Scribd
Upload
73 views

The Variational Equation Notes For CSCI4446/5446

This document discusses the variational equation, which describes how small perturbations to a dynamical system evolve over time along a trajectory. The variational equation models how volumes or perturbations in state space change by looking at the differential slopes of the system using a Jacobian matrix. Solving the variational equation simultaneously with the system equation allows modeling how initial perturbations grow or shrink over time based on the landscape around the trajectory.

Uploaded by

Pedro Mendes
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
73 views

The Variational Equation Notes For CSCI4446/5446

This document discusses the variational equation, which describes how small perturbations to a dynamical system evolve over time along a trajectory. The variational equation models how volumes or perturbations in state space change by looking at the differential slopes of the system using a Jacobian matrix. Solving the variational equation simultaneously with the system equation allows modeling how initial perturbations grow or shrink over time based on the landscape around the trajectory.

Uploaded by

Pedro Mendes
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

The Variational Equation

Notes for CSCI4446/5446


Liz Bradley
Department of Computer Science
University of Colorado
Boulder, Colorado, USA 80309-0430
c
1999
[email protected]

The variational equation is the mathematics that goes with pictures like this:

This equation describes how pertubations (or state-space volumes) evolve along a trajectory. This kind of information is essential to any kind of in-depth understanding of the
landscapes of dynamical systems: stability, Liouvilles theorem and dissipation, un/stable
manifolds, s, etc.
Given some state-space point ~x0 and some dynamical system ~x = F~ (~x), its easy to
generate the state-space trajectory (~x0 ) that emanates from that point; just use any old
ODE solver like RK4. What wed like to know, now, is how that would change if someone
wiggled ~x. Consider the Rossler system

(y + z)
fx (x, y, z)
x

~
F (~x) = y = fy (x, y, z) = x + 0.398y
2 + z(x 4)
fz (x, y, z)
z
with ~x0 = [0, 1, 2]T . The state-space trajectory (~x0 ) from that point is shown below:
x

0
( x

(1)

Imagine, now, that we add a small perturbation to ~x0 , in the x-direction. That perturbation will evolve as the system moves along the trajectory:

The same is also true of perturbations in the other two directions:

The change in length and orientation of each arrow, as well as the resultant to which
all three add up in the end (which is equal to the volume into which the original box has
grown), depends on the differential slope of the ODE landscape around the trajectory:
If a particular perturbation points out over a sharp dropoff, that arrow will lengthen
If it points over a gentle rise, that arrow will shrink
If it points out diagonally from a saddle point (not in the direction of either the stable
or the unstable direction), it will be pulled sideways: it will grow along the unstable
direction and shrink along the stable one
If it points out over a region of constant slope, it will not change length.
The formal way to describe these differential slopes to which the variations react is with
partial derivatives:
fx
y

for instance, is how much the x-slope (fx ) changes if you wiggle y. In n dimensions, there
are n variables and n slopes, so you need n2 of these different partials to describe everything.
One way to write these down is in an n n matrix called the Jacobian:

Dx F~ =

fx
x
fy
x
fz
x

fx
y
fy
y
fz
y

fx
z
fy
z
fz
z

For the Rossler system, the Jacobian is

0 1
1

1 0.398
z
0
x4
The last thing we need to set up the variational equation is a way to describe the
variations. The notation is a little complicated; xy is the component of the x variation
that came from the previous y variation, and so on. Again, there are nine of these:

xx yx zx

[] = xy yy zy
xz yz zz
The column sums of this matrix are the lengths of the x, y, and z coordinates (respectively)
of the evolved variation. The rows are the coordinates of the vectors into which the original
x, y, and z components of the variation have evolved.
= Dx F~ [], or
The variational equation is []

xx yx zx


xy yy zy =

xz yz zz

fx
x
fy
x
fz
x

fx
y
fy
y
fz
y

fx
z
fy
z
fz
z

yx zx
xx
xy yy zy

xz yz zz

For the Rossler system, this looks like:

xx yx zx
xx yx zx
0 1
1

0
xy yy zy
xy yy zy = 1 0.398
xz yz zz
z
0
x4
xz yz zz

To figure out what happens to the variations, you have to solve the variational equation
and the system equation simultaneously (because the variations react to the differential
slopes at each point along the trajectory). To do this, you work with a new n + n2 length
augmented state vector:
(x y z xx xy xz yx yy yz zx zy zz )
and its derivative:

(x y z xx xy xz yx yy yz zx zy zz )
3

=
...and you stack the system equation ~x = F~ ~x on top of the variational equation []
Dx F~ []. The equations are:
x = (y + z)
y = x + 0.398y
z = 2 + z(x 4)
xx = xy xz
xy = xx + 0.398xy
xz = zxx + (x 4)xz
yx = yy yz
yy = yx + 0.398yy
yz = zyx + (x 4)yz
zx = zy zz
zy = zx + 0.398zy
zz = zzx + (x 4)zz
If you wrote your RK4 for arbitrary length vectors and derivatives, you should have no
problem here; its just a 12D system instead of Rosslers three.
For example, if you want to see how unit variations grow near the point [0, 1, 2]T , youd
fire up your ODE solver starting at
[0 1 2 1 0 0 0 1 0 0 0 1]
and let it go. Say, for the purposes of this example, that after 100 steps at 0.05 seconds,
the new state of this augmented system were:
[13.087 23.455 24.204 0.79 0.95 0.89 0.61 1.23 0.79 0.14 0.48 0.66
To figure out the actual variations from this, you need to look at the appropriate rows or
columns. The total x variation at t = 5sec is xx +xy +xz = 0.79+0.95+0.89; the original
x perturbation (at t = 0) has grown into the vector xx x +yx y+zx z = 0.79
x +0.61
y 0.14
z:
^
^
^
0.79X+0.61Y-0.14Z
^ ^ ^
X+Y+Z

You might also like