, A.Z. Statistical signatures of photon localization.
Nature 404, 850, (2000).
Pradhan, P., Sridar, S, “Correlations due to localization in quantum eigenfunctions od
disordered microwave cavities”, PRL 85, (2000)
Sound
Weaver, R.L. Anderson localization of ultrasound. Wave Motion 12, 129-142 (1990).
f = 3.04 GHz f = 7.33 GHz

Localized State Extended State

Anderson Insulator Anderson Metal
 Localization of cold atoms
Billy et al. “Direct observation of Anderson localization
of matter waves in a controlled disorder”. Nature 453,
891- 894 (2008).
87Rb

Roati et al. “Anderson localization of a non-interacting

Bose-Einstein condensate“. Nature 453, 895-898 (2008).

Q: What about electrons ?

A: Yes,… but electrons interact with each other
Anderson • Lattice - tight binding model

Model • Onsite energies εi - random

• Hopping matrix elements Iij
j i
Iij
I i and j are nearest
-W < εi <W
uniformly distributed
{
Iij =
0
neighbors

otherwise

Anderson Transition I c = f (d ) ∗ W
I < Ic I > Ic
Insulator Metal
All eigenstates are localized There appear states extended
Localization length ξ all over the whole system
 Why arbitrary

Q: weak hopping I is
not sufficient for
the existence of
the diffusion
? j
Iij
i

Einstein (1905): Marcovian (no memory)

process g diffusion

Quantum mechanics is not marcovian

There is memory in quantum propagation !
Why ?
 I
ε2
ε1
Hamiltonian
 ε I   E 0
H = 
ˆ 1
 diagonalize H = 
ˆ 1

 I ε2  0 E2 

E2 − E1 = (ε 2 − ε1 ) 2
+I 2
  ε I   E 0
Hˆ =  1
 diagonalize H = 
ˆ 1

 I ε2  0 E2 
ε 2 − ε 1 ε 2 − ε 1 >> I
E2 − E1 = (ε 2 − ε1 ) 2
+I ≈2

I ε 2 − ε 1 << I

von Neumann & Wigner “noncrossing rule”

Level repulsion

v. Neumann J. & Wigner E. 1929 Phys. Zeit. v.30, p.467

What about the eigenfunctions ?

  ε I  ε 2 − ε 1 ε 2 − ε 1 >> I
H = 
ˆ 1
 E2 − E1 = (ε 2 − ε1 )
2
+I ≈
2

 I ε2  I ε 2 − ε 1 << I

What about the eigenfunctions ?

φ1 , ε1 ; φ2 , ε 2 ⇐ ψ 1 , E1 ; ψ 2 , E2
ε 2 − ε 1 >> I ε 2 − ε 1 << I
 I 
ψ 1, 2 = ϕ1, 2 + O ϕ 2,1
ε −
 2 1ε ψ 1, 2 ≈ ϕ1, 2 ± ϕ 2 ,1
Off-resonance Resonance
Eigenfunctions are In both eigenstates the
close to the original on- probability is equally
site wave functions shared between the sites
 Anderson insulator Anderson metal
Few isolated resonances There are many resonances
and they overlap

Transition: Typically each site is in the

resonance with some other one
 Condition for
Localization:
energy mismatch
I< # of n.neighbors

energy =εi − ε j =
W
mismatch typ

# of nearest
neighbors = 2d
A bit more precise:

Logarithm is due to the resonances, which are not nearest neighbors

 Condition for Localization:

Q:Is it correct ?
For low dimensions – NO. I = ∞ for d = 1, 2
A1:All states are localized. Reason – loop trajectories
c

  ϕ1 = ϕ 2 The particle
ϕ = ∫ pdr can go around
the loop in
Phase accumulated two directions
when traveling
along the loop
O Memory!
 For d=1,2 all states are localized.
  ϕ1 = ϕ 2 The particle
ϕ = ∫ pdr can go around
the loop in
Phase accumulated two directions
when traveling
along the loop
O Memory!

Weak Localization:
The localization length ς can be large

Inelastic processes lead to dephasing, which is

characterized by the dephasing length L
ϕ
If ς >> Lϕ , then only small corrections to a
conventional metallic behavior
 Condition for Localization:

Q:Is it correct ?
For low dimensions – NO. I = ∞ for d = 1, 2
A1:All states are localized. Reason – loop trajectories
c

A2:Works better for larger dimensions d > 2

A3:Is exact on the Cayley tree
W is the
Ic = , K branching K =2
K ln K number
Anderson Model on a Cayley tree
 extended
Eigenfunctions

localized

Does anything interesting

Q: happen with the spectrum ?
Density of States Density of States

I =0 W =0

energy energy
−W W −#I #I

Density of States
I ≠ 0; W ≠ 0

Lifshits tail extended Lifshits tail

energy

Mobility edge Mobility edge

 Anderson Transition

I > Ic I < Ic
localized and extended
never coexist!
all states are
localized
extended

DoS DoS

- mobility edges (one particle)

 Temperature dependence of the conductivity
one-electron picture
Chemical
potential

DoS DoS DoS

E c −ε F
−
σ (T → 0) > 0 σ (T ) ∝ e T
σ (T ) = 0 ∀T
Temperature dependence of the conductivity
one-electron picture

Assume that all the

states
are localized;
e.g. d = 1,2 DoS

σ (T ) = 0 ∀T
 Inelastic processes
transitions between localized states

β energy
mismatch
α

T =0 ⇒ σ =0 (any mechanism)
 Phonon-assisted hopping
ω
ω = ε α − ε β
β
α σ (T = 0) = 0

Variable Range
Hopping
N.F. Mott (1968)

Mechanism-dependent Optimized
prefactor phase volume

Any bath with a continuous spectrum of delocalized

excitations down to ω = 0 will give the same exponential
Lecture1.
2. Phononless conductivity
in Anderson insulators
with e-e interaction
 Anderson
Common Insulator Phonon assisted
belief: weak e-e hopping transport
interactions

Can hopping conductivity

exist without phonons ?
Given: 1. All one-electron states are localized
2. Electrons interact with each other
3. The system is closed (no phonons)
4. Temperature is low but finite

Find: DC conductivity σ(T,ω=0)

(zero or finite?)
Q: Can e-h pairs lead to phonon-less variable range
hopping in the same way as phonons do ?

A#1: Sure
1. Recall phonon-less
AC conductivity:
Sir N.F. Mott (1970)

2. Fluctuation Dissipation Theorem:

there should be Johnson-Nyquist noise

3. Use this noise as a bath instead of phonons

4. Self-consistency (whatever it means)

 Q: Can e-h pairs lead to phonon-less variable range
hopping in the same way as phonons do ?

A#1: Sure
A#2: No way (L. Fleishman. P.W. Anderson (1980))
Except maybe Coulomb interaction in 3D

is contributed by
rare resonances

R →∞
γ β
δ
α
matrix
element 0
R→∞ ⇒ vanishes
 No ??? No
phonons transport
∀ T

Problem:
If the localization

}
length exceeds L , At high enough
then – metal. ϕ temperatures
conductivity should
In a metal e–e
interaction leads to be finite even
a finite L without phonons
ϕ
 Q: Can e-h pairs lead to phonon-less variable range
hopping in the same way as phonons do ?
A#1: Sure
A#2: No way (L. Fleishman. P.W. Anderson (1980))
A#3: Finite temperature Metal-Insulator Transition
(Basko, Aleiner, BA (2006))

Drude
metal
insulator

σ=0
 Finite temperature Metal-Insulator Transition

Many body wave functions are Drude

localized in functional space
Many body
localization!
metal
Interaction
insulator strength

σ=0 δ ζ ≡ (νζ )
d −1Localization
spacing

Definitions:
Insulator σ =0 Metal σ ≠0
not dσ dT < 0 not dσ dT > 0
3. Localization beyond
real space
 Kolmogorov – Arnold – Moser (KAM) theory

=0
A.N. Kolmogorov,
Dokl. Akad. Nauk
SSSR, 1954.
Proc. 1954 Int. Integrable classical Hamiltonian Ĥ 0 , d>1:
Congress of
Mathematics, North-
Holland, 1957 Separation of variables: d sets of
action-angle variables
= I1 , θ1 2=
πω1t ; ... , I 2 , θ 2 2πω2t ;..

Quasiperiodic motion:
set of the frequencies,ω1 , ω2 ,.., ωd which are
in general incommensurate. Actions I i are
Andrey integrals of motion ∂I i ∂t = 0
Kolmogorov
θ1 θ2
I1 I2
⊗ ⊗…=>
Vladimir
Arnold
Jurgen tori
Moser
 Integrable dynamics:
Each classical trajectory is quasiperiodic
and confined to a particular torus, which
is determined by a set of the integrals of
motion

space Number of dimensions

real space d
phase space: (x,p) 2d
energy shell 2d-1
tori d

Each torus has measure zero on the energy shell !

 Kolmogorov – Arnold – Moser (KAM) theory
A.N. Kolmogorov, Integrable classical Hamiltonian Ĥ 0 , d>1:
Dokl. Akad. Nauk
SSSR, 1954. Separation of variables: d sets of action-angle
Proc. 1954 Int. variables I , θ = 2πω t ;.., I , θ = 2πω t ;..
1 1 1 2 2 2
Congress of Quasiperiodic motion: set of the frequencies,
Mathematics, North-
Holland, 1957 ω1 , ω2 ,.., ωd which are in general incommensurate
Actions Ii are integrals of motion ∂I i ∂t = 0
θ1 θ2
I1 I2
⊗ ⊗…=>
Will an arbitrary weak perturbation
Andrey
Kolmogorov
Q: Vˆ of the integrable Hamiltonian Ĥ 0
destroy the tori and make the motion
ergodic (when each point at the energy
?
shell will be reached sooner or later)
Most of the tori survive KAM
Vladimir
Arnold
Jurgen A: weak and smooth enough theorem
perturbations
Moser
 Kolmogorov – Arnold – Moser (KAM) theory
A.N. Kolmogorov, Will an arbitrary weak perturbation
Dokl. Akad. Nauk
SSSR, 1954. Q: Vˆ of the integrable Hamiltonian Ĥ 0
destroy the tori and make the
Proc. 1954 Int.
Congress of
motion ergodic (i.e. each point at
the energy shell would be reached
Mathematics, North-
Holland, 1957 sooner or later) ?
Most of the tori survive KAM
A: weak and smooth enough theorem
perturbations

Andrey
Kolmogorov

Vladimir Jurgen
Arnold Moser
 KAM Most of the tori survive weak and
theorem: smooth enough perturbations
I2 I2

Vˆ ≠ 0
I1 I1
Each point in the space of the Finite motion.
integrals of motion corresponds
to a torus and vice versa
Localization in the space
?
of the integrals of motion
 KAM Most of the tori survive weak and
theorem: smooth enough perturbations
I2

I1
Rectangular billiard
Lx I2 ≠0
Ly

I1
Two integrals of motion
πn πm
px = ; py =
=I1 p=
x; I1 p y Lx Lx
 KAM Most of the tori survive weak and
theorem: smooth enough perturbations
I2 I2

Vˆ ≠ 0
I1 I1
I2 ≠0

I1
Energy shell
 Vˆµ ,ν
µ ν
Matrix element of
the perturbation

 (µ )
µ = I
 (µ )
{
(µ ) (µ )
I = I1 ,..., I d } One can speak about localization
provided that the perturbation
is somewhat local in the space
of quantum numbers of the
original Hamiltonian
AL hops are local – one can distinguish “near” and “far”
KAM perturbation is smooth enough
 Glossary
Classical Quantum
Integrable Integrable 
() Hˆ 0 = ∑ Eµ µ µ ,
 µ = I
H0 = H0 I
µ

KAM Localized
Ergodic – distributed all
over the energy shell Extended ?
Chaotic
Strong disorder localized
Weak disorder extended

Strong disorder localized

Moderate disorder extended
No disorder chaotic extended
No disorder integrable localized
Too weak disorder int. localized
Consider an integrable system.
Each state is characterized by a set of
quantum numbers.
It can be viewed as a point in the space of
quantum numbers. The whole set of the states
forms a lattice in this space.

A perturbation that violates the integrability

provides matrix elements of the hopping
between different sites (Anderson model !?)

Is it possible to tell if the

Q: states are localized (in some
unknown basis) or extended. ?
Density of States is not singular
at the Anderson transition

This applies only to the

average Density of States !
Fluctuations ?
4. Spectral statistics and
Localization
 Spectral
RANDOM MATRIX THEORY statistics
ensemble of Hermitian matrices
N×N with random matrix element N→∞
Eα - spectrum (set of eigenvalues)

δ 1 ≡ Eα +1 − Eα - mean level spacing,

determines the density of states
...... - ensemble averaging
Eα +1 − Eα
s≡ - spacing between nearest
δ1 neighbors

P (s ) - distribution function of nearest

neighbors spacing between

Spectral Rigidity P (s = 0) = 0
Level repulsion P (s << 1) ∝ s β
β =1,2,4
 Wigner-Dyson; GOE Gaussian
Poisson Orthogonal
Ensemble

Orthogonal
β=1
Unitary
β=2
Simplectic
β=4

Poisson – completely
uncorrelated
levels
 RANDOM MATRICES
N×N matrices with random matrix elements. N→∞

Dyson Ensembles
Matrix elements Ensemble β realization
real orthogonal 1 T-inv potential
complex unitary 2 broken T-invariance
(e.g., by magnetic
field)
2 × 2 matrices simplectic 4 T-inv, but with spin-
orbital coupling
Anderson • Lattice - tight binding model

Model • Onsite energies εi - random

• Hopping matrix elements Iij
j i
Iij
-W < εi <W
uniformly distributed

Is there much in common between Random Matrices

and Hamiltonians with random potential ?

Q: What are the spectral statistics

of a finite size Anderson model ?
 Anderson Transition
Strong disorder Weak disorder

I < Ic I > Ic
Insulator Metal
All eigenstates are localized There appear states extended
Localization length ξ all over the whole system

The eigenstates, which are Any two extended

localized at different places eigenstates repel each other
will not repel each other

Poisson spectral statistics Wigner – Dyson spectral statistics

 Anderson Localization and
Spectral Statistics
0 Ic I
Localized states Extended states
Insulator Metal
Poisson spectral Wigner-Dyson
statistics spectral statistics
Consider an integrable system.
Each state is characterized by a set of
quantum numbers.
It can be viewed as a point in the space of
quantum numbers. The whole set of the states
forms a lattice in this space.

A perturbation that violates the integrability

provides matrix elements of the hopping
between different sites (Anderson model !?)

Weak enough hopping:

Localization - Poisson
Strong hopping:
transition to Wigner-Dyson
Extended Level repulsion, anticrossings,
states: Wigner-Dyson spectral statistics

Localized Poisson spectral statistics

states:

Invariant
(basis independent)
definition
Many-Body
Localization

BA, Gefen, Kamenev & Levitov, 1997

Basko, Aleiner & BA, 2005. . .
Example: Random Ising model in the perpendicular field
Will not discuss today in detail
N N N
=
Hˆ ∑ i i ∑ ij i j ∑ i 0 ∑ i
B
i=
1
σˆ z
+ J σ
i≠ j
ˆ σ
z z
ˆ + I σ
i=
1
ˆ x
≡ ˆ
H + I σ
i=
1
ˆ x

Perpendicular
Random Ising model field
in a parallel field

σi - Pauli matrices, σ = ±i
1
z

2
i 1, 2,..., N ; N >> 1

σ
Without perpendicular field all i
z
commute with the Hamiltonian, i.e.
they are integrals of motion
 N N N
=
Hˆ ∑ i i ∑ ij i j ∑ i 0 ∑ i
B σˆ
i=
1
z
+ J σ
i≠ j
ˆ σ
z z
ˆ + I σˆ
i=
1
x
≡ ˆ
H + I σˆ x

i=
1

Perpendicular
Random Ising model field
in a parallel field
 Withoutz perpendicular field
σ i - Pauli matrices all σ i commute with the
Hamiltonian, i.e. they are
i 1, 2,..., N ; N >> 1 integrals of motion

Anderson Model on {σ z } determines a site

i
N-dimensional cube
H {σ } ( )σ= σ + σ
x + −
ˆ ˆ ˆ
0 i
hoping between
onsite energy nearest neighbors
 N N N
=
Hˆ ∑ i i ∑ ij i j ∑ i 0 ∑ i
B
i=
1
σˆ z
+ J σˆ σ
z z

i≠ j
ˆ + I σˆ x

i=
1
≡ ˆ
H + I σˆ x

i=
1

Anderson Model on N-dimensional cube

Usually: Here:
# of dimensions d → const # of dimensions d= N → ∞
system linear size L→∞ system linear size L =1
6-dimensional cube 9-dimensional cube
 σ>0
Conductivity σ
insulator metal
σ=0
δ ζ ≡ (νζ )
d −1 localization
spacing
Many body
localization! interaction
strength

Bad metal Drude metal

temperature T

Definitions:
Insulator σ =0 Metal σ ≠0
not d σ dT < 0 not dσ dT > 0
Many-Body Localization
1D bosons + disorder
 1D Localization

Exactly solved: Gertsenshtein & Vasil'ev,

all states are localized 1959

Conjectured: Mott & Twose, 1961

...

correct for
1-particle problem bosons as well
as for fermions
 Bosons without disorder
•Bose - Einstein condensation

•Bose–condensate even at weak enough repulsion

•Even in 1d case at T=0 – “algebraic superfluid”

•Finite temperature – Normal fluid

T
Normal fluid
 Localization of cold atoms
Billy et al. “Direct observation of Anderson localization
of matter waves in a controlled disorder”. Nature 453,
891- 894 (2008).
87Rb

Roati et al. “Anderson localization of a non-interacting

Bose-Einstein condensate“. Nature 453, 895-898 (2008).

No interaction !
Thermodynamics of ideal
Bose-gas in the presence
of disorder is a pathological Need
problem: all particles will repulsion
occupy the localized state
with the lowest energy

Q: 1D Bosons + disorder
+ weak repulsion ?
 Weakly interacting bosons

•Bose - Einstein condensation

•Bose–condensate even at weak enough repulsion

•Even in 1D case at T=0 – “algebraic superfluid”

1. No interaction 2. No disorder 3. Weak repulsion

disorder disorder

insulator
Normal fluid T
For any
Superfluid-
energy at
insulator
finite
(insulator)

transition
disorder Superfluid-
glass

superfluid
1D insulator
localization transition
T=0 Superfluid – Insulator Quantum Phase Transition

disorder

insulator
Berezinskii Kosterlitz

superfluid
Thouless
transition in 1+1 dim.

relatively
T. Giamarchi and H. J. Schulz, Phys. Rev., strong
B37, #1(1988). interaction

E. Altman, Y. Kafri, A. Polkovnikov & G.

Refael, Phys. Rev. Lett., 100, 170402 (2008).
G.M. Falco, T. Nattermann, & V.L. Pokrovsky,
Phys. Rev., B80, 104515 (2009).
} weak
interaction
 disorder

superfluid insulator

?
T
Normal fluid

Is it a normal fluid at any temperature ?

 There can be no phase transitions
Dogma at a finite temperature in 1D
Van Howe, Landau

Thermal fluctuation destroy any

Reason long range correlations in 1D

T=0 Normal fluid – Insulator Phase Transition:

Neither normal True phase transition:

fluids nor glasses singularities in
(insulators) still transport (rather
exhibit long range than thermodynamic)
correlations properties
What is insulator?
Perfect Zero DC conductivity at
Insulator finite temperatures
Possible if the system is decoupled from any outside bath

Normal Finite (even if very small)

metal DC conductivity at finite
(fluid) temperatures
1D Luttinger liquid: bosons = fermions ?
Bosons with infinitely
strong repulsion ≈ Free fermions

Free bosons ≈ Fermions with infinitely

strong attraction

Weakly interacting
bosons ≈ Fermions with strong
attraction
U U

f
f
b
b b f

x x

As soon as the occupation numbers become large

the analogy with fermions is not too useful
1D Weakly Interacting Bosons + Disorder
Aleiner, BA & Shlyapnikov, 2010, Nature Physics, to be published
cond-mat 0910.4534
disorder
1. No interaction 3. T=0
disorder

(insulator)
glass
K-T
For any transition
temperature
and any
(insulator)

finite
glass

disorder

“Algebraic
suprfluid”
1D disorder
localization

2. No disorder
T
?
Normal fluid T
Density of States ν(ε) in one dimension

ν(ε)

No disorder
ν (ε ) =
m
Quadratic spectrum
1 2 2π  ε
2 2

ε= p
2m - singularity

ε
Density of States ν(ε) in one dimension

ν(ε) ν(ε)
No disorder In the presence
of disorder the
Quadratic spectrum singularity is
smeared
ν (ε ) =
m
2π 2  2ε

ε ε
 Density of States ν(ε) in one dimension

ν(ε)
Lifshitz tail:
exponentially
small Density
of States

ν (ε ) =
m
2π 2  2ε

ε
Weak disorder – random potential U(x)

Random potential U(x):

ν(ε) Amplitude U0
Correlation length σ
2

Short range disorder: U 0 <<
mσ 2

Localization length ζ>>σ

ε
Characteristic scales:
13

Energy
U σ m 
4 2
E∗ ≡  0

ν(ε)
2
  

(ς ∗ E∗ )−1    4
13

Length ς ∗ ≡  2  >> σ
 U0σ m 
E∗
All states are localized
Localization length:
ς∗ ε ~ E∗
ε ς (ε ) ~ ε
ς∗ ε >> E∗
E∗
 ν(ε) ν(ε)
(ς ∗ E∗ )−1 (ς ∗ E∗ )−1

E∗ E∗

Low energy
High energy
ε ε
 Finite density Bose-gas with repulsion
Density n
Two more energy scales
2 2
 n
Temperature of quantum degeneracy Td ≡
m
Interaction energy per particle ng
κ ≡ E∗ ng Characterizes the
Two strength of disorder
dimensionless
parameters Characterizes the
γ ≡ ng Td interaction strength

Strong disorder κ >> 1

Weak interaction γ << 1
Dimensionless temperature t = T ng

Critical temperature Tc tc = tc (κ , γ )
Critical disorder κ c = κ c (t , γ )

Phase transition line on the t,κ - plane

 Finite temperature
phase transition in 1D
κ ≡ E∗ ng
κ c (t )
κc = t 13

κ c = (t γ )
2 13

κc ~ 1

1 1 γ 1γ
t ≡ T ng
 Conventional Anderson Model
•one particle,
•one level per site,
•onsite disorder
•nearest neighbor hoping
labels
Basis: i , i sites

=
ˆ
Hamiltonian: H Hˆ 0 + Vˆ
Hˆ 0 = ∑ ε i i i Vˆ = ∑I i j
i i , j = n.n.

Transition: happens when the hoping matrix element

exceeds the energy mismatch
The same for many-body localization
 Many body Anderson-like Model
• many particles,
• several particles Basis: µ ≡ {n } i
per site.
• interaction
i label sites

ni = 0,1, 2,3,...
occupation numbers
 Many body Anderson-like Model
µ
• many particles,
• several particles Basis:
µ = { ni }
per site.
• interaction
i labels sites
occupation
ni = 0,1, 2,... numbers

Hamiltonian:
 Ĥ 0 = ∑ Eµ µ µ
= Hˆ 0 + Vˆ
H µ

Vˆ = ∑ I µ η (µ ) l
µη µ
, ( )
k
η ( µ ) =.., ni − 1,.., n j − 1,.., nk + 1,.., nl + 1,..
δ j

i, j, k , l = n.n. i
 Conventional Many body Anderson-
Anderson like Model
Model Basis: µ , µ = { niα }
Basis: i ni = 0,1, 2,...
labels
i labels sites
i sites
occupation numbers

Hˆ ∑ε i
i i i +=Ĥ ∑µ Eµ µ µ + ∑ I µ ν (µ )
µν µ
, ( )
∑
i , j = n.n.
I i j

“nearest ν ( µ ) =.., ni − 1,.., n j − 1,.., nk + 1,.., nlδ + 1,..

neighbors”: i, j, k , l = n.n.
Transition temperature: Tc ≡ tc ( ng )
i , j ⇒ k ,l j
transition k
i l
Transition temperature: Tc ≡ tc ( ng )
i , j ⇒ k ,l j
transition k
i l
∆ ij ,kl ≡ ε i + ε j − ε k − ε l energy
mismatch
I ij ,kl matrix element Decay of a state i
∆ typical mismatch
N1 typical # of channels
Anderson condition: I typical matrix element

>> ∆ ( T ) N1 ( T ) extended
I (T )
<< ∆ ( T ) N1 (T ) localized
High temperatures: T >> Td t >> γ −1
Bose-gas is not degenerated;
occupation numbers either 0 or 1

I~g ς

Matrix element of the transition

I ~ g ς ( ε = T ) ~ ( gE∗ ) (ς ∗T )
should be compared with the minimal energy
( )
mismatch (νς ) −1 ( nς ) ~ ν nς 2T 2 −1 E 2
∗ ∗

κ c (t ) ∝ t tγ >> 1
13
Localization Number of
spacing δ channels
ς
Intermediate temperatures: γ −1 2
<< t << γ −1

1. T << Td tγ << 1

2. Bose-gas is degenerated; occupation numbers

either >>1.

3. Typical energies |µ|=T2/Td , µ is the chemical

potential. Correct as long as multiple T
N (ε ) ~
µ >> ng , E t γ >> 1 occupation ε
*

<< T
4. Characteristic energies ε ~ µ
>> ng , E*
We are still dealing with
the high energy states
Intermediate temperatures: γ −1 2 << t << γ −1
µ = T Td >> ng , E*
2
T << Td

Bose-gas is degenerated; typical energies ~

|µ|>>T occupation numbers >>1 matrix
elements are enhanced

g T
IN1 ~
ς (ε ) ε

κ c (t ) ∝ t γ
23 13
γ << tγ << 1
Low temperatures: t << γ −1 2 Start with T=0
Bosons occupy only
Suppose κ ≡ E∗ ng >> 1 µ << E∗ small fraction of low
energy states ε i < µ
Low temperatures: t << γ −1 2 Start with T=0
Bosons occupy only
Suppose κ ≡ E∗ ng >> 1 µ << E∗ small fraction of low
energy states ε i < µ

Localization length ς ∗
Occupation #: ( µ − εi )ς ∗ g µ 2
µ = E∗ κ
n=
ν (ε ) = ( E∗ς ∗ )
−1
DoS: 2 gE∗
ς∗ l (κ ) ς ∗ κ >> ς ∗
l (κ ) =
Occupation
nl (κ )=
ς ∗ γ −1 2 >> 1
x
Low temperatures: t << γ −1 2
Occupation
nl (κ )=
ς ∗ γ −1 2 >> 1
κ ≡ E∗ ng >> 1 “lakes”
Distance
l (κ ) ς ∗ κ >> ς ∗
=
ς∗
l (κ )

x
l (κ ) >> ς ∗ Strong
insulator

κ → κc Insulator – Superfluid transition in

l (κ ) << ς ∗ a chain of “Josephson junctions”
Low temperatures: t << γ −1 2
Strong
κ ≡ E∗ ng >> 1

}
insulator

T = 0 transition κ c ~ 1
κ c ~ 1 for t << γ −1 2
Disordered interacting bosons in two dimensions
Disordered interacting bosons in two dimensions

Justification:
1. At T=0 normal state is unstable with respect to either
insulator or superfluid.
2. At finite temperature in the vicinity of the critical
disorder the insulator can be thought of as a collection
of “lakes”, which are disconnected from each other.
The typical size of such a “lake” diverges. This means
that the excitations in the insulator state are localized
but the localization length can be arbitrary large.
Accordingly the many –body delocalization is unavoidable
at an arbitrary low but finite T.
Phononless conductance

Many-body Localization
of fermions
 σ>0
Conductivity σ
insulator metal
σ=0
δ ζ ≡ (νζ )
d −1 localization
spacing
Many body
localization! interaction
strength

Bad metal Drude metal

temperature T

Definitions:
Insulator σ =0 Metal σ ≠0
not d σ dT < 0 not dσ dT > 0
 Many body Anderson-like Model
• many particles,
• several levels Basis: µ
µ = { ni }
α
per site,
• onsite disorder
• local
interaction
i labels
sites
α labels
levels
Hamiltonian: α occupation
Ĥ 0 = ∑ Eµ µ µ ni = 0,1 numbers
Hˆ = Hˆ + Vˆ + Vˆ 0 1 2 µ Vˆ1
Vˆ1 = ∑ I µ ν (µ )
µν µ
, ( ) I
ν ( µ ) = .., ni − 1,.., n j + 1,.. , i, j =n.n.
α β

Vˆ2
Vˆ2 = ∑ U µ η (µ ) U
µη µ , ( )
ν ( µ ) = .., niα − 1,.., niβ − 1,.., niγ + 1,.., niδ + 1,..
Conventional Many body Anderson-
Anderson like Model
Model Basis: µ , µ = { ni }
α

Basis: i niα = 0,1

i labels i labels
sites
α labels
levels occupation
sites numbers
=Ĥ ∑µ Eµ µ µ +
Hˆ ∑ε i i i +
i
∑ I µ ν (µ ) +
∑ I i j µν µ
, ( )

∑ µ η (µ )
i , j = n.n.
U
µη µ
, ( )
Two types of ν ( µ ) = .., niα − 1,.., n βj + 1,.. , i, j =n.n.
“nearest
neighbors”: η ( µ ) = .., n α
i − 1,.., ni
β
− 1,.., ni
γ
+ 1,.., n δ
i + 1,..
Anderson’s recipe:
1. take descrete spectrum Eµ of H0 insulator
2. Add an infinitesimal Im part iη to Eµ
3. Evaluate ImΣ µ

1 2
4 1) N → ∞
limits 2) η →0

4. take limit η → 0 but only after N → ∞ metal

5. “What we really need to know is the

!
probability distribution of ImΣ, not
its average…”
Probability Distribution of Γ=Im Σ
η is an infinitesimal width (Im
part of the self-energy due to
metal a coupling with a bath) of
insulator
one-electron eigenstates

Look for:

V
 Stability of the insulating phase:
NO spontaneous generation of broadening
Γα (ε ) = 0 ε → ε + iη
is always a solution linear stability analysis
Γ Γ
→ πδ (ε − ξα ) +
(ε − ξα ) + Γ
2 2
(ε − ξα ) 2

After n iterations of n
the equations of the η  λT 1 
Self Consistent Pn (Γ) ∝ 3 2 const ln
Born Approximation Γ  δ ζ λ 
first
then
(…) < 1 – insulator is stable !
Physics of the transition: cascades
Conventional wisdom:
For phonon assisted hopping one phonon – one electron hop

It is maybe correct at low temperatures, but the higher

the temperature the easier it becomes to create e-h pairs.

Therefore with increasing the temperature the typical

number of pairs created nc (i.e. the number of hops)
increases. Thus phonons create cascades of hops.

Typical size β

of the
cascade
≈ Localization
length
α
Physics of the transition: cascades
Conventional wisdom:
For phonon assisted hopping one phonon – one electron hop

It is maybe correct at low temperatures, but the higher

the temperature the easier it becomes to create e-h pairs.

Therefore with increasing the temperature the typical

number of pairs created nc (i.e. the number of hops)
increases. Thus phonons create cascades of hops.

At some temperature T = Tc nc (T ) → ∞. β
This is the critical temperature.
α
Above Tc one phonon creates
infinitely many pairs, i.e., phonons
are not needed for charge transport.
Many-body mobility edge

mobility
transition ! edge
Many-body mobility edge
Large E (high T): extended states
(good metal)
Fermi Golden Rule
hopping (bad metal)
mobility
transition ! edge
Finite T normal metal – insulator transition
is another
example of the many-body localization

σ>0
Conductivity σ

insulator metal
σ=0
δ ζ ≡ (νζ )
d −1 localization
spacing
Many body
localization! interaction
strength

nonergodic
Bad metal ergodic
Drude metal
temperature T
Definition: We will call a quantum state µ
ergodic if it occupies the number N µ
of sites N µ on the Anderson lattice,
which is proportional to the total
number of sites N :
Nµ Nµ
N→ 0
→∞
N→ const > 0
→∞
N N
nonergodic ergodic

Localized states are N µ N → const

obviously not ergodic: → ∞

Q: Is each of the extended state ergodic ?

A: In 3D probably YES, for d>4 - probably NO
Nonergodic states
Cayley tree
(Bethe lattice)
W
Ic =
K ln K
is the
branching
K number

Ic < I < W
Extended but
not ergodic
 W
I≈ ⇒ N µ ≈ ln N << N
K nonergodic
 ???
nonergodic glassy
Main postulate of the Gibbs StatMech-
equipartition (microcanonical distribution):
In the equilibrium all states with the same
energy are realized with the same
probability.
Without interaction between particles the
equilibrium would never be reached – each
one-particle energy is conserved.
Common believe: Even weak interaction
should drive the system to the equilibrium.

Is it always true?
Lecture 3.
4. Speculations
 Conductivity σ insulator metal σ>0
σ=0
δ ζ ≡ (νζ )
d −1 localization
spacing
Many body
localization! interaction
strength

Bad metal Drude metal

temperature T

Q: What happens in the classical limit → 0 ?

Speculations: 1.No transition Tc → 0
2.Bad metal still exists
Reason: Arnold diffusion
 Arnold diffusion
I2 I2

Vˆ ≠ 0
I1 I1
Each point in the space of the
integrals of motion corresponds ?
Finite motion
to a torus and vice versa

d =2 All classical trajectories

correspond to a finite motion

d >2 Most of the trajectories

correspond to a finite motion
However small fraction of the
trajectories goes infinitely far
 Arnold diffusion
1. Most of the tori survive – KAM
2. Classical trajectories do not cross each
other
space # of dimensions
real space d
phase space 2d
energy shell 2d-1
tori d
d =2 ⇒ d en.shell − dtori =1 d =2 ⇒ d en.shell − dtori =1
Each torus A torus does not have
has “inside” “inside” and “outside” as
inside
and “outside” a ring in >2 dimensions
Speculations:
1. Arnold diffusion Nonergodic (bad)
metal

2. Appearance of the transition (finite Tc ) –

quantum localization of the Arnold diffusion
 Conclusions
Anderson Localization provides a relevant language
for description of a wide class of physical
phenomena – far beyond conventional Metal to
Insulator transitions.
Transition between integrability and chaos in
quantum systems
Interacting quantum particles + strong disorder.
Three types of behavior:
ordinary ergodic metal
“bad” nonergodic metal
“true” insulator
A closed system without a bath can relaxation to a
microcanonical distribution only if it is an ergodic
metal