8.

03 at ESG

Supplemental Notes

Coupled Linear Oscillators

You need not memorize any of this, but please dont toss it; well be needing
some of the results soon.
In these notes, matrices will be denoted by boldface captial letters.
For a system of coupled oscillators subject to linear restoring forces, we found
that the equations of motion many be written in matrix form;
(t) = K
x(t),
M
x

(1)

where
x(t) will be taken as a column vector, each of whose components is one of
the dependent variables, M is a diagonal matrix with positive elements, and K is
symmetric. For example, we have seen that for the simple system of two masses
connected by springs, as shown,

m1

the coupled equations

m2

1 = kx1 k
(x1 x2 )
m1 x
m2 x
2 = kx2 k
(x2 x1 )

became

m1
0

0
m2



x
1
x
2

k k

=
k

k

k k




x1
.
x2

It will be helpful to use row vectors also; if

x1
.

x = .. ,
xT = [x1 , . . . , xn ] .
xn
The T superscript is for transpose. From the symmetry of M and K, note that

xT M = (M
x)

,
1

xT K = (K
x) .

Note also that
xT
x is a scalar; in fact,
xT
x =
x
x.

So, lets look for solutions to (1) of the form
= 2
x.

x

(2)

Note that, if
x0 is a constant vector,
x =
x0 eit ,
x =
x0 eit ,
x =
x0 cos t,

x =
x0 sin t or any linear combination will suce. Note that in this usage,
x0
is not necessarily
x(0). The specic functional form wont matter, as long as (2)
holds.
Substitution into (1) yields



K + 2 M
x =
0.

(3)



The only way for non-trivial solutions of (3) to exist is to have det K + 2 M = 0.
This allows us to solve algebraically for 2 , and hence for
x0 (within constant
multiples). Specically, let
x =
x0 ei t . Now, then, what are meant by normal
modes? Who are we to judge anyone elses normalcy? This is, after all MIT. Well,
consider





xT0 K
x0 =
xT0 (K
x0 ) =
xT0 2 M
x0 = 2
xT0 M
x0 .
But,

So, we have



T

xT0 K
x0 =
xT0 K
x0 = (K
x0 )
x0

T
T
= 2 M
x0
x0 = 2 (M
x0 )
x0


= 2
xT0 M
x0 .




2
xT0 M
x0 = 2
xT0 M
x0 ,


 2
2
xT0 M
x0 = 0,

or

which means that if 2 = 2 ,

xT0 M
x0 = 0,
and this is the interpretation of normal modes that we need.
If 2 = 2 for some = , the above relation still holds, if
x0 and
x0
are chosen properly. First, note that if 2 = 2 , (3) is satised for 2 = 2 and

x = a
x0 + b
x0 for any scalars a and b. Thus,
x0 and
x0 may be chosen such
that
xT0 M
x0 = 0 if
x0 and
x0 are linearly independent, and a result from linear
2

algebra shows that they will be. However, the physics shows why the vectors
x0
and
x0 may always be chosen to be normal.
The specic values of 2 and the components of
x0 will depend on the values of
the masses (i.e., the diagonal elements of M). If any frequency appears as a multiple
root of (3), changing a value of some mass slightly will remove the degeneracy, the
2 will be dierent, and the modes will be necessarily normal. A small change in
the elements of M cannot change the normality of the vectors. Please note that this
explanation does not constitute a proof; it relies on our belief, on physical grounds,
that a small change in the linear system cannot grossly aect the normalcy criteria.
We saw in class that if the masses are all the same, M is a scalar multiple of
the indentity matrix, and the normality condition reduces to
x0
x0 = 0. We also
saw that if the masses are not identical, the modes are normal when weighted by
the masses.
Before moving on, lets take advantage of the physics to help us do some math;
1

dene the matrix M 2 as the diagonal matrix whose elements are the non-negative
1
1
square roots of the masses, so that M 2 M 2 = M. Then, rewrite (1) as
1

= M 2 M 2
x
= K
x = K M 2 M 2
x
M
x

or

1
= M 12 K M 12 M 12
x,
M 2
x
1
1
1
. This last point may seem trivial, but fractional powers of
where M 2 = M 2
matrices are not as easily dened or determined as for scalars; our form for M makes
1
1
it easy. Also, note that if the masses are not the same, M 2 K = K M 2 . So,
1

dene new coordinates by
y (t) = M 2
x(t). With K
M 2 K M 2 , (1) becomes

y = K

y .

(4)

Our conditions for normal modes are then



det K
+ 2 I = 0,

y0
y0 = 0.

Note that this is a mathematical convenience; if the x1 , . . . , xn represent lengths,

changing to y1 , . . . , yn changes each length by a dierent factor, and that aint
physics. We will want the mathematical convenience at a later date.
For our purposes, we will need real solutions, and it would be nice to know
that real solutions exist; this would mean that is real for all (if is real, all
3

of the elements of
x0 may be chosen real). All we need is the fact that both K
and M are real and symmetric. Then, if K
x0 = 2 M
x0 ,
T

xT0 K = (K
x0 ) = 2 (M
x0 )

and,


T  2   T 
 T 
T
=

x0 M .

x0 K = (K
x0 ) = 2 M
x

Thus, we have

 2



x0 ) =
xT
x0 = 2
xT
x0 ,

xT
0 (K

0 M

0 M

 T 


  T 
  T


x0 K
x0 = 2

x0 M
x0 = 2

x0 M
x0 ,

 
so 2 = 2 , and 2 is real. Those familiar with the terminology of linear algebra
will note that we have introduced the Hermitian conjugates of the
x0 , and that
because both K and M are real and symmetric, they are Hermitian, as is K
.
Showing that 2 is non-negative, so that is real, is best done by appeal to
the phyiscs. Specically,
x =
0 must be a stable conguration point (or neutrally
stable, if = 0 for some ). A way to see this is to consider the potential energy
V (x1 , . . . , xn ) of the system due to the forces represented by K; each component of
K is then
2 V 
Kjk =
.

xj xk
x=
0
Apart from the minus sign, K is a Hessian matrix, a term you might have encountered in 18.02 or the equivalent. Well just use the calculus result, that V is
a minimum at
x =
0 if the eigenvalues of K are negative. We could then use the
result from linear algebra that if the eigenvalues of K are negative, the eigenvalues of K
(which are not the eigenvalues of K) are negative. Or, we can dene

V
(y1 , . . . , yn ) = V ( m1 y1 , . . . , mn yn ), using yi = xm1 1 , as in (4), and then observe that if V is a minimum at
x =
0, V
is a minimum at
y =
0.
So, now that we know that they exist, lets look for real solutions (e.g., sines
and cosines), specically

x(t) =

(aj
xj0 cos j t + bj
xj0 sin j t) ,

x(0) =
x0 ,

x (0) =
v0 ,

(5)

j=1

where the aj , bj are constants, and
xj0 and j are as found previously. Note that

x0 , without a j or subscript, is the initial value
x(0), and is not, in general, equal
to
xj0 for any j. In terms of the vectors
xj0 ,

x0 =

aj
xj0 ,

v0 =

j=1

n

j=1

j bj
xj0 .

To nd the aj and bj , use the linearity of M and the normalcy condition;

M
x0 =

xTk0 M
x0 =

aj M
xj0 ,

j=1

aj
xTk0 M
xj0 .

j=1

The sum is of n terms, all but one of which vanish. That term will be for j = k, so

xTk0 M
x0 = ak
xTk0 M
xk0
and so
ak =

xTk0 M
x0
.

xTk0 M
xk0

Similarly,

/(ii), where A

= Axi + Ay j + Az k.
This is analogous to Ax = i A
bk =

1
xTk0 M
v0
,
k
xTk0 M
xk0

and the solution to (5) is

x(t) =

n

j=1

xj0
T

xj0 M
xj0



1 T
T

xj0 M
x0 cos j t +
xj0 M
v0 sin j t .
j

(The situation where j = 0 for some j is not hard to incorporate, and is often
made part of an assignment.) This may look like a mess, but consider; once we nd
the
xk0 and k , and compute M
x0 and M
v0 , thats it!.
Well, whats the point? Consider now a driven system,
= K
x + F

(t),
M
x
where F
is a column vector representing the external forces. Then, look for

x(t) =

gk (t)
xk0 ,

=

x

k=1

gk
xk0 ,

k=1

where the
xk0 are the vectors found previously. Then,

=
M
x

gk (t) M
xk0 =

k=1

gk K
xk0 + F
(t).

k=1

But, remember that K
xk0 = k2 M
xk0 , so

n




gk + k2 gk M
xk0 = F
(t),

k=1

and, as before,

n





.
gk + k2 gk
xTj0 M
xk0 =
xTj0 F
=
xj0 F

k=1

But the sum vanishes except for the j = k term; then,

gk + k2 gk =

xTk0

1

(t).

xk0 F
M
xk0

The intial conditions for gk are similarly found to be

gk (0) =

xTk0 M
x0
,

xTk0 M
xk0

g k (0) =

xTk0 M
v0
.

xTk0 M
xk0

This is a second order inhomogeneous equation. The solution to the homogeneous part is well known; the complete solution depends, of course, on F
, and
may be found by variation of parameters, undetermined coecients, annihilators,
Greens functions, Laplace transforms or Tarot cards.
The main point here is; if any component of F
is sinusoidal with any frequency
k (unless
xk0 F
vanishes), there will be one normal mode of the system that will
be driven at resonance. This is when interesting things happen.