A brief introduction to Anderson Localization

Chenguang Guan, Xingyue Guan

Department of Physics, Massachusetts Institute of Technology
(Dated: May 18, 2019)
This paper is aimed to make a thorough review of the famous phenomenon called Anderson Lo-
calization, both analytically and numerically. Properties of localization in 1, 2 and 3D would be
discussed through various approaches including classical anology, transfer matrix methods, locator
expansions and even scaling theory. At the end of this paper, the properties and results reviewed
would extend to cutting-edge research topics, showing the significance of Anderson localization, as
has been witnessed by over 50 years.

I. INTRODUCTION like a free particle, the return probability of the

electron would decay, and thus its wave function
P. W. Anderson stated in his 1958 paper[1] the propagates. Therefore, a non-decaying return prob-
phenomena of localization in random potential. And ability would indicate localization. This is also An-
the localization phenomena leads to insights on Metal- derson’s method in his article where he used the
Insulator Transition (MIT) and various kinds of applica- method called locator expansion[2] for the resol-
tion. The following more than 50 years have witnessed vent, using Green functions as well as analogy to
its importance with innovative derivations and numeri- random walking to figure out that, sufficiently large
cal results, as well as experimental observation that stand randomness would cause localization. Besides, one
the cutting-edge of every age. can also set the initial state (like a Gaussian wave
packet) and use other time-dependent observables
• Strong disorder: In Anderson’s seminal paper, to study localization.
he proposed that if the disorder is strong enough,
the localization of the states will happen no mat- • Localization of eigenstates Localization of
ter which the dimension of the system d is. How- wavefunction can also be characterized by the in-
ever, delocalization and crtical behaviour are not verse participation ratio.This fits well with matrix
well studied in this article. diagonalization and eigen-system calculation using
computers. Also this method would be easy to vi-
• low dimension d = 1, 2:[5][4] Motivated by An-
sualize and to study the localization length of the
derson’s work, people tried to solve the problem
system.
when the disorder is finite.
If d = 1, it can be exactly proved that the local- • Spectrum continuity The energy spectrum of lo-
ization must happen whenever how small the dis- calized states and delocalized states are different,
order is. As for d = 2, there is no exact solution. which are absolute continuous spectrum and pure
However, we can solve it approximately by Renor- point spectrum respectively.
malization Group (RG) analysis, which shows that
localization must happen in 2d system no matter • Spectral statistics The spectral statistic study
how small the disorder is. Many solid numeric ex- the statistics of the normalized gap between neig-
Ei+1 −Ei
periment supported the RG results. boring energy levels si = hE i+1 −Ei i
.
• high dimension d = 3: Decolization phenomenon
In this paper, we study Anderson Localization in dy-
only exist when dimension d is higher than 3. In
namical analysis and also eigenstates calculation, dealing
addition, the critical disorder strength is dependent
with locator expansion and transfer matrices. Besides,
on the energy of the eigenstate.
easier ways like classical analogy and scaling theory can
also give intuitive explanations on the properties of local-
II. BASIC METHODS[6] ization in different dimensions. The spectrum continuity
and spectral statistics will not be introduced here.
Anderson localization is the result of quantum reflec-
tions in the lattice that make the wave function halt.
III. CLASSICAL ANALOGY OF RANDOM
Localization means the wave would decay and would not WALK
be scattered out of the lattice.
To determine whether localization happens in the lat-
tice, usually we can have four different methods. Anderson localization gives the theory of MIT (Metal-
Insulator Transition), which would only exist in 3D situ-
• Dynamical transport Place an electron, set its ations. In 1D and 2D systems, however, even tiny disor-
initial state and then wait. For delocalized states ders lead to localization. We can try to make sense of it
A brief introduction to Anderson Localization 2

using a naive but intuitive analogy from classical random Therefore,

walk.
Consider symmetric random walk on an one- hXi = ∞ ⇔ ρ = 1
(8)
dimensional lattice. At each time step, the particle jumps hXi < ∞ ⇔ ρ < 1
to the right with probability 21 and left with probability
1 It indicates that random walk starting from the origin
2 . The distance of each jump is l. would surely return in 1D situations.
As for 2D situations, similarly, we have P (x2n = 0) =
! N +  2
* N
X X N
X 1 2n
x2N =


xi  xj  = hxi xj i 42n n
i=1 j=1 i,j=1 We can get:
N (1) ∞ ∞
1X1
X X
x2i + 2


hxi i hxj i
X
= hXi = P (x2n = 0) ∼ =∞ (9)
i=1 1≤i<j≤N i=1
π i=1 i
= N l2
So the return probability of 2D systems is also 1.
2

 2 The story would Pbecome different yet interesting in 3D
It follows σN ≡ x2N − hxN i = 4N l2 q(1 − q) since systems. hXi3D ∝ i=1 1/i3/2 < ∞. So the return prob-
hxi i = 0. ability of 3D random walk ρ3D < 1. This naive classical
One can also compute the probability mass function random walk analogy yields the results consistent with
of xN . The probability that the particle jumps right k MIT theory from Anderson Localization.
times among the first N steps is
 
N 1 IV. BEHAVIOUR OF DISORDER-FREE AND
p(k, N ) = (2)
k 4k INFINITE DISORDER[8][6]

Perhaps it is more illuminating to consider the N → ∞

We have used the analogy of classiacl random walk to
limit. According to the central limit theorem, xN tends
2

intuitively sketch the Anderson Localization of different
to distribute normally xN ∼ N µN , σN :
dimension. Before we move to the rigorous and solid
methods such as 2nd -order pertubation theory or transfer
( x−µN )2 matrix, we can discuss the disorder-free case as well as
1 −
2σ 2
p(x, N ) = √ e N (3) the infinite disorder case.
2πσN

The probability that the particle returns to the origin A. Tight-binding Hamiltonian
after 2n steps is:

1

2n

1 The original model is tight-binding model in the lattice
P (x2n = 0) = 2n ∼√ (4) system, which we would also use here in this paper.
2 n πn

Define
P∞the indicator random variable Xn = 1 if xn = 0. 1. Spatial discretization
X = i=1 Xi counts the total number of returns. Then:
∞ ∞ ∞
X X 1 X 1
hXi = hXi i = P (x2i = 0) ∼ √ √ = ∞,
π i=1 i 1 2 X
i=1 i=1 (− ∇ + Vk (r − rn ))Ψ(r) = EΨ(r) (10)
(5) 2m
k
which diverges.
The expected number of returns hXi is related to the where rk is the position of k th atom, and the potential
probability of return. Denote ρ as the probability that is the superposition of each atom’s central potential.
the particle ever returns to the origin, and ρk as the prob- In the tight-binding approximation, the electron wave-
ability that the particle returns to the origin exactly k function is a linear combination of orbitals of adjacent
times. We have: atoms.

ρk = ρk (1 − ρ) (6) X
Ψ((r)) = φj χj (r − rn ) (11)
We have:
ρ where χj (r − rj ) denotes the orbital of the j th atom,
hXi = (7) and φn is corresponding amplitude. Therefore, this
1−ρ
A brief introduction to Anderson Localization 3

method is also known as ”Linear Combination of Atom where ki = 2π

L n, n = 0, 1, ..., L−1 is the i
th
component
Orbitals (LCAO)”. of the wave-vector on discretized lattice.
If we substitute the tight-binding wavefunction into the Corresponding eigenvalues can be easily computed:
Schödinger equation and integrate the equation, we can
d
get a compact algebric equation: X
Ek = −2g cos(ki ) (18)
X X i=1
[(j − E)An,j + Bn,k,j ]φj = 0 (12)
j k6=j
1. Return probability
where
If the initial state is localized at site j, the wavefunc-
Z
An,j = drχ∗n (−rn )χj (r − rj ) tion will be a δ-function.
Z (13)
Bn,k,j = drχ∗n (r − rn )vk∗ (r − rk )χ∗j (r − rj ) 1 X iki ji
Ψ(t = 0, j) = d e (19)
L
k
We neglect the overlap between the orbitals of neigh-
boring atoms An,j = δn,j . And we assume that the non- The different k component of the wavefunction will
vanishing element of the hopping integrals are only the evolve with time independently.
nearest-neigboring Bn,n,n+z , and the their values don’t 1 X iki ji +2itg Pdi=1 cos(ki )
fulctuate Bn,n,n+z = −g. Ψ(t, j) = e (20)
Ld
k
X
n Ψn − g Ψn+z = EΨn (14) In the thermodynamic limit, the series summation will
z be substituted by integration.
Therefore, the wavefunction in continuous Z Pd d
Y
space can be transformed into a discretized prob- Ψ(t, j) = eiki ji +2itg i=1 cos(ki )
= id Jji (2tg)
lem on lattce. k∈[0,2π]d i=1
(21)
Therefore
2. second-quantization
e−2idtg dd x −igx2
Z
Ψ(t = 0, j) ≈ e (22)
The Hamiltonian can be written in second- td/2 x∈Rd 2π d
quantization representation.
X we can then write the return probability and mean
H= {n |ni hn| − g[|ni hn + 1| + |n + 1i hn|]} (15) square displacement explicitly:
n
p(t) = |Ψ(t = 0, j)|2 ∼ 1/td (23)
{n c†n cn − g[c†n+1 cn + c†n cn+1 ]}
X
= (16)
n and
P
n n |ni hn| terms refer to on-site energy of a particle
and the sum of −g[|ni hn + 1| + |n + 1i hn|] is called hop- ∆x2t = 2dgt2 (24)
ping term, which moves the particle to its neighboring
We can see that when the time evolves to infinity, the
sites.
return probability decays to zero. Therefore, the state
In the following part, we will discuss the problem
will be delocalized in disorder-free case.
mainly on square lattice with periodic boundary condi-
In addition, what we should note that is the energy
tion.
spectrum of the disorder-free case is continuous in the
thermodynamic limit, which is an important criterion of
B. Disorder-free Case
the localization transition.

In the absence of disorder n = 0, the Hamiltonian C. Infinite disorder case

comes back to the free electron gas, which is translational
invariant. Therefore, the eigenstates of Disorder-free case
when disorder is large enough or hopping is small
are naive plane waves. g
enough, W can be seen as zero. we can write the eigen-
states of this case without calculation.
1
Ψk (j) = √ eiki ·ji , i = 0, 1, ..., d (17)
Ld Ψi (j) = δj,i , Ei = i (25)
A brief introduction to Anderson Localization 4

The corresponding return probability and mean square B. Resistivity

displacement are trivial.
1. Goal and Workflow
p(t) = 1, ∆x2t = 0 (26)
what we should note that is the energy spectrum of In this approach, our goal is to calculate the resis-
infinite disorder case is pure point spectrum. tivity of wave and prove that it (its expectation
value actually) grows exponentially with the num-
ber of sites in lattice (N ). We use transfer matrices
V. 1D ONE-DIMENSION and S-matrices to connect disordered sites and boundary
conditions. Then we get expression of hρi, the expected
To show localization in 1D lattice, here we use the value of resistivity.
transfer matrix method analytically to prove that Transfer matrices come from the Hamiltonian of the
the resistivity grows exponentially with the total system while S-matrices bring their physical insight of
number of sites in the lattice. reflection and transmission. The combination of them
can thus yield the expression of the resistivity of the sys-
tems expressed by the certain Hamiltonian.
A. Transfer Matrix in the lattice

Firstly, we setup relations between neighboring sites,

which are of vital importance for further derivation to
connect the sites as well as sites and boundaries.
Note an as the wavefunction amplitude of the N-th
lattice site.
For eigen-energy E, we have
Ean = n an + Vn,n+1 an+1 + Vn−1,n an−1 (27)
It can be written in the form of matrix equation:
! FIG. 1: Disordered lattice and non-disordered boundary con-
  E−n −Vn−1,n 
an+1 V V an dition
= n,n+1 n,n+1 (28)
an 1 0 an−1
! Consider electron flow shooting from the left into the
E−n −Vn−1,n random lattice. In this approach, we would show 1D
The matrix Vn,n+1 Vn,n+1
is called transfer ma- localization by proving that the resistivity of the electron
1 0
wave function, defined as ρ = R T where R and T refer
trix and noted as pn . With these transfer matrices, once to transmission and reflection rate, grows exponentially
we are given the wavefunctions of a pair of neighboring with N, the total number of random lattice sites.
sites, we can know that of the sites next to them. Itera- The key in this approach is its boundary condition that
tively, we can know the wavefunctions
Qn of all sites in the out of the lattice, there is no disorders so the wavefunc-
random lattice. Note PN = i=0 pi : tions there are just plane waves.
  n    
an+1 Y a1 a1
= pi = PN (29) 2. Derivation
an a0 a0
i=0

This leads to the recursion relations as follows: Here n is the label of a site in the lattice of disorder
and d is the width between neighboring sites:
E − n 11 Vn−1,n 11
Pn11 = Pn−1 − P
Vn,n+1 Vn,n+1 n−2
Pn21 = Pn−1
11
(30)
E − n 12 Vn−1,n 12
Pn12 = Pn−1 − P
Vn,n+1 Vn,n+1 n−2
Pn22 = Pn−1
12
FIG. 2: Plane waves outside the lattice
And if only diagonal disorders are considered, the
transfer matrix would just be
 E−
n
−1
 an = Aeikdn + Be−ikdn −∞ < n ≤ 1
pn = V (31) ikdn −ikdn
(32)
1 0 an = Ce + De n≥N
A brief introduction to Anderson Localization 5

From the wavefunctions above, the relation between E using the elements of PN . And the expectation value of ρ
and k can also be determined as E = 2V0 cos(kd), which (also the iterative relations for FN , FN −1 and fN , fN −1 )
comes from dispersion in 1D lattice with no disorder. can be derived from the recursion relations by squaring
Next, we relate two boundaries of the random lattice and taking the expactation values of Equation 30. Defin-
using the matrix T and express the resistivity ρ using the ing Fn = h(PN11 )2 i and fn = h(PN12 )2 i would give hρi a
matrix elements: neat look:
Define T :
1 1
   
B D hρi = [FN + FN −1 + fN + fN −1 ] − (40)
T = (33)
A C 4 2

Scattering matrix (S-matrix), which also relates two The relation yielded for FN and FN −1 (Similar for fN
boundaries of the random lattice and thus looks pretty and fN −1 ) is written as
similar to the transfer matrix T , would help us build the   2 1
h ih V 2 i hV 2 ih V12 i
  
FN FN −1
relationship between T and ρ, where r, t and r̄, t̄ respec- = (41)
FN −1 1 0 FN −2
tively give the reflected and transmitted amplitudes for
waves incident from the left and right.  2 1
h ih V 2 i hV 2 ih V12 i

Let R = . We would use its
1 0
eigenvalues and corresponding eigen vectors, as well as
 
r t̄
S= (34) the boundary conditions to calculate the expressions of
t r̄
Fn , Fn−1 in the expectation values of 2 and V 2 together
with the eigenvalues of R.
The properties of the eigenvalues of R are the key for
   
A B
S = (35) further derivation:
D C
r
Our reflection and transmission rate can be expressed 1 2 1 1 1 1
λ± = h ih 2 i ± ( h2 ih 2 i + hV 2 ih 2 i) (42)
using the elements of S-matrices: R = |r|2 and T = |t|2 . 2 V 2 V V
Relating T and S we have: The acuatual expression for the expectation value of ρ
T12 is as following. But we do not have to pay much atten-
r = S11 =− tion to its details since we only want to research on its
T11
exponentially growing behavior when N → +∞:
1
t = S21 = (36)
T11 h2 i
 
1
R hρi = 1 + 2 (λN N
+ − λ− )
⇒ρ = ∗
= (−T12 )(−T12 ) = |T12 |2 4(λ+ − λ− ) V0
T  2
hV i

2 1 1 N −1 −1
+ + V 0 h i + h 2
ih i · (λ+ − λN
− )
To know how the boundaries relate to the sites inside V02 V2 V2
the random lattice, we are supposed to find the relation- 1

1
N −2 −2
ship between T and PN . In the following +hV 2 ih 2 i · (λ+ − λN ) −
 ikN  −ikdequations
 we V −
2
e d 0 e eikd (43)
note θ as and Λ as :
0 e−ikN d 1 1 According to Schwartz Inequality of expectationvalues:
   −ikd ikd   −ikN d  
aN +1 e e e 0 D 1
= E(X 2 )E(Y 2 ) ≥ (E(XY ))2 ⇒ hV 2 ih i ≥ 1 (44)
aN 1 1 0 eikN d C V2
  (37)
−1 D Together with the condition that h2 i > 0, we have:
= Λθ
C
λ+ > 1, |λ− | < λ+ (45)
Similarly, we can also derive the relation between a1 , a0
and A, B Therefore, when N → +∞:
   −ikd ikd     
a1 e e B B hρi ∝ λN N ln λ+
= =Λ (38) + =e (46)
a0 1 1 A A

Recalling Equations 33, 37 and 38, we can get Since λ+ > 1, hρi grows exponentially with N as
N → +∞, leading to the localization of wave, no
T = θΛ−1 PN Λ (39) matter how big or tiny the randomness is.
As for the situation of diagonal disorders only,
Setting E = 0 here for simplicity and convinience, since hV 2 ih V12 i = 1. Still we have the same properties for
we have known that ρ is just T12 , ρ can be now expressed the eigenvalues, which lead to localization.
A brief introduction to Anderson Localization 6

C. Numeric results of 1D system 1. Exact Diagonalization

After analyzing the 1D model analytically, we can solve

the problem using three different numeric methods in- If the system is located on the orthogonal lattice with
cluding Exact Diagonalization (ED), transfer matrix and Periodic Boundary Condition (PBC), the Hamiltonian of
time propagation of wavepacket. the system can be written in the matrix representation.

a1 1 −g ... 0 −g a1
    
 a2  −g 2 ... 0 0   a2 
E  ...  =  ... ... ... ... ...   ...  (47)
    
a
L−1
  0 0 ... 
L−1 −g  aL−1 
aL −g 0 ... −g L aL

Then the Anderson Localization becomes an eigen-

value problem.
Here we set −g as 1, and sample the on-site energy n
from an uniform distribution [−W/2, W/2]. Then we can
get eigenstates with different disorder strength.
If the disorder strength W = 0, the eigenstates are
simple Bloch waves.

FIG. 4: Localized States: W = 0.5

FIG. 3: Bloch waves: W = 0

If the disorder strength W = 0.5, the states are local-

ized.
If we further improved the disorder strength W = 2.0,
the localization of the wavefunction were strengthened.
Apart from the naive eigenstates, we can use Inverse
Participation Ratio (IPR) to characterize the localiza-
tion. Here, we define IPR as:
X
I= |Ψ(j)|4 (48)
j

we can find that when the system is approaching ther-

modynamic limit, the correlation length versus energy
curve will converges to a stabe one. FIG. 5: Localized States: W = 2.0
A brief introduction to Anderson Localization 7

FIG. 6: Inverse participation ratio versus energy with differ-

ent system size: W = 2.0 FIG. 8: Correlation length versus energy with different system
size
If the IPR is higher, the states are more localized and
vice versa. Realizing the model twenty times to avoid If we can calculate the correlation length with more
fluctuation, we can calculate the IPR distribution over different disorder strength. We can find that if the energy
eigen energies with different disorder strength of the eigenstate is fixed, the correlation length scales as
Furthermore, IPR can be used to construct a phase ξ = α · W 2.
diagram in disorder strength W and energy E parameter
space. Unfortunately, because there is no delocalization
in 1D system, there is merely one phase in 1D case. 3. Time propogation

Dynamical process can also support this problem.

Here, we use the quantum dynamics to investigate the
time evolution of an wave packet.
Starting with initial state as δ function, we can cal-
culate the time-dependent Schödinger equation. When
t = 100s, we can get the wavepackets with different dis-
order strength.

FIG. 7: Phase diagram in disorder strength W and energy E

parameter space

2. Transfer matrix

The correlation length is another important informa-

tion in the Anderson localization.
By applying transfer matrix on the initial vector
(a1 , a0 ) thousands of times, we can get the average am- FIG. 9: Time evolution at t = 100s with different disorder
pltitude an , which can be used to calculate the correla- strength
tion length.
we can find that when the system is approaching ther- If the disorder strength is stronger, the wavepacket at
modynamic limit, the correlation length versus energy t = 100s is more localized.
curve will differ.
A brief introduction to Anderson Localization 8

VI. SCALING THEORY g → 0: In the limit of very strong disorder, wave-

functions will behave as exponentially localized
Now we finished the derivation of 1D Anderson local-
d ln(g) L
ization using transfer matrices. However, using similar g(L) ∝ e−L/ξ ⇒ = − = ln(g) (53)
approaches in 2D or 3D situations would be conceivably d ln(L) ξ
difficult. To deal with such difficulties, here we introduce Thus we have
a smart and easy way to investigate 2D and 3D lattices.
In this part L refers to the length of the lattice in one lim β(g) = ln(g) (54)
direction (sidelength) and A is the cross section: A = g→0

Ld−1 , where d is the dimension of the lattice. Define the

Having the limiting behavior of β(g) for g → 0 and
conductance g as the inverse of resistivity we have defined
g → +∞,a plot can be made (FIG10), where the function
before: g = ρ1 , which is related to the transmission in one
we used in plotting is β(g) = (d − 1) − (1 + g) ln(1 + g −1 ).
direction as According to the limiting behavior and the plot of β(g):
1 T (L) T (L) For 1D and 2D systems, β never exceeds x-axis so
g= = = (49) the conductane would always decrease under the exten-
ρ R(L) 1 − T (L)
sion in one direction since β(g) = dd ln
ln g
L is always nega-
Scaling Theory methods applying to Anderson Local- tive. In the limit of L → +∞, there would be localization
ization dates back to “Gang of four”: Abrahams, Ander- since its conductance would approach 0.
son, Licciardello, and Ramakrishnan[4], where the crucial For 3D systems, however, it is special that there
step was to claim that the conductance at changed scale is an turning point gc where β(gc ) = 0, indicating the
is the function of its previous scale and the scale itself, behavior of transition from insulator to metal. Therefore,
which can be expressed as localization would only exist with large disorder. As for
tiny disorder, the density of wavefunctions may not decay
g(nL) = F (n, g(L)) (50) to cause localization. gc refers to a certain degree of
disorder connecting insulator and metal.
Define β = dd ln(L)
ln(g)
to show how g would change when
tuning L. And we would research on the limiting behav- VII. PERTUBATIVE METHODS ON HIGHER
ior of β when g is extremely large or small to get the DIMENSION
properties of the conductance in 1/2/3D situations.
Apart from the transfer matrix, some advanced per-
tubative methods such as locator expansion and forward
approximation were proposed in past several years.

A. Locator expansion for the resolvent

The locator expansion was firstly proposed by Ander-

son, which used Random Walk representation to expand
pertubatively the resolvent at the vicinity of infinite dis-
order case (compeletly localized state).
The non-pertubative resolvent is
1
G(E) := (55)
E−H
1
where G(a, b, E) = ha| E−H |bi is the localized matrix
FIG. 10: Scaling Theory
element.
If we expand the resolvent at the vicinity of the com-
g → +∞: In the case of a good conductor g  1, peletely localized state (W −→ ∞), we can get expansion
according to Ohm’s law series similar with the Dyson series.
L L
R=ρ = ρ d−1 ⇒ g = σ0 Ld−2 (51) +∞
A L X n
G = G0 + g n (G0 T ) G0 (56)
Then we can easily obtain n=1

d ln(g) Using different approximations/representations, we

=d−2 ⇒ lim β(g) = d − 2 (52)
d ln(L) g→+∞ can get different expansions.
A brief introduction to Anderson Localization 9

FIG. 11: RW: naive pertubation expansion for the resolvent vetween a and b; SAW: Self-avoid random walk representation
using renormalized pertubation theory; DP: Forward approximation merely considering shortest path [2]

Naive Random Walk

+∞ n
δa,b 1 X X Y 1
G(b, a, E) = + (−g)n
E − a E − a n=1 s=1
E − π(s)
π:a→b,|π|=n
(57)
Self-avoiding Random Walk
n
X X Y 1
G(b, a, E) = (−g)n P{π(0),··· ,π(s−1)}
SAW π:a→b,|π|=n s=1 E − π(s) − π(s) (E)
(58)
FIG. 12: An eigenstate of 2d lattice

B. Forward Approximation

If we only keep the shortest path in our expansion, we

can get a more rough expression.
n
X Y 1
ψα (b) ' (−g)n (59)
 − π(s)
s=1 α
DP π:a→b

It is called the Forward Approximation.

This was a approxiamted expansion of the resolvent,
therefore the localization might be under-estimated dur-
ing our calculation. For example, the critical disorder FIG. 13: An eigenstate of 2d lattice
strength in 2D case calculated by Forward Approxima-
tion is not zero Wc |2d 6= 0.
2. 2D lattice wavepacket evolution

C. Numerci results of Higher dimension Apart from the eigenstates, the time-evolution of the
wavepacket can also be calculated.
1. 2D lattice eigenstates

Similarly for 2D, a 20 × 20 lattice, periodic boundary 3. 3D lattice eigenstates

conditions are applied to make the 2D lattice a torus.
We still choose E0 = 10. Again, two eigen-states are From the Scaling Theory part we have seen that 3D sit-
randomly chosen and presented in FIG 12 and 13. uation is special because of the existence of turning point
The shape of resulting eigen-states looks like Bugles, indicating that there would be a transition from insula-
showing the behavior of localization. tor to metal. It is stated[? ] that the critical disorder
A brief introduction to Anderson Localization 10

value depends on the band energy E. So in the simulation

where we tune the randomness trying to show the transi-
tion, we chose the eigenstates of nearly the same energy
(eigenvalues),which is the one nearest to 0 in different
randomness.
FIG. 17 shows the situations where E0 is respectively
1,10 and 25. We can see the eigenstates transition from
delocalization to localization. But the specific turning
point cannot be determined here.

VIII. EXPERIMENTAL OBSERVATION

Billy et al.[3] are the first to directly observe expo-

nential spatial localization of matter-waves. They re-
ported the observation of exponential localization of a FIG. 14: Wave packet time evolution of 2d lattice whith W =
Bose-Einstein condensate released into a one-dimensional 2.0 at t = 50
waveguide with controlled disorder created by laser
speckle. Here we briefly introduce their experiments.
They operated weak disorder such that localization
would result from quantum reflections instead of bounded
by huge potential basin and walls. Besides, they con-
trolled the atomic density to be very small thus the in-
teraction can be neglected.
In a horizontal optical waveguide providing strong
transverse confinement, a magnetic longitudinal trap and
a weak disorderd optical potential are applied. In the be-
ginning, a Bose-Einstein condensate is formed in the trap.
When the longitudinal magnetic trap is turned off, the
Bose-Einstein condensate begins to expand then becomes
localized.
In FIG18, a) is the beginning state of trapped Bose-
Einstein condensate and b) is when the magnetic trap
goes off and it is localized. c) and d) show the original
and fitting curve of the atomic density, where pink lines
refer to the initial state and red lines refer to observed FIG. 15: Wave packet time evolution of 2d lattice whith W =
localization. Logarithmic y coordinate is applied in d), 3.0 at t = 50
showing that the density decays exponentially.
Apart from the matter wave observation, anderson lo-
calization of solitons are also overserved, which is a very
interesting phenomenon.[11]

FIG. 16: Wave packet time evolution of 2d lattice whith W =

10.0 at t = 50
A brief introduction to Anderson Localization 11

FIG. 17: 3D lattice eigenstate: Left: E0 = 1. No localization

is seen. Middle: E0 = 10. Weak wavefunction localization
can be seen. Right: E0 = 25. We can see strong localization. IX. SUMMARY

In the whole paper we made exploration of Anderson

Localization. We have done the derivation of 1D using
transfer-matrices and used scaling theory to show the
difference of 1/2D and 3D systems: Only in 3D lattices
we can have transition from metal to insulator. Besides,
we did simulation of 1, 2 and 3D localization to visual-
ize our results of scaling theory. Finally we introduced
an experimental observation of Anderson localization in
BEC. Actually there are far more fancy content related
to Anderson localization. For example, anderson local-
ization can be observed even in light[10]. Sixty years have
FIG. 18: Anderson localization of a Bose-Einstein condensate, witnessed the importance of Andserson localization, yet
taken from [3] it would still yield brand new discoveries.

[1] Anderson, P. (1958). Absence of Diffusion in Certain https://doi.org/10.1088/0022-3719/4/10/023

Random Lattices. Physical Review, 109(5), pp.1492- [7] Anderson, P. W., Thouless, D. J., Abrahams, E., &
1505. problem 10.58, page 481. Fisher, D. S. (1980). New method for a scaling the-
[2] http://nanotheory.itp.ac.ru/teaching/Basko042015/lecture1 ory of localization. Physical Review B, 22(8), 3519–3526.
.pdf, Anderson localization in 3D, locator expansion https://doi.org/10.1103/physrevb.22.3519
[3] Billy, J., Josse, V., Zuo, Z., Bernard, A., Ham- [8] Dominguez-Adame, F., & Malyshev, V. A. (2004). A sim-
brecht, B., Lugan, P., . . . Aspect, A. (2008). Direct ple approach to Anderson localization in one-dimensional
observation of Anderson localization of matter waves disordered lattices. American Journal of Physics, 72(2),
in a controlled disorder. Nature, 453(7197), 891–894. 226–230. https://doi.org/10.1119/1.1593660
https://doi.org/10.1038/nature07000 [9] Localization and low temperature transport in disordered
[4] Abrahams, E., Anderson, P., Licciardello, D. and Ra- one-dimensional systems, Alfred Douglas Stone, Doctor-
makrishnan, T. (1979). Scaling Theory of Localization: ate in Philosophy, Massachusetts Institute of Technology,
Absence of Quantum Diffusion in Two Dimensions. Phys- Jan. 1983.
ical Review Letters, 42(10), pp.673-676. [10] Segev, M., Silberberg, Y., & Christodoulides, D. N.
[5] Thouless, D. J. (1972). A relation between the density of (2013). Anderson localization of light. Nature Photonics,
states and range of localization for one dimensional ran- 7(3), 197–204. https://doi.org/10.1038/nphoton.2013.30
dom systems. Journal of Physics C: Solid State Physics, [11] Sacha, K., Müller, C. A., Delande, D., & Za-
5(1), 77–81. https://doi.org/10.1088/0022-3719/5/1/010 krzewski, J. (2009). Anderson Localization of
[6] Herbert, D. C., & Jones, R. (1971). Local- Solitons. Physical Review Letters, 103(21).
ized states in disordered systems. Journal of https://doi.org/10.1103/physrevlett.103.210402
Physics C: Solid State Physics, 4(10), 1145–1161.