Lecture notes for 12.006J/18.353J/2.

050J, Nonlinear Dynamics: Chaos

D. H. Rothman, MIT
September 15, 2022

Contents
1 Flows in one dimension 1
1.1 Fixed points and stability . . . . . . . . . . . . . . . . . . . . 2
1.1.1 General statements . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Example: logistic growth . . . . . . . . . . . . . . . . . 3
1.2 Linear stability analysis . . . . . . . . . . . . . . . . . . . . . 5

2 Bifurcations in one dimension 7

2.1 Saddle-node bifurcation . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Normal form . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Example: oxygenation of the atmosphere . . . . . . . . 11
2.2 Transcritical bifurcation . . . . . . . . . . . . . . . . . . . . . 12
2.3 Pitchfork bifurcation . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.1 Supercritical pitchfork bifurcation . . . . . . . . . . . . 14
2.3.2 Subcritical pitchfork bifurcation . . . . . . . . . . . . . 15

1 Flows in one dimension

Reference: Strogatz, Chapter 2 [1].

The simplest dynamical systems concern the evolution of only one variable.
Here we consider flows in which the the dynamics

ẋ = f (x)

describe the continuous evolution of x(t).

1
1.1 Fixed points and stability

1.1.1 General statements

Suppose, for example, that

f (x) = x2 − 1.

Since a graph of f (x) is equivalent to a graph of ẋ as a function of x, we have

O X

The sign of f (x) determines the directions of the flow of x; we signify this by
arrows.

f (x) intersects the x-axis when f (x) = 0. These are called fixed points (or
equilibrium solutions) of the system. We shall denote these by x⇤ . Here we
have
x⇤ = ±1.

A fixed point is stable if flows are attracted to it.

A fixed point is unstable if flows are repelled from it.

By inspection we see that

f 0 (x⇤ ) < 0, x⇤ stable

f 0 (x⇤ ) > 0, x⇤ unstable

2
1.1.2 Example: logistic growth

Consider the growth of some population, say, rabbits, bacteria, or people.

Let
N = population size
r = growth rate.

The simplest possible growth model has N growing at rate r > 0:

dN
= rN, r > 0.
dt
The solution predicts exponential growth:
N (t) = N (0)ert .
Such a model works works well when the resource that a population requires
for growth is sufficiently abundant.

In cases of such unlimited growth,

1 dN
r= = ‘per capita’ or ‘specific’ growth rate
N dt
is constant, independent of N .

Eventually, however, resources become depleted and/or the population be-

comes overcrowded.

Thus as N increases, we expect that r will decrease.

Moreover it can even become negative, meaning that the death rate exceeds
the birth rate.

Graphically, we expect that r(N ) behaves like

r

i K N

3
The point at which r(N ) = 0 is a special population size which neither grows
nor decays. It corresponds to N = K, where
K = carrying capacity.
In most cases we can’t really know the shape of r(N ), but the notion of a
carrying capacity remains reasonable.

So the simplest assumption would be to assume that r(N ) decreases linearly:

n

I N
Then ◆ ✓
N
r(N ) = r 1 −
K
and our growth model now reads
✓ ◆
dN N
= rN 1 − , (1)
dt K
known as the logistic equation.⇤

The logistic equation can be solved exactly but in this course we are generally
interested in qualitative analyses that lead to physical (and mathematical)
insight.

Here, qualitative analysis begins by plotting dN/dt vs. N , i.e., we plot the
RHS of the logistic equation:
N

0
in to
We consider only N ≥ 0 since populations cannot be negative.
⇤
The Belgian mathematician Verhulst published and named it in 1838, but he gave no reason for the
name.

4
The fixed points occur where Ṅ = 0. We find

N1⇤ = 0, N2⇤ = K.

The slope of the curve or the directions of flow tell us that N1⇤ = 0 is unstable
N2⇤ = K is stable.

Indeed, we see that as long as we initiate growth with N (0) > 0 , the popu-
lation always evolves to the carrying capacity.

How it evolves depends on the initial condition:

• Small populations N < K grow in an accelerating fashion, and then

later slowly approach K. This is sometimes called S-shaped or sigmoid-
shaped growth.
• Large populations N > K or populations K/2 < N < K continously
decelerate as they approach K.
N

É I e

1.2 Linear stability analysis

We now address stability quantitatively. Recall that we are studying one-

dimensional flows
ẋ = f (x).
Let
⌘ = x − x⇤
be a small perturbation away from a fixed point x⇤ .

5
Now note that
d
⌘˙ = (x − x⇤ )
dt
= ẋ
= f (x)
= f (x⇤ + ⌘).
Expand f (x⇤ + ⌘):
f (x⇤ + ⌘) = f (x⇤ ) + ⌘f 0 (x⇤ ) + O(⌘ 2 ).
Since x⇤ is a fixed point, f (x⇤ ) = 0. Then
f (x⇤ + ⌘) = ⌘f 0 (x⇤ ) + O(⌘ 2 ).
Assuming that f 0 (x⇤ ) 6= 0, the terms of O(⌘ 2 ) are negligible and we have
⌘˙ = ⌘f 0 (x⇤ )
This simple linear equation shows us what we determined earlier from the
slope of f (x) in our graphs:

• f 0 (x⇤ ) < 0 ) x decays exponentially ) x⇤ is stable.

• f 0 (x⇤ ) > 0 ) x grows exponentially ) x⇤ is unstable.

But we now learn something else: the characteristic timescale of the expo-
nential growth or decay is 1/|f 0 (x⇤ )|.

What happens when f 0 (x⇤ ) = 0? The higher-order terms then matter; we

look at this later.

Example: Let’s return to logistic growth. Here

✓ ◆
N
f (N ) = rN 1 −
K
with fixed points N ⇤ = 0 and N ⇤ = K. Since
✓ ◆
0 2N
f (N ) = r 1 −
K
6
we find
f 0 (0) = r and f 0 (K) = −r.
As we found graphically, N ⇤ = 0 is unstable and N ⇤ = K is stable.

Moreover, the timescale of growth or decay is 1/r.

2 Bifurcations in one dimension

Reference: Strogatz, Chapter 3 [1].

Thus far, in our discussion of ẋ = f (x), nothing terribly interesting happens:

either x ! x⇤ or x ! ±1.

Now we consider instances in which f (x) depends on a control parameter in

a way that fixed points can be created or destroyed, or their stability can
change.

Such qualitative changes are called bifurcations.

Examples: stick-slip motion as function of applied stress; the transition from

conduction to convection as a fluid is heated from below.

Although the representation of bifurcations in one dimension may seem ar-

tificial or too simple, it turns out to be highly relevant in more general,
multi-dimensional settings.

2.1 Saddle-node bifurcation

2.1.1 Normal form

Here we illustrate how fixed points can be created or destroyed. We write

ẋ = r + x2
where r, the control parameter, can be positive, negative, or zero. The three
cases are qualitatively di↵erent:

7
 x

Hoax had x x

Nco no r 0
The bifurcation occurs at r = 0.

Fixed points exist only if r < 0. We then have

p
x⇤ = ± −r, r<0

We can infer their stability graphically.

Now let’s plot a bifurcation diagram in the plane of r and x.

Strogatz [1], Fig. 3.1.4 See image credit on Page 19.

This type of bifurcation is called a saddle-node bifurcation (or a fold bifurca-
tion). It occurs quite commonly. Let’s consider a few other ways in which it
can appear.

The first is trivial: write

ẋ = r − x2
This is simply the reverse case: instead of two fixed points disappearing as
r is increased past zero, the parabolas are concave down, two fixed points
appear as r is increased past zero, and their stability is exchanged relative to
the case above:

8
 X X X

xno
no mo
Max r 0

The second amounts to a tilt. Suppose that

ẋ = r − x − e−x

Consider separate plots of r − x and e−x :

Strogatz [1], Fig. 3.1.6 See image credit on Page 19.

Fixed points occur at the intersections of the line and the curve, and their
stability corresponds to the case we just considered.

The bifurcation occurs where the graphs of r−x and e−x intersect tangentially.
This requires that
r − x = e−x
and that their slopes are equal:
d d −x
(r − x) = e .
dx dx
The second condition gives

−1 = −e−x ) x=0

and the first gives r = 1; therefore the bifurcation occurs at x = 0 and rc = 1.

We now show why the quadratic form is general (and is therefore called a

9
normal form). Near the bifurcation, we write
e−x
ẋ = r − x − ✓ ◆
x2
= r−x− 1−x+ + ...
2
x2
= r−1− + ...
2
This has the same form as ẋ = r − x2 , and could agree exactly by rescaling
r and x.

Essentially we are saying that ẋ appears parabolic near r = rc , where rc is

the critical value of r where the fixed points collide (or first appear).

Strogatz [1], Fig. 3.1.7 See image credit on Page 19.

This point can be made precise by Taylor expansion of f (x, r) near x = x⇤
for r = rc :
ẋ = f (x, r)
� � �
@f � @f � 1 @f �
= f (x⇤ , rc ) + (x − x⇤ ) � + (r − r ) � + (x − x⇤ 2
) � + ...
@x �x⇤ ,rc @r �x⇤ ,rc 2 @x2 �x⇤ ,rc
c

where we have neglected terms of O(x − x⇤ )3 and O(r − rc )2 .

The first term vanishes because x⇤ is a fixed point.

The second term vanishes because the assumption of a saddle-node bifurca-

tion requires that f (x) be tangent to the x-axis at r = rc . Then
ẋ = a(r − rc ) + b(x − x⇤ )2
where a and b can be read o↵ the Taylor expansion. Appropriate rescalings
then bring us back to the normal form ẋ = r + x2 .

10
2.1.2 Example: oxygenation of the atmosphere

Many problems of dynamics may be expressed as

ẋ = inputs − outputs

As an example, let’s ask how the atmosphere became oxygenated.

Today’s level of O2 is stable. Conceptually that means there are sources and
sinks that balance in a stable way [2]:
sink
rate
Edt
X

son e
La

There was essentially no O2 in the atmosphere until about 2.5 billion years
ago, when oxygen-producing photosynthesis developed.

But the first aerobic organisms presumably consumed it voraciously. And

we know from various observations that O2 levels remains low for at least
hundreds of millions of years.

How, then, could it have increased if the stability argument just given is
correct?

One idea is that the source function had a non-monotonic dependence on O2

levels, and the sink function became weaker with time. A first we would have
a situation like this: source − sink
d[O ] / dt ˜
rate °

0
2

source
sink
0
0
˜ ° [O2] ˜
ox

Then, as the sink function weakened, two new fixed points would appear (via
a first saddle-node bifurcation):

11
 source − sink

d[O ] / dt ˜
rate °
0

2
source
sink
0
0
˜ ° [O2] ˜
ox

Further weakening of the sink then causes the low-O2 fixed point to collide
with the unstable fixed point via a second saddle-node bifurcation, leaving
only the high-O2 state:
source − sink

d[O2] / dt ˜
rate °

source
sink
0
0
˜ox ° [O2] ˜

For further details and evidence that such a picture may be correct, see
Ref. [3].

2.2 Transcritical bifurcation

In some problems a fixed point must always exist. For example, in the logistic
equation, the zero-population fixed point must always be present.

However the stability of the permanent fixed point may change as a parameter
is varied.

The normal form of the transcritical bifurcation accounts for this possibility.
Its normal form reads
ẋ = rx − x2
This looks like the logistic equation, but now x and r are no longer constrained
to be positive.

Graphically, we have

12
 Strogatz [1], Fig. 3.2.1 See image credit on Page 19.
⇤
We see that the fixed point x = 0 always occurs, but as r increases past zero
the unstable fixed point x⇤ = r for r < 0 becomes stable when r > 0.

Here’s the bifurcation diagram in the plane of r and x:

Strogatz [1], Fig. 3.2.2 See image credit on Page 19.

2.3 Pitchfork bifurcation

Many physical problems exhibit symmetry, such that our one dimensional
dynamics are invariant under x ! −x. Examples:

• Buckling beam. Put a weight on a beam confined between two rigid

walls. If the weight is sufficiently heavy, the beam will buckle to either
the left or right.
• Thermal convection. Heat a fluid from below. Eventually, hot fluid rises
and cold fluid sinks, forming convective “rolls.” In two dimensions, the
motion may be either clockwise or counter-clockwise.

Such symmetry leads to a pitchfork bifurcation. There are two types: super-
critical and subcritical.

13
2.3.1 Supercritical pitchfork bifurcation

The normal form is

ẋ = rx − x3
Note that the equation is invariant for x ! −x.

Here are the flows:

Strogatz [1], Fig. 3.4.1 See image credit on Page 19.

⇤
When r < 0, there is only one fixed point, x = 0.

When r = 0, the origin is still stable, but only weakly because the linear term
has vanished. Thus there is no longer an exponential decay towards x⇤ = 0.
This is called critical slowing down.

When r > 0, the origin becomes unstable and two new fixed points occur, at
p
x⇤ = ± r. Both are stable.

The bifurcation diagram looks like a pitchfork:

Strogatz [1], Fig. 3.4.2 See image credit on Page 19.

Note that there is now bistablity: depending on the initial condition, we either
end up at the positive or negative fixed point. Thus the initial condition
“breaks” the symmetry.

Another way of viewing the problem is to derive the dynamics ẋ from a

14
potential V (x). We define the potential by
dV
ẋ = −
dx
so that x tends to go in the direction of smaller V (i.e., downhill), without
inertia, at a rate proportional to the slope of the potential.

Integrating the right-hand side, we obtain

1 1
V (x) = − rx2 + x4 ,
2 4
where the integration constant is set to zero. Here are plots for di↵erent r:

Strogatz [1], Fig. 3.4.5 See image credit on Page 19.

This sort of potential occurs in studies of phase transitions where either of two
states are possible, such as spontaneous magnetization or phase separation.
In this case the parameter r is typically the temperature, and the critical
temperature corresponds to r = 0.

We can also now better appreciate the notion of critical slowing down: at
r = 0 the potential is nearly flat near the origin, so fluctuations relax very
slowly.

2.3.2 Subcritical pitchfork bifurcation

In the supercritical case, the cubic term is stabilizing: its sign is opposite
that of the instability caused by the linear term.

To see what this means, consider x very small; then the nonlinear term is
negligible so that, for r > 0,
ẋ ⇠ rx ) x ⇠ ert .

15
But eventually the cubic term becomes significant. If we write the supercrit-
ical normal form as ✓ ◆
x2
ẋ = rx 1 − ,
r
p
we see that “saturation” of the instability occurs when x = ± r, i.e., at a
fixed point.

However, nothing prevents the lowest-order nonlinear term from also being
destabilizing. Thus we could have
ẋ = rx + x3 .
The flows are now flipped compared to previously:
x x x

from
to
x

MCO no no
⇤
p
The non-zero fixed points x = ± −r are now unstable, and they exist only
for r < 0. (Thus we see some reason for the name “subcritical.”)

And the origin is stable for r < 0 but unstable otherwise:

Strogatz [1], Fig. 3.4.6 See image credit on Page 19.

Of course, now the cubic term drives trajectories out to ±1, so we still need
a stabilizing term.

The lowest-order nonlinear term that still maintains symmetry is then x5 ,

leading to the normal form
ẋ = rx + x3 − x5 .

16
Although things now become somewhat complicated algebraically, the e↵ect
of the damping term must create a new set of stable fixed points above the
dashed lines.

Here are some plots for di↵erent values of r:

r=-0.25 r=-0.2
 
x x
0.2

0.10
0.1
0.05

-1.0 -0.5 -1.0 -0.5

x x
0.5 1.0 0.5 1.0

-0.05
-0.1
-0.10

-0.2

r=0 r=0.1
 
x x
0.2

0.2
0.1
0.1

-1.0 -0.5 -1.0 -0.5

x x
0.5 1.0 0.5 1.0

-0.1
-0.1
-0.2

-0.2

The bifurcation diagram then looks like this:

Strogatz [1], Fig. 3.4.7 See image credit on Page 19.

Here rs = −0.25. Otherwise there is plenty to say:

• When rs < r < 0, both the origin and the new large-amplitude fixed
points are stable. The initial condition determines the final outcome.
Thus the origin is stable to small perturbations, but not to large pertur-
bations.
• Jumps. When r is increased to r = 0, the system jumps to a large-
amplitude fixed point.

17
 • Hysteresis. If r is then decreased, the system stays on the large ampli-
tude branch until it reaches rs , and then it jumps back to the origin.
The dynamics are not reversible.
x

0
rs 0 r

Strogatz [1], Fig. 3.4.8 See image credit on Page 19.

• As r is increased past rs from below, both an unstable and a stable fixed

point appear, below and above the origin. Our earlier discussion of such
spontaneous appearance of stable and unstable fixed points implies that
this occurs by a saddle-node bifurcation, as one can see in the plot of ẋ
vs. x for r = −0.25.

A major di↵erence between the supercritical and subcritical pitchfork bifur-

p
cation is that the former leads to continuous changes x⇤ ⇠ r as r is increased
past zero, whereas the latter results in a finite jump.

Because the jump can be quite large, subcritical bifurcations are sometimes
called hard or dangerous, while supercritical bifurcations are called soft or
safe.

And in the analogy with phase transitions, supercritical bifurcations are re-
lated to continuous or second-order phase transitions (e.g., the ferromagnetic
transition), while subcritical bifurcations are discontinuous or first-order (e.g.,
the freezing of water into ice).

References
1. Strogatz, S. Nonlinear Dynamics and Chaos: With Applications to Physics,
Biology, Chemistry, and Engineering (CRC Press, 2018).

18
2. Holland, H. D. The Chemistry of the Atmosphere and Oceans (John Wiley
& Sons, New York, 1978).
3. Shang, H., Rothman, D. H. & Fournier, G. P. Oxidative metabolisms
catalyzed Earth’s oxygenation. Nature communications 13, 1–9 (2022).

Image Credit
Images on Pages 8–10 and 13–18 from Strogatz, S. Nonlinear Dynamics and Chaos: With
Applications to Physics, Biology, Chemistry, and Engineering (CRC Press, 2018) © Informa
UK Limited. All rights reserved. This content is excluded from our Creative Commons
license. For more information, see https://ocw.mit.edu/help/faq-fair-use.

19
MIT OpenCourseWare
https://ocw.mit.edu

12.006J/18.353J/2.050J Nonlinear Dynamics: Chaos

Fall 2022

For information about citing these materials or our Terms of Use, visit: https://ocw.mit.edu/terms.