Sie sind auf Seite 1von 13

Mech 506

Exploration of Anderson localization


using n-pendula model

Author:
Miayan Yeremi

Student #:
18213074

December 5, 2014

Introduction

It is well known that vibration propagate through coupled lattice system such as phonons through a
solid. Electron conductivity can also be thought of in a similar manner. Simplistically speaking, one
can view the electron as a wave propagating on a periodic lattice of atoms. In a paper, published
by Anderson in 1958, Anderson localization was first introduced. Anderson localization describes
the phenomena of vibration localization on lattices with weak disorder (i.e. small irregularities in
the periodic lattice) and weak coupling. This phenomena helped explain reduced conductivity in
some solids, and eventually led to a Nobel prize in Physics.
Experiments exhibiting Anderson localization in solid state physics field can be quite complicated
and not accessible with out specialized equipment. Anderson localization can actually also be
observed in mechanical system such as masses on a string, or more complicated systems such as,
shrouded blades of jet engine rotors.
A paper published by Hodges described the theoretical analysis and experimental setup of a fairly
simple example of Anderson localization in a mass on a string system. In addition, to describing
the experimental setup Hodgess paper outlined a theoretical connection between the masses on a
string system and a system of coupled pendula (An example of two coupled pendula is shown in
Figure 1). In the next sections of my report, I will re-derive the connection between the system of
coupled pendula and the masses on a string. Next, I will conduct a thorough analysis of the system
of pendula both in strong/weak coupling regimes with weak disorder ( disorder is introduced by
slight variation of the pendulas nominal length ). I will use perturbation methods to analyze the
system of pendula, and validate the perturbations validity by comparing my results to the exact
answer, which I will also derive and compute using Matlab.
Lastly, I would like to mentioned that I borrowed heavily out of the three references I provided
at the end of the report. I used Pierres paper to guide my study of the n-pendula system using
perturbation methods. I used Hodgess paper to come up with the derivation for the connection
between the pendula system and the masses on a string system. Lastly, I used Hinchs book to work
out any issues I had with understanding the perturbation methods presented in Pierres paper.

Connection between masses on a string and a system of


n-pendula

In this section I will present a derivation of the equations of motion (EOM) of a coupled n-pendula
system in the frequency domain, in order to make a direct analogy to the system of n heavy masses
on a light string. In the proceeding sections I will re-derive the EOM in the time domain. Imagine
we have n-pendula and we want to write the EOM of an interior pendulum with a natural frequency
of i . Lets label the displacement of the is pendulum as xi = Li i , where Li = L(1 + Li ). L is a
nominal length of the pendula and Li the length deviation from the nominal value L. Lastly, let
k be the coupling parameter, and assume equal masses for all the pendula.
So if we proceed to right the EOM of the ith interior pendulum, assuming small pendulum
displacement, we get,
m
xi = k(xi+1 xi ) k(xi1 xi )

mgxi
L(1 + Li )

Lets divide by m, convert to frequency domain, and notice that the natural frequency of the ith

g
pendulum is i = L(1+L
. In addition, lets label the frequency domain analogue of displacement
i)
k
. From this we get:
as ci and V = m

2 ci = (2V + i )ci + V ci+1 + V ci1


Lastly, if we label Ui = 2 2V i . We get the exact equation presented in Hodgess paper.
V ci+1 + Ui ci + V ci1 = 0

(1)

Lets turn our attention to the system of masses on a string. We have to discuss the analogies
between the two systems, and in addition discuss what it means to have coupling in the string
system. So, lets start out making a couple of observations. If the masses on the string were
infinitely massive then the pieces of string in between masses, which I will refer to as bays, will not
feel the effect of vibrations of the neighbouring bays. This is like having no coupling in the system.
So heavy masses is equivalent to weak coupling between bays. It is also worth mentioning that
the bays are the analogues of the pendula. Finally, we can guess the mode shapes of the masses
on a string system. We can say the masses vibration amplitude is small compared to the string
amplitude vibration. The masses can be thought of as being displaced from each other by straight
lines, and the bays can be thought of as having sinusoidal mode shapes. So, if we say the masses
vibration amplitude is bi and the bays vibration amplitude is ai , where bi << ai , we can write the
guessed mode shapes as follows:
vi (x) = ai sin(

x
nx
+ bi1 + (bi bi1 )
L(1 + Li )
L(1 + Li )

Now we can calculate potential energy of the given mode as follows,


U=

Z
Z n
t X L(1+Li ) 0
(nt) X
(ai )2
(cos(x))2 dx
(vi (x))2 dx
2
2
L(1
+
L
)
i
0
bays 0
bays
n2 2 t X
(ai )2
=
,
4
L(1 + Li )
bays

where t is the tension in the string. We can get away with this approximation because bi << ai .
Next we have to calculate the corresponding modes kinetic energy. The approximation in this part
is a little bit more involved, because we have to consider several small and large parameters. bi s
and (i.e. string/bay density) are small, whereas ai s and m are large. Keeping this in mind the
kinetic energy looks like,
Z
X L(1+Li )
m X
T =
(vi (x))2 dx +
(bi )2
2
2
0
bays
bays
=

Z
X L(1+Li )
nx
nx
bi bi1
((ai )2 (sin(
))2 +2ai (sin(
)(bi1 +
x)+O((bi1 )2 , (bi1 )2 )dx
2
L(1
+
L
)
L(1
+
L
)
L(1
+
L
)
i
i
i
bays 0
m X
+
(bi )2
2
bays
2



2ai (bi1 (1)n bi )
m X
X
(ai )2

L(1 + Li )
+
+
(bi )2
2
2
n
2
bays
bays
Lastly, we apply Rayleigh Ritz method by writing the following two equations:
2 T )
(U 2 T )
= 0, and (Ub
= 0. From these equations we get the following,
ai
i



n2 2 tai
ai
bi1 (1)n bi
2
L(1 + Li )
+
=0
2L(1 + Li )
2
n

(2)

(ai+1 L(1 + Li+1 ) (1)n ai L(1 + Li )) + mbi = 0


(3)
n
Equation 3 can be rearranged to solve for bi , which in turn can be stuck into Equation 2 to
produce an equation just in terms of the ai coordinates, which in our analogy to the pendula are
the displacements of the pendula.
Finally, if we group and label the parameters and variables above as follows:
2 2 L
n2 2 t
n
i2 = L2 (1+L
2 , V = n2 2 m , and ci = (1) ai L(1 + Li ).
i)
We get the following equation:
2 i2
ci = V (2ci ci1 ci+1 )
(1 + Li )
which to a very good approximation is the same as what we got in Equation 1, because |Li | <<
1. This proves that the weakly coupled/weakly disordered n-pendula model in relevant to the study
of heavy masses on a light string. In the proceeding sections I will give an extensive analysis of the
n-pendula model, starting with a simple case of two pendula for both strong and weak coupling.

Figure 1: Two pendula example

Two pendula example

First I will start by deriving the equations of motion (EOM) for the simple two pendula example.
After, this derivation I will generalize the equation to n number of pendula with appropriate
boundary conditions.
Start by writing out the kinetic and potential energies of the system.
T =

m
m
(L1 )2 + (L2 )2
2
2

k
(L1 L(1 + L)2 )2 + mgL(1 cos(1 ) + +mgL(1 + L)(1 cos(2 )
2
Now from the kinetic and potential energies, I applied Lagranges equations to arrive at the
following EOM (note, I assumed that the angles are small in order to linearise the equations).
U=

mL2 1 + kL(L1 L(1 + L)2 ) + mgL1 = 0


mL2 (1 + L)2 2 + kL(1 + L)(L1 L(1 + L)2 ) + mgL(1 + L)2 = 0
Next, I write the EOM in matrix form.



 
k
k
+ gl )
(1 + L)
( m
d2
1
1
m
=
k
1
k
1
( m
+ gl (1+L)
)
2
dt2 2
m (1+L)
From the above equation it is trivial to write down the characteristic equation used to find the
eigenvalues of the system.
4

4
1,2


+

1
2k g g
+ +
m
l
l (1 + L)

2
1,2
+

gk
gk
g2
+
+ 2
=0
Lm (1 + L)m L (1 + L)
2

It is convenient to non-dimensionalize above equation by dividing above expression by Lg 2 and


2

kL
2
2
. After, non-dimensionalization 1,2
becomes
1,2
= 1,2
defining R2 = mg
g . Next, it is easy to show
L
using the quadratic equation that,
s
2
1
1
2
2
2
1,2 = 2R + 1 +
.
(4)
4R4 + 1
+
(1 + L)
(1 + L) (1 + L)2

At this point, I would like to introduce some terminology. I am studying the effects of weak
disorder, which means small values of L, and both cases of strong/weak coupling, which can be
viewed in the non-dimensional form as large/small values of R2 .
The two regimes I am interested in, lend themselves well to Taylor expansions of equation 4.
Lets Taylor expand the strong coupling case, which gives:


(L)2
L (L)2
2
+
R2 1 +
+ O(L3 ).

1,2
= 1 + R2
2
2
8R4
This approximation works well for strong coupling, where R2 is large, but it is worth noticing
that in the weak coupling regime this approximation breaks down. Therefore, the moral of this
story is that the smallness of R2 has to be considered as well in the weak coupling case. In the next
sections, when I generalize these ideas to n pendula I will treat the perturbation methods in both
regimes differently to make sure I do not run into divergent solutions when switching between the
strong and weak coupling cases.

Exact Equations of Motion for n-pendula

The first thing I will present is an extension of the equations of motions (EOM) arrived at in Section
3. Note, I am assuming small angular values through out entire discussion.
To start out, I will write the EOM for a pendulum in the interior of n-pendula. An interior
pendulum has two pendula on either side of it, so the equations presented in Section 3 can be
generalized to:
mL2 (1 + Li )2 i + kL(1 + Li )(L(1 + Li )i L(1 + Li+1 )i+1 ) +
+ kL(1 + Li )(L(1 + Li )i L(1 + Li1 )i1 + mL(1 + L)gi = 0
From here, I non-dimensionlize (i.e. same method used in Section 3) above equation and write
it in matrix form:
d2 ~
~
= A,
dt2

R2 (1 + Li1 )
1
R2 (1 + Li+1 )
2
A=B
; 2R +
;
.
1 + Li
1 + Li
1 + Li
B is a tridiagonal banded matrix. Lastly, I assumed fixed-fixed boundary conditions.


Strong coupling weak disorder n-Pendula Perturbation

In order to develop a strong coupling perturbation we take the equations of motion (EOM) provided
in Section 4 and Taylor expand using Li as a small parameter. The two terms that need to be
expanded are,
1
= 1 Li + (Li )2 + H.O.T.
1 + Li
1 + Lj
= 1 + (Lj Li ) + (L2i Li Lj ) + H.O.T.
1 + Li
H.O.T stands for higher order terms. Let Li = Li , where Li is order one and  << 1.
With these expansions we can create the following second order perturbation expansion.

A0 + A1 + 2 A2 + x = x

A0 = B R2 ; 2R2 + 1 ; R2 .

A1 = B R2 (Li1 Li ) ; Li ; R2 (Li+1 Li ) .

A2 = B R2 (L2i Li Li1 ) ; L2i ; R2 (L2i Li Li+1 ) .
x = x0 + x1 + 2 x2 +
= 0 + 1 + 2 2 +
A0 , x0 , and 0 are the unperturbed 0th order problem that is easy to solve, and assume as
known. In the next part I will present the theoretical method of solving the perturbation problem
assuming the zero order problem is known. To show that this method is working well, I will use it
to get a perturbation solution for the ground state of a n = 80 pendula system with weak disorder
and strong coupling and compare it to the exact solution.
To solve the perturbation problem we have to work our way up order by order. So assuming we
know the answer to the zeroth order we turn our attention to solving the first order problem.
O() : (A0 0i )x1i = A1 x0i + 1 x0i .
Next, multiply by the right handed eigenvector of A0 , x0i , which gives,
x0i A1 x0i + 1 x0i x0i = 0,
and from this we get that,
1i =

x0i A1 x0i
x0i x0i

Next, lets compute the eigenvector 1st order correction.


(A0 0i )x1i = A1 x0i +

x0i A1 x0i
x0i x0i

x0i .

The only way this equation can be solved is if x1i is orthogonal to x0i , which is what we implicitly
imposed when solving for 1i . So to solve for x1i we can just find its projections on to the rest of
the eigenvectors of the unperturbed problem. If we do this we get the following,
x1i =

X
j

x0j A1 x0i
1
x0j i 6= j.
(0i 0j ) x0j x0j

Finally, we proceed to the second order perturbation problem.


O(2 ) : (A0 0i )x2i = A1 x1i + 1i x1i + 2i x0i A2 x0i
Again, multiply by x0i and solve for 2i , which gives,
2i =

x0i (A1 x1i + A2 x0i )


x0i x0i

By a very similar process as described for the first order perturbation we get the following for
the second order correction for the eigenvector,
x2i =

X
j

x0j (A1 x1i + A2 x0i )


1
x0j i 6= j.
(0i 0j )
x0j x0j

There is one slight issue I have been ignoring here. I have been implicitly assuming there are
no multiple eigenvectors with the same eigenvalues. If there is a degeneracy in the eigenvalues than
the equation above have to be modified.
So how well is this perturbation method actually working? Lets try an example with Matlab.
Like I mentioned previously, I calculated the correction to the ground state of a system with
strong coupling and smaller disorder. In this example I wanted to demonstrate that the perturbation
method performs well even if the disorder is to weak. Now, because the equations I am working
with are non-dimensionalized I can compare the number I used for the order of the disorder and the
order of the coupling. I set R2 = 1 and epsilon = 0.1 so the disorder is only one order of magnitude
smaller than the coupling parameter. In addition, I randomly and uniformly set the changes in
pendula lengths using a uniform distribution. The eigenvalues I got were as follows:
exact = 1.1779
0i = 1.0015 0th order perturbation
1i = 1.1533 1st order perturbation
2i = 1.1750 2nd order perturbation
Now, lets see how the method performed in terms of guessing the eigenvector correction, see
Figure 2.
The second order perturbation approximated the exact solution very closely, even for a system
with fairly large disorder. Another interesting observation is that even though the system is fairly
disordered there is still transmission through the system at the zero harmonic (i.e. the mode spans
through all the pendula and is not localized). This as we shall see in the next section is not true for
system with weak coupling. This difference in behaviour was hinted at in Section 1, when it became

Strong coupling weak disorder (n=80)


0.25
0th order perturbation
exact solution
1th order perturbation
2th order perturbation

0.2

0.15

0.1

0.05

10

20

30

40

50

60

70

Figure 2: Perturbation eigen vectors vs. Exact solution (strong coupling)


clear that the same approximation methods could not be used for both weak and strong coupling.
In the weak coupling weak disorder regime we actually get Anderson localization, as opposed to
global mode transmission.

Weak coupling weak disorder n-pendula

For the weak coupling weak disorder system, we have to rethink the perturbation scheme. What
this amounts to is a different Taylor approximation of the exact equations presented in Section 4.
We still have the same structure for the perturbation scheme,

A0 + A1 + 2 A2 + x = x
Except the A0 , A1 , and A2 matrices are different. In addition, I am relabelling the coupling
R2
2
2
= R2 , where R2 is order one, and I am assuming Lwi is order one. I
parameter to Rw
, and Rw
am also assuming all the Li are unique.
A0 = diag (1 + Li ) .

A1 = B R2 ; 2R2 ; R2 .

A2 = B R2 (Li1 Li ) ; L2i ; R2 (Li+1 Li ) .
x = x0 + x1 + 2 x2 +
= 0 + 1 + 2 2 +
8

80

Weak coupling weak disorder (n=20)


1
0th order perturbation
exact solution
1st order perturbation
2nd order perturbation

0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1

10

12

14

16

18

20

Figure 3: Perturbation eigen vectors vs. Exact solution (weak coupling)


The same exact theoretical frame work, as presented in Section 5, goes into solving the perturbation problem in this section.
2
For this example, Rw
, and Li were on the order of 0.01, I found that anything bigger than
this order seemed to worsen the performance of the perturbation; in addition, the exact solution for
2
, and Li was less localised. This lead me to conclude that 0.01 was the order
bigger values of Rw
were I could assume weak coupling and weak localization. I used 20 pendula and assigned their
deviations in length randomly, using a uniform distribution. The eigenvalues I got were as follows:
exact = 1.0443
0i = 1.0238 0th order perturbation
1i = 1.0438 1st order perturbation
2i = 1.0441 2nd order perturbation
Figure 6 displays the results of the eigenvector perturbation corrections. After running this calculation for several random length deviation distribution I noticed that the first order perturbation
works quite well in the weak coupling weak disorder parameter regime. I believe that this is do to
the localised nature of this problem.
Next, I will briefly discuss the affects of having two of the length are equal.

Weak coupling weak disorder n-pendula with degenerate


length

An interesting case arises if we allow m (m < n) of the pendula to have equal lengths. In order to
understand what happens when m of the pendula have equal length, it is most convenient to start
with the first order perturbation matrix.
A0 = diag (1 + Li ) .
If two of the pendula length are the same than this matrix has degenerate eigenvalues of multiplicity m. The problem with the ordinary perturbation method is that when we go to solve for the
eigenvectors we have to divide by the difference of the unperturbed eigenvalues. This is of course
problematic if two, or more of the eigenvalues are the same.
What we have to do is approach the perturbation problem a little differently. Lets just look at
the first order problem.
(A0 + A1 + ) x = x,
where this time we will write x a little differently (note, I am performing an expansion for the
eigenvectors that are associated with the degenerate eigenvalue).
X
x=
j x0j + x1 + , where j runs over the degenerate eigenvectors.
j

= 0 + 1 +
When we proceed to solve this problem we actually have introduced multiple unknowns (i.e.
j s), but we can use the multiple x0j s to come up with multiple equations which allow to solve for
the j s and the new multiple eigenvalue corrections. In the process of solving for the j s and first
order eigenvalue corrections we get the following new eigenvalue problem.


x01 A1 x02
x01 A1 x0m
x01 A1 x01

x01 x01
x01 x01

x01 x01

~
~ = 1

x0m A1 x01
x0m x0m

x0m A1 x02
x0m x0m

x0m A1 x0m
x0m x0m

Now that we have the m


~ , and s we can proceed with the perturbation to higher orders. The
first order corrections will usually lift the degeneracy in the unperturbed problem. The lifting of degeneracy can be seen in many fields in physics. For example, degeneracy lifting due to perturbation
is seen in fine structure splitting, Zeeman effect, Stark effect, etc.
In the pendula systems I am considering there is an interesting effect taking place due to the
structure of the perturbing matrices, more specifically the perturbing matrices are tridiagonal.
Because the perturbing matrices are tridiagonal the machinery develop above only has to be
used for certain orders of perturbation corrections depending on the locations of the equal length
pendula. Essentially if the spacing between the two equal pendula is k pendula then the degenerate
perturbation method only needs to be employed at the k th order perturbation. This suggest a scaling for the amplitude transmission between two degenerate pendula. The amplitude transmission
10

nearest neighbour degeneracy (deg.)

next (n.) nearest neighbour deg.


1
eigenvector location

eigenvector amplitude

1
0.5
0
0.5
1

5
10
pendula location

0.5
0
0.5
1

15

n. n. nearest neighbour deg.


0.5
eigenvector amplitude

eigenvector amplitude

15

n. n. n. nearest neighbour deg.

0.5

0.5

5
10
pendula location

5
10
pendula location

0.5

15

5
10
pendula location

15

Figure 4: Degenerate pendula at different distances from each other


might take on an order that is proportional to k . To be more precise, I am defining amplitude
of transmission as magnitude of vibration felt by one pendulum do to an excitation in a different
pendulum.
I did not pursue this idea any further do to time constraints, but I did use the exact solution to
see if the amplitude of transmission between two degenerate pendula decreases as the degenerate
pendula are moved further apart from each other (this is not very surprising physically speaking).
Figure 4 shows the results of this experimentation. I positioned the first pendulum at the beginning
(i.e. location 1)of the pendula chain and then move the degenerate pendulum further and further
away from location 1. The results do suggest some amplitude of transmission scaling with distance
between degenerate pendula, that is proportional to .

Future Work
I think the most interesting extension of this project would be to set the experimental portion
presented in Hodges paper and observe Anderson localization in the lab.
I would also like to further investigate the scaling law for the transmission of two, or more
degenerate pendula with the parameter 
Lastly, if I could work the previous part out, I think it would be interesting to come up with
an experimental setup validating the scaling proportionality.

11

References
[C. Pierre, E. H. Dowell, 1986] C. Pierre, E. H. Dowell (1986) Localization of vibration by structural irregularity.
Journal of sound and vibration.
[C. H. Hodges, J. Woodhouse, 1983] C. H. Hodges, J. Woodhouse (1983) Vibration isolation from irregularities in
a nearly periodic structure: Theory and measurements. Journal of acoustics.
[E. J. Hinch ] E. J. Hinch Perturbation Methods.

12

Das könnte Ihnen auch gefallen