18.03 Class 37, May 7 Inhomogeneous linear equations.

The first order companion system of a second order differential equation.

If we have a second order linear ODE
x" + bx' + kx = f(t)
we can form a first order system of equations by the usual device: set
y = x'
so
y' = x" = - kx - bx' + f(t)
and we have the first order system, the "companion system,"
x' = y
y' = - kx - by + f(t)
It is autonomous when k , b , and f(t) are all constant.
This association leads to a dictionary relating concepts relevant to the
second order equation to concepts relevant to the corresponding first order system. Here it is, all at once:

DICTIONARY: Second order equation x x' [x ; x' ] x" + bx' + kx = 0 First order system x y u u' = Au , A = [ 0 1 ; -k -b ]

x" + bx' + kx = f(t) s and p(s) roots of char. poly. normalized soln. pair b , k >= 0 undamped underdamped critically damped overdamped

u' = Au + [ 0 ; f(t) ] lambda and p_A(lambda) eigenvalues top entries of $e^{At} NW part of (Tr,Det) plane center spiral defective node node

For example, if
x" + 2x' + 2x = 0 (*)

has char. poly. s^2 + 2s ^ 2 with roots - 1 +- i
and hence the normal modes are
e^{(-1+-i)t}
and the basic real solutions are
e^{-t} cos(t) and e^{-t} sin(t)

The equation is underdamped.
The companion matrix is
A = [ 0 1 ; -2 -2 ]
Its characteristic polynomial is lambda^2 + 2 lambda + 2 and the
eigenvalues are -1 +- i . With lambda1 = -1 + i , alpha1 satisfies
[ 1-i , 1 ; -2 , -1-i ]

and so alpha1 = [ 1 ; -1+i ] or any complex multiple. (For example, [ 1+i ; -2 ] is another eigenvalue for the same eigenvector.) The corresponding normal mode is e^{(-1+i)t} [ 1 ; -1+i ]
= e^{-t} ( cos(t) + i sin(t)) [ 1 ; -1+i ]
with real part e^{-t} [ cos(t) ; - cos(t) - sin(t) ]

and imaginary part e^{-t} [ sin(t) ; cos(t) - sin(t) ]
The top entries of these solutions are exactly the two solutions of (*)
and the bottom entries are their derivatives.
The trajectories are stable spirals.
The confusing story about normalized pairs of solutions is just the story
of the matrix exponential in this case. We have for fundamental matrix
Phi(t) = e^{-t} [ cos(t) , sin(t) ; - cos(t) - sin(t) , cos(t) - sin(t) ]
This isn't yet the exponential, since
Phi(0) = [ 1 , 0 ; -1 , 1 ]
Phi(0)^{-1} = [ 1 , 0 ; 1 , 1 ]
so
e^{At} = Phi(t) Phi(0)^{-1}
= e^{-t} [ cos(t)-sin(t) , sin(t) ; -2sin(t) , cos(t)-sin(t) ]
e^{At} is a fundamental matrix, and is characterized among all fundamental
matrices by being "normalized at t = 0" in the sense that its value at t = 0 is the identity matrix. Correspondingly, its columns are solutions x1 , x2 , of (*) , such that x1(0) = 1 x1'(0) = 0 x2(0) = 0 x2'(0) = 1

That is, the top entries of e^{At} form a normalized pair of solutions.

Now, as it turns out, this very same matrix provides a model for a revised
version of the romance of Romeo and Juliet;
R' = J
J' = -2R - 2J
so Romeo pays no attention to his own feelings, but if he sees she loves him
he comes to love her more. Juliet is suspicious of both his affection and hers.
Now, this young couple does not exist in isolation. The Montagues and Capulets
do what they can to make each of them like the other less. We could model this
by
R' = J - 1
J' = -2R - 2J - 2
This is now an INHOMOGENEOUS linear equation. We could write it as
u' = Au + q(t)
where here q(t) = [ -1 ; -2 ] is a constant vector.
Its general solution has the form
u = up + uh
where up = [ Rp ; Jp ] is any "particular" solution, and uh is the
general solution of the homogenous equation, which we just found.
What can we hope for by way of up? Since the "input signal" [ -1 ; -2 ] ,
is constant, we might hope for up constant. In this case, up' = 0 , so
0 = A up + q or up = - A^{-1} q
For us, A = [ 0 1 ; -2 -2 ] so
A^{-1} = (1/2) [ -2 -1 ; 2 0 ] = [ -1 -1/2 ; 1 0 ] up = - [ -1 -1/2 ; 1 0 ] [ -1 ; -2 ] = [ 3/2 ; -1 ] .
This is the "equilibrium solution." In it, R = 3/2 and J = -1 .
and

The forces just balance. Romeo's love (with value 3/2) is never requited, and Juliet's dislike (with value -1 ) is never relieved. In fact, if the Montagues and Capulets wanted to produce any given outcome, they can arrange it. To produce an equilibrium at [ a ; b ] , they just have to take q = - Au You can see though that (1) they have to understand their kids, and (2) they have to cooperate with each other!

Introduction to general nonlinear autonomous systems. Recall that an ODE is autonomous if x' depends only on x and not on t: x' = f(x) For example, I know an island in the St Lawrence River in upstate New York where there are a lot of deer. When there aren't many deer, they multiply with growth rate k ; x' = kx . Soon, though, they push up against the limitations of the island; the growth rate is a function of the population, and we might take it to be k(1-(x/a)) where a is the maximal sustainable population of deer on the island. So the equation is x' = k(1-(x/a))x , the "logistic equation." On this particular island, k = 3 and a = 3 , so x' = (3-x)x . There are "critical points" at x = 0 and x = 3 . When 0 < x < 3 , x' > 0. When x > 3 , x' < 0 , and, unrealistically, when x < 0 , x' < 0 too. I drew some solutions, and then recalled the phase line:

------<-----*---->------*------<-------­

One day, a wolf swims across from the neighboring island, pulls himself up the steep rocky coast, shakes the water off his fur, and sniffs the air. Two wolves, actually.

Wolves eat deer, and this has a depressing effect on the growth rate of deer.
Let's model it by
x' = (3-x-y)x
where y measures the population of wolves.
Now, wolves in isolation follow a logistic equation too, say
y' = (1-y)y (no deer)

But the presence of deer increases their growth rate, say
y' = (1-y+x)y
We have a nonlinear autonomous system
x' = (3-x-y)x
y' = (1-y+x)y
The general case would be
x' = f(x,y)
y' = g(x,y)
We will be interested in the equilibrium solutions, and how other solutions
converge to them.