Answers to P-Set # 02, (18.353/12.006/2.

050)j
MIT (Fall 2021)
Rodolfo R. Rosales (MIT, Math. Dept., room 2-337, Cambridge, MA 02139)
September 24, 2021

Contents
1 Find and classify bifurcations problem #01 2
1.1 Statement: Find and classify bifurcations problem #01 . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Answer: Find and classify bifurcations problem #01 . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Find and classify bifurcations problem #02 2

2.1 Statement: Find and classify bifurcations problem #02 . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Answer: Find and classify bifurcations problem #02 . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 Toy model for column buckling 6

3.1 Statement: Toy model for column buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.2 Answer: Answer: Toy model for column buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
Force balances and a bit of physical intuition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

4 Problem 16.09.17 - (exponential to algebraic decay transition) 10

4.1 Problem 16.09.17 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.2 Problem 16.09.17 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

5 Problem 02.05.06 - Strogatz (The leaky bucket) 11

5.1 Problem 02.05.06 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.2 Problem 02.05.06 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

6 Problem 03.02.06 - Strogatz (Eliminate the cubic term) 12

6.1 Problem 03.02.06 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
6.2 Problem 03.02.06 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

7 Problem 03.04.11 - Strogatz (An interesting bifurcation diagram) 14

7.1 Problem 03.04.11 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
7.2 Problem 03.04.11 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

8 Problem 16.09.20 (growing yeast) 15

8.1 Problem 16.09.20 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
8.2 Problem 16.09.20 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

List of Figures
1.1 Bifurcation diagram with a saddle-node . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.1 Bifurcation diagram with two saddle-nodes and a soft pitchfork . . . . . . . . . . . . . . . . . . . . . 4
2.2 Bifurcation diagram with a hard pitchfork . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Bifurcation diagram with a non-canonical hard pitchfork . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Bifurcation diagram with a hard pitchfork . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Bifurcation diagram with a non-canonical hard pitchfork . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.1 Toy model for column buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Bifurcation for the toy model for column buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.3 Forces in the toy model for column buckling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.1 Problem 16.09.17. Algebraic versus exponential decay at a critical slowing down . . . . . . . . . . . . 11

1
18.353 MIT, (Rosales) Find and classify bifurcations problem #01. 2

7.1 Problem 03.04.11. Bifurcation diagram for ẋ = rx − sin(x) . . . . . . . . . . . . . . . . . . . . . . . . 14

8.1 Problem 16.09.20. Bifurcation diagram for ẋ = x (1 − x) − h . . . . . . . . . . . . . . . . . . . . . . . 16

1 Find and classify bifurcations problem #01

1.1 Statement: Find and classify bifurcations problem #01
For equation (1.1) below, find the values of r at which a bifurcation occurs, and classify them as saddle-node,
transcritical, supercritical pitchfork, or subcritical pitchfork. Finally, sketch the bifurcation diagram of fixed points
x∗ versus r.
dx x2
=r− . (1.1)
dt 1 + x2

1.2 Answer: Find and classify bifurcations problem #01

The critical points are given by r
r
x=± for 0 ≤ r < 1, (1.2)
1−r
x2 x2
r
r
with no others. Furthermore ẋ > 0 for r > 2
, and ẋ < 0 for r < 2
. It follows that the branch x =
r 1 + x 1 + x 1 − r
r
is stable, and the branch x = − is unstable, with a saddle-node bifurcation at (r, x) = (0, 0). The bifurcation
1−r
diagram is shown in figure 1.1
2 2
Bifurcation diagram for dx/dt = r − x /(1+x ).
3

In each region (yellow or cyan), the black arrows

1 indicate the direction of the flow for the equation
x2
x0 ẋ = r − .
1 + x2
Stable branches of critical points are plotted in solid
−1 blue, and unstable branches in dashed red. The
black dot indicates the bifurcation point (saddle-
−2 node).

−3
−0.5 0 0.5 1
r
Figure 1.1: Bifurcation diagram for equation (1.1).

2 Find and classify bifurcations problem #02

2.1 Statement: Find and classify bifurcations problem #02
The only extra question that you need to answer is: does the PCS apply across r = 0? The others are optional.
Note that the problem has three parts, and that in each you have to answer the same set of questions.
18.353 MIT, (Rosales) Find and classify bifurcations problem #02. 3

Part 1 of 3. For equation (2.1) below, find the values of r at which a bifurcation occurs, and classify them as
saddle-node, transcritical, supercritical pitchfork, or subcritical pitchfork. Finally, sketch the bifurcation diagram of
fixed points x∗ versus r.
dx x3
= rx− . (2.1)
dt 1 + 2 x2 + x4
Extra questions: Something “strange” happens for r = 0 in the bifurcation diagram. Is some bifurcation taking
place there? If so, which type? Does the PCS apply across r = 0?

Hint 2.1 Look at the equation satisfied by y = 1/x. What happens near (y, r) = (0, 0)? ♣

Remark 2.1 The Principle of Conservation of Stability (PCS) says that: Consider the ode ẋ = f (x, r), where f is
smooth and r is a parameter. Assign a weight w = 1 to each stable critical point, a weight w = −1 to each unstable
critical point, and a weight w = 0 to each semi-stable critical point. Then the sum of the weights (the stability index
S) does not change as r crosses a bifurcation value. ♣

dx x
Part 2 of 3. Consider the equation = rx− √ , (2.2)
dt 1 + x2

and repeat the analysis in part 1. Important: Be careful when doing the transformation to the variable y, as things
are not entirely smooth at y = 0. It follows that what happens near (y, r) = (0, 0) does not fit the “standard”
canonical forms studied in the lectures. Nevertheless, you should be able to do it with minimum effort.
dx
Part 3 of 3. Consider the equation = r x − x sech(x), (2.3)
dt

and repeat the analysis in part 1. Note: the situation near (y, r) = (0, 0) is even less “friendly” than the one in
part 2. Yet, it is still tractable if you are careful.

2.2 Answer: Find and classify bifurcations problem #02

Part 1. The critical points of (2.1) are given by
r
1 1 1
x = 0 for any r and x = ± √ ± − 1 for any 0 < r ≤ . (2.4)
r 4r 4

x2
The second formula arises from the equation r = , with four solutions for 0 < r < 1/4, one solution (x = 0)
(1 + x2 )2
for r = 0, and two solutions (x = ±1) for r = 1/4. The stability of the critical points is easy to ascertain by looking
at the sign of ẋ in each of the four regions that (2.4) splits the plane (r, x) into — this sign is represented by the
arrows in figure 2.1.
We conclude that three bifurcations occur: two saddle-nodes (at (r, x) = (1/4, ±1)) and a super-critical (soft)
pitchfork (at (r, x) = (0, 0)). The bifurcation diagram is shown in figure 2.1
The stability index S satisfies: S = 1 for r < 0, and S = −1 for r > 0. This seems to indicate a violation of the
PCS. However, consider the equation satisfied by y = 1/x

dy y3
= −(r y − ). (2.5)
dt 1 + 2 y2 + y4

This is the same as (2.1) backwards in time. Since y = 0 corresponds to x = ∞, we conclude:

a. x = ∞ is a critical point for (2.1), unstable for r < 0 and stable for r > 0. With this extra critical point: S ≡ 0
and the PCS holds true.
b. At (r, x) = (0, ∞) a subcritical (hard) pitchfork bifurcation occurs.
18.353 MIT, (Rosales) Find and classify bifurcations problem #02. 4

3 2 2
Bifurcation diagram for dx/dt = rx − x /(1+x ) .
3

In each region (yellow or cyan), the black arrows indicate

1
the direction of the flow for the
x0 x3
equation ẋ = r x − .
(1 + x2 )2

−1
Stable branches of critical points are plotted in solid
blue, and unstable branches in dashed red. The black
dots indicate the bifurcation points: two saddle-nodes at
−2
(r, x) = (1/4, ±1), and a supercritical (soft) pitch-
fork at r = x = 0.
−3
−0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5
r
Figure 2.1: Bifurcation diagram for equation (2.1).

Part 2. The critical points of (2.2) are given by

r
1
x = 0 for any r and x = ± − 1 for any 0 < r ≤ 1. (2.6)
r2
1
The second formula arises from the equation r = √ , with two solutions for 0 < r < 1, and one solution (x = 0)
1 + x2
for r = 1. Proceeding as in part 1, we arrive at the results summarized in figure 2.2. In particular: a subcritical
(hard) pitchfork bifurcation occurs at (r, x) = (1, 0).
2 1/2
Bifurcation diagram for dx/dt = rx − x/(1+x ) .
3

1
In each region (yellow or cyan), the black arrows indicate
the direction of the flow for the
x0 x
equation ẋ = r x − √ .
1 + x2
−1
Stable branches of critical points are plotted in solid
blue, and unstable branches in dashed red. The black
dot indicates the bifurcation point: a subcritical (hard)
−2
pitchfork at (r, x) = (1, 0).

−3
−0.5 0 0.5 1 1.5
r
Figure 2.2: Bifurcation diagram for equation (2.2).

As in part 1, a violation of the PCS seems to occur, with S = 1 for r < 0, and S = −1 for r > 0. Thus, we inspect
the behavior near x = ∞ by considering the equation satisfied by y = 1/x
dy y |y| dy
= −r y + p =⇒ ≈ (−r + |y|) y for |y|  1. (2.7)
dt 1 + y2 dt
18.353 MIT, (Rosales) Find and classify bifurcations problem #02. 5

The second equation here yields the (non-canonical) bifurcation diagram in figure 2.3 — note that, unlike the parabolic
shape of the bifurcating branches that occurs for canonical pitchfork bifurcations, here a curve with a corner appears.

Bifurcation diagram for dy/dt = (−r+|y|) y.

3

In each region (yellow or cyan), the black arrows indicate

1
the direction of the flow for the
y0 equation ẏ = −r y + |y| y.
Stable branches of critical points are plotted in solid
−1 blue, and unstable branches in dashed red. The black
dot indicates the bifurcation point: a (non-canonical)
−2
subcritical (hard) pitchfork at r = y = 0. Note that
the branches of critical points join with a corner at the
−3
bifurcation.
−1 −0.5 0 0.5 1 1.5 2 2.5 3
r
Figure 2.3: Bifurcation diagram for the second equation in (2.7).

Since y = 0 corresponds to x = ∞, we conclude that

c. x = ∞ is a critical point for (2.2), unstable for r < 0 and stable for r > 0. With this extra critical point: S ≡ 0
and the PCS holds true.
d. At (r, x) = (0, ∞) a (non-canonical) subcritical (hard) pitchfork bifurcation occurs.

Part 3. The analysis of (2.3) is entirely similar to that of the two previous cases, with critical points

x = 0 for any r and x = sech−1 (r) for any 0 < r ≤ 1, (2.8)

where we note that sech−1 (r) is double valued for 0 < r < 1. The results summarized in figure 2.4. In particular: a
subcritical (hard) pitchfork bifurcation occurs at (r, x) = (1, 0).
Again, we resolve the (apparent) failure of the PCS at r = 0 by inspecting the equation satisfied by y = 1/x
    
dy 1 dy 1
= −r y + y sech =⇒ ≈ −r + 2 exp − y for |y|  1. (2.9)
dt y dt |y|

Unlike the situation in (2.7), the right hand side here is smooth. However, the perturbation to the linear part
−r y, and all its derivatives, vanish at y = 0 — which makes this a very special problem. The second equation here
yields the (non-canonical) bifurcation diagram in figure 2.5 — note that, unlike the parabolic shape of the bifurcating
branches that occurs for canonical pitchfork bifurcations, here an “infinitely flat” curve appears. In conclusion:
e. x = ∞ is a critical point for (2.2), unstable for r < 0 and stable for r > 0. With this extra critical point: S ≡ 0
and the PCS holds true.
f. At (r, x) = (0, ∞) a (non-canonical) subcritical (hard) pitchfork bifurcation occurs.
18.353 MIT, (Rosales) Toy model for column buckling. 6

Bifurcation diagram for dx/dt = rx − x sech(x).

3

1 In each region (yellow or cyan), the black arrows indicate

the direction of the flow for the
x0
equation ẋ = r x − x sech(x).

−1
Stable branches of critical points are plotted in solid
blue, and unstable branches in dashed red. The black
dot indicates the bifurcation point: a subcritical (hard)
−2
pitchfork at (r, x) = (1, 0).

−3
−0.5 0 0.5 1 1.5
r
Figure 2.4: Bifurcation diagram for equation (2.3).

Bifurcation diagram for dy/dt = (−r + 2 exp(−1/|y|)) y.

0.6

0.4

In each region (yellow or cyan), the black arrows indicate

0.2
the direction of the flow for the
y 0 equation ẏ = −r y + 2 y exp(−1/|y|).

−0.2
Stable branches of critical points are plotted in solid
blue, and unstable branches in dashed red. The black
−0.4 dot indicates the bifurcation point: a (non-canonical)
subcritical (hard) pitchfork at r = y = 0. The unsta-
−0.6 ble branches at the bifurcation point are infinitely flat,
without the parabolic shape of a canonical pitchfork.
−0.4 −0.2 0 0.2 0.4 0.6 0.8 1
r
Figure 2.5: Bifurcation diagram for the second equation in (2.9).

3 Toy model for column buckling

3.1 Statement: Toy model for column buckling
Imagine a vertical cylindrical elastic column, on which you push down along its axis by putting a weight on top. If
the load is small enough, the column compresses a little, and the elastic response can balance the weight — with the
cylinder staying straight. But, if the load is too large, this configuration is not stable, and the column buckles under
the weight. This behavior arises because of the interplay of three forces: (i) the load; (ii) the elastic force along the
axis of the cylinder; and (iii) the restoring force that is generated when the cylinder bends. When the axial forces
are too large, the bending resistance is not enough to keep the straight state stable.
In this exercise we consider a very simple (1-D) toy model, exhibiting the essentials of the behavior described in the
prior paragraph. Note though that it is an over-simplified “toy” model, where all the richness of the original setting
is gone, and only the column buckling bifurcation remains.
A sketch depicting the model is shown in figure 3.1. Further assumptions and notation are:
18.353 MIT, (Rosales) Toy model for column buckling. 7

k A bead of mass m (black square) can slide along a rigid

d
horizontal rod (in red). The bead is connected by two
rod s x equal springs (in blue), with spring constants k, to two
supports placed symmetrically a distance d above and be-
m low the rod. A third spring, with spring constant s, pulls
the bead towards the middle of the vertical line connecting
k the supports for the other springs. Everything is friction-
frictionless hinge less, except for the friction force opposing the motion of
the bead along the rod. See the text for further details.

Figure 3.1: Toy model for column buckling.

1. Note that the whole device is restricted to a plane, with the bead moving along a line.
2. Let x be the distance, along the rod, of the bead from the vertical line joining the spring supports. Let x > 0
if the bead is to the right of the supports and x < 0 if to the left. Note the left-right (x → −x) symmetry
of the set-up.
3. The two main springs are equal, with a rest lengths L > 0 and spring constants k > 0. Each generates a force
along its axis of magnitude (Hook’s law) F = k (` − L), where ` is the spring length. They push if ` < L, and
pull if ` > L.
4. The spring aligned with the rod has zero rest length and a spring constant s > 0. This spring generates a
restoring force F = −s x along the rod, pulling the bead towards x = 0. Note: a “better” model would have
the restoring force provided by a torsion spring located at the hinge between the two springs on the bead. Such
a spring would generate a restoring torque proportional to the angle between the two main springs. However,
there is no qualitative difference between these two set-ups — and the one here yields simpler algebra. 1
5. When the bead slides along the rod, the motion is opposed by a friction force of magnitude b ẋ, where b > 0 is
a constant.
6. The distance of the main spring supports from the rod is d > 0. Instead of considering the behavior of the
system as a function of an applied compression force, we will consider it as a function of the total “imposed”
length 2 d of the “column”.
7. Because the rod is rigid, we need to consider only the horizontal components of the various forces that act on
the bead. These are the forces provided by the three springs, and friction along the rod.

PROBLEM TASKS:
A. Derive an ode for the bead position, and write it in appropriate a-dimensional variables. 2
B. Assume that friction is large, so that inertia can be neglected. Exactly which a-dimensional number has to be
small for friction to be “large”?
C. Analyze the bifurcations that occur for the equation resulting from item B, as the distance d changes (with
everything else fixed). What type of bifurcation(s) occur?
D. Consider the model that results from neglecting inertia. The equation for this model can, with an appropriate
scaling, be written in such a way that it contains a single a-dimensional parameter. Exhibit this form.
L s
To standardize the notation used in the answers, define a= and γ = . (3.1)
d 2k
1 Both models are over-simplifications of the situation described in the first paragraph of the exercise. There is no point in worrying about

getting small details right, when whooping simplifications occur elsewhere.

2 Suggestion: to a-dimensionalize use d for length and b/(2 k) for time.
18.353 MIT, (Rosales) Toy model for column buckling. 8

3.2 Answer: Toy model for column buckling

Newton’s law for the motion of the bead takes the form
x  p 
m ẍ + b ẋ = 2 k √ L − x2 + d2 − s x, (3.2)
x2 + d2
p
where the factor 2 k arises because there are two springs, and the factor x/ x2 + d2 is to compute the projection
along
p the rod of the spring’s forces. Note also that the signs are correct: when the springs are under compression
( x2 + dp2 < L), and x > 0, the springs should be pushing x to the right — with the force sign switching if either

x < 0 or x2 + d2 > L.
b
Select a-dimensional variables via x = d x̃ and t = t̃. The equation then becomes
2k
x  p 
 ẍ + ẋ = −γ x + √ a − 1 + x2 , (3.3)
1 + x2
where we have not written the tildes to simplify the notation,
2km L s
= , a= , and γ = . (3.4)
b2 d 2k
If   1, we can neglect inertia. Thus we arrive at the final equation (the toy model equation)
√ !
1 + x2 − a  p  x
ẋ = − γ + √ x = − (1 + γ) 1 + x2 − a √ = f (x, a, γ). (3.5)
1+x 2 1 + x2

a < 1+! y
a > 1+!

Plots of the function f = f (x, a, γ) on the right of equa-

a = 1+! x tion (3.5), with γ fixed. As a = L/d increases (d decreases
— see figure 3.1), the critical point at x = 0 (straight col-
umn) looses stability,
p and two new (stable) critical points
appear at x = ± (a/(1 + γ))2 − 1 — column buckles. A
supercritical (soft) pitchfork bifurcation.

Figure 3.2: Bifurcation for the toy model for column buckling.

To understand the bifurcations in equation (3.5), we now study its critical points and their stability — see figure 3.2.
Three cases arise:
c1. Case a < 1 + γ. There is a single critical point, which is stable: 3 x2 = 0 (straight column). Note that, if
d > L, the column is under tension and the straight state should be stable — which it is, since then a < 1.
c2. Case a > 1 + γ.pThere are three critical points: p
x1 = − (a/(1 + γ))2 − 1, x2 = 0, and x3 = (a/(1 + γ))2 − 1.
It is easy to see that both x1 and x3 are stable, while x2 is unstable — the column buckles. Note that x3
corresponds to the configuration shown in figure 3.1, while x1 corresponds to the mirror-image configuration.
 p 
3 This is a simple consequence of the fact that, in this case, (1 + γ) 1 + x2 − a > 0 for all values of x.
18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 9

c3. Case a = 1 + γ. Only one critical exists: x2 = 0, which is stable. At a = 1 + γ a supercritical (soft) pitchfork
bifurcation occurs. As a decreases through 1 + γ, x2 looses stability, and the system moves to either x1 or x3 .
There is no abrupt transition here, because both x1 and x3 are small for 0 < a − (1 + γ)  1.

Finally, note that, in terms of τ = (1 + γ) t, equation (3.5) takes the form

dx  p  x
= r − 1 + x2 √ , (3.6)
dτ 1 + x2
which has the single non-dimensional parameter
a L 2k
r= = . (3.7)
1+γ d s + 2k
This is the single parameter controlling the behavior of the system. Everything else can be absorbed into a scale
change (either time or space).

Force balances and a bit of physical intuition

The behavior of the system in this problem is controlled by the following forces:
e1. The force F1 = −s x by the spring lined up with the rod, which pulls always inwards. 4
x  p 
e2. The force F2 = 2 k √ L − x2 + d2 resulting from the other two springs. This force can either push
x2 + d2
outwards (compressed springs) or pull inwards (stretched springs).
Note that, no matter what the system parameters are, this force pulls inwards if |x| is large enough. As we will
see, this is what makes the resulting bifurcation “soft”.
e3. Large enough friction along the rod, which allows us to simplify the equations by neglecting inertia. After this
simplification, friction imposes no further qualitative changes in the system behavior (but it influences the time
scale over which things happen).

In figure 3.3 we plot the non-dimensional version of these two forces:

x  p 
F1 = −γ x and F2 = √ a − 1 + x2 , (3.8)
1 + x2
with γ fixed and two illustrative values for a. The figure shows how the balance of these two forces yields critical
points, and bifurcations. As illustrated by the figure, the key facts in producing a supercritical (soft) pitchfork
bifurcation are:
k1. The forces are odd.
k2. For x > 0, y = F2 is more concave than y = −F1 . That is, F2 acts as a soft-spring, weakening with distance.
This is produced by the geometry of the problem.

The bifurcation then occurs as the difference in the effective spring constants at the origin (derivatives at x = 0)
switches sign.

Problems motivated by Strogatz’s book

The problems below are either from Strogatz’s book, or are motivated by problems there. If different, do the
version here, not the one in the book.
4 Here inwards means: towards x = 0, while outwards means away from x = 0.
18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 10

y
F , a > 1+!
2
Plot of y = −F1 (x) — for an illustrative value of γ, and of
y = F2 (x) — for two illustrative values of a. The plots are for
−F x ≥ 0 only — x ≤ 0 follows by reflection across the origin.
1
Critical points arise when the curves intersect. Their stability
F , a < 1+! can be assessed from the sign of F2 + F1 . A supercritical (soft)
2 pitchfork bifurcation arises because F2 is concave: when F2 is
above −F1 near the origin (thus x = 0 is unstable), a second
(stable) critical point occurs at some x∗ > 0.
x

Figure 3.3: Forces in the toy model for column buckling — see equation (3.8).

4 Problem 16.09.17 - (exponential to algebraic decay transition)

4.1 Statement for problem 16.09.17
Consider the equation 5 ẋ = −r x − x3 , with initial data x(0) = x0 . In the limit r ↓ 0, the solution to this equation
transitions from exponential decay as t → ∞ (when r > 0) to algebraic decay when r = 0.
Why? Because when x becomes very small, p the behavior of the solution is controlled by the term −r x if r > 0. On the other
√
hand, if r = 0, the exact solution x = x0 / 1 + 2 x20 t decays like 1/ 2 t. However, how do we know that the solutions decay when
r > 0? Because then ẋ/x = −(r + x2 ) < r, so that the solution size is less than |x0 | e−r t .

Assume that 0 < r  1. Then the solution should decay exponentially fast
x0
as t → ∞, yet (because r is small) the solution should also be “close” to the x∗ = p (4.1)
solution to the problem when the term −r x is neglected. That is: (4.1). 1 + 2 x20 t
How can these two things happen simultaneously?
(a) Use separation of variables (or any other method) to solve ẋ = −r x − x3 , x(0) = x0 , r > 0, analytically.
Hint: what equation does y = 1/x2 satisfy? Once you know y, finding the sign of x is easy.
(b) For the solution in item a, show that x ∼ C e−r t for t → ∞, where C is a constant that you should compute.
(c) For the solution in item a, show that x ∼ x∗ for 0 ≤ r t  1, where x∗ is as in (4.1). Notice that, as r ↓ 0, the
time interval over which this is valid gets larger and larger.
Hint: The solution in part (a) will involve e−rt . Considers what happens when r is small and 0 ≤ t  1/r, so
that rt is small. Use this to approximate e−rt in the solution.
(d) To get some intuition on what is going on, plot the exact solution you obtained in item a versus x∗ in (4.1)
[plot #1]. In addition, plot the exact solution you obtained in item a versus the approximation in item b [plot
#2]. Use the same parameter values for both plots, but different time ranges. Suggestion: Use x0 = 1 and
r = 0.01. Then, for plot #1 use the range 0 < t < 50, and for plot #2 the range 0 < t < 100.

4.2 Answer for problem 16.09.17

√
(A) It is easy to see that y = 1/x2 solves ẏ = 2 r y + 2. x0 r e−r t
x = xe (t) = p . (4.2)
From this, or from separation of variables, we obtain r + x20 (1 − e−2 r t )
√
x0 r
(B) From (4.2) it follows that, as t → ∞, xe ∼ C e−r t , where C=p . (4.3)
r + x20
(C) Let 0 ≤ rt  1, with 0 < r  1. Substitute e−rt ∼ 1 into the numerator of (4.2), and e−2rt ∼ 1 − 2rt into the
denominator. What follows is x∗ in (4.1). Hence xs ∼ x∗ for 0 ≤ t  1/r.
(D) See figure 4.1.
18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 11
1 1

0.8 0.8
x = solutions

x = solutions
0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 10 20 30 40 50 0 20 40 60 80 100
t = time t = time

Figure 4.1: (Problem 16.09.17). Plot of the solution to ẋ = −0.01 x − x3 , x(0) = 1, (black, solid line), versus: (i) Left panel:
x∗ in (4.1) (red, dashed line); and (ii) Right panel: C e−r t , (blue dashed line), where C is as in (4.3). Note that here r = 0.01 and
x0 = 1. The left panel illustrates the match between the exact solution and x∗ for 0 < t  1/r, while the right panel illustrates the
onset of exponential decay for t  1/r.

5 Problem 02.05.06 - Strogatz (The leaky bucket)

5.1 Statement for problem 02.05.06
(The leaky bucket). The following example 6 shows that in some physical situations, non-uniqueness is natural
and obvious, not pathological.
Consider a water bucket with a hole in the bottom. If you see a water bucket with a puddle beneath it, can you figure
out when the bucket was full? No, of course not! It could have finished emptying 7 a minute ago, ten minutes ago,
or whatever. The solution to the corresponding differential equation must be non-unique when integrated backwards
in time.
Here is a crude model for the situation. Let h(t) = height of the water remaining in the bucket at time t; a = area
of the hole; A = cross-sectional area of the bucket (assumed constant); v(t) = velocity of the water passing through
the hole.
a. Show that a v(t) = A ḣ. What physical law are you invoking? Warning: since ḣ < 0, this presumes that
we assign a negative value to the velocity v. This is a weird choice, implicit in the problem statement, but
acceptable.
b. To derive an additional equation, use conservation of energy. First, find the change in potential energy in the
system, assuming that the height of the water in the bucket decreases by an amount ∆h, and that the water
has density ρ. Then find the kinetic energy transported out of the bucket by the escaping water. Finally,
assuming all the potential energy is converted into kinetic energy, derive the equation v 2 = 2 g h — g = gravity
acceleration.
√ ap
c. Combining a and b, show that ḣ = −C h, where C = 2 g.
A
d. Given h(0) = 0 (bucket empty at t = 0), show that the solution for h(t) is non-unique in backwards time, i.e.,
for t < 0.

The description/derivation above ignores surface tension. Briefly discuss the effect surface tension has on the outcome.
5 Note that this equation is one of the two possible normal forms for pitchfork bifurcations.
6 Hubbard, J. H., and West, B. H. (1991) Differential Equations: A Dynamical Systems Approach, Part I (Springer, New York).
7 Note that, in this problem, evaporation effects are neglected.
18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 12

5.2 Answer for problem 02.05.06

a. Since water is conserved (conservation of mass), the rate at which the water leaves the bucket — i.e.: a v(t),
must equal the rate at which the water in the bucket decreases — i.e.: A ḣ. Hence: a v(t) = A ḣ. Warning:
since ḣ < 0, this presumes that we assign a negative value to the velocity v. This is a weird choice, implicit in
the problem statement, but acceptable.
Z h
1
b. The potential energy of the water in the bucket is given by V = g A ρ y dy = g A ρ h2 . The rate at which
0 2
1
kinetic energy is transported out of the bucket by the escaping water is K̇ = ṁ v 2 , where ṁ = ρ A ḣ is the
2
1
rate at which mass leaves the bucket. Hence K̇ = ρ A ḣ v 2 . We now invoke conservation of energy, neglecting
2
2
frictional losses, and equate V̇ = K̇ =⇒ v = 2 g h.
Note that an alternative expression for the rate at which kinetic energy is transported out of the bucket by
1
the escaping water is given by K̇ = ρ a v 3 — since in an infinitesimal time interval dt the kinetic energy
2
1
transported out is dK = (ρ a v dt) v 2 . Using a, it is easy to see that this is equivalent to the expression
2
derived in the prior paragraph.
c. A simple calculation using a and b (be careful with the signs, recall that: v < 0 and ḣ < 0) then shows that
√ ap
ḣ = −C h, where C = 2 g > 0, and the implicit restriction on the solutions given by h ≥ 0 applies.
A
d. Given h(0) = 0 (empty bucket), then the solution(s) to the o.d.e. derived in item c are:
- Since ḣ ≤ 0, and h must remain non-negative, the solution is unique for t > 0. Namely: h(t) ≡ 0.
- Using separation of variables, we can find the following (infinite number of) solutions, valid for t < 0 :
(
0 for t0 ≤ t ≤ 0,
h(t) = C

2
2 (t0 − t) for t ≤ t0 .

where t0 ≤ 0 is arbitrary.

What about surface tension? Surface tension affects this problem in (roughly) three ways: (i) On the top surface
it will create an extra force that resists the emptying of the bucket. However, by assumption, the top surface is
“large”, so this is not an important effect [certainly not for a bucket-sized container]. (ii) The jet that comes out
of the hole may be fractured into droplets by surface tension. Whether or not this happens is not relevant for this
problem. Once the water leaves the bucket, it no longer affects h. (iii) If the hole at the bottom is small enough,
surface tension may be able to stop the flow before h = 0. What happens beyond this depends on whether or not
the bucket surface is wetting. If it is not, then the interface at the hole will be pinned, with surface tension across
it able to support the pressure by a nonzero h in the bucket — a stable, steady, situation. If the bucket surface is
wetting, then the interface will spread, allowing more water to flow through the hole, eventually making the interface
unstable. Then a drop forms and falls, re-starting the process. The bucket continues to empty out, but not through
a jet, but drop-by-drop. Eventually h may be too small to support even this, and this process stops.

6 Problem 03.02.06 - Strogatz (Eliminate the cubic term)

6.1 Statement for problem 03.02.06
Consider the system
dX
= RX − X 2 + aX 3 + O(X 4 ), (6.1)
dt
18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 13

where R 6= 0. We want to find a new variable x such that the system transforms into
dx
= Rx − x2 + O(x4 ). (6.2)
dt
This would be a big improvement, since the cubic term has been eliminated and the error term has been bumped
to fourth order. 8 In fact, the procedure to do this (sketched below) can be generalized to higher orders. 9 This
generalization is the subject matter of problem 03.02.07.
Let x = X + bX 3 + O(X 4 ), where b is chosen later to eliminate the cubic term in the differential equation for
x. This is called a near-identity transformation, since x and X are practically equal: they differ by a cubic term. 10
Now we need to rewrite the system in terms of x; this calculation requires a few steps.
1. Show that the near-identity transformation can be inverted to yield X = x + cx3 + O(x4 ), and solve for c.
2. Write ẋ = Ẋ + 3bX 2 Ẋ + O(X 4 ), and substitute for X and Ẋ on the right hand side, so that everything
depends only on x. Multiply the resulting series expansions and collect terms, to obtain ẋ = Rx − x2 +
kx3 + O(x4 ), where k depends on a, b, and R.
3. Now the moment of triumph: choose b so that k = 0.
4. Is it really necessary to make the assumption that R 6= 0? Explain.

6.2 Answer for problem 03.02.06

We now fill in the steps outlined in the problem statement:
1. Replacing X = x + cx3 + O(x4 ) into x = X + bX 3 + O(X 4 ) yields:
x = (x + cx3 ) + b(x + cx3 )3 + O(x4 ) = x + (c + b)x3 + O(x4 ). Thus, it must be c = −b.
This process can be carried out to any order. If x = X + aX 2 + bX 3 + cX 4 + · · · + O(X N ), we can find the
inverse transformation X = x + Ax2 + Bx3 + Cx4 + · · · + O(xN ) by successively selecting the coefficients A,
B, C, . . . to eliminate the coefficients of the powers x2 , x3 , x4 , . . . in a substitution like the one above.
2. Write ẋ = Ẋ + 3bX 2 Ẋ + O(X 4 ), use equation (6.1) to eliminate Ẋ on the right hand side, and substitute
X = x − bx3 + O(x4 ) — as obtained in the first step — to eliminate X. This yields:
ẋ = Ẋ + 3bX 2 Ẋ + O(X 4 )
= (RX − X 2 + aX 3 ) + 3bX 2 (RX − X 2 + aX 3 ) + O(X 4 )
= RX − X 2 + (a + 3bR)X 3 + O(X 4 )
= R(x − bx3 ) − (x − bx3 )2 + (a + 3bR)(x − bx3 )3 + O(x4 )
= Rx − x2 + kx3 + O(x4 ), where k = a + 2 b R.
a
3. Now choose b so that k = 0. That is b=− . (6.3)
2R
4. Equation (6.3) shows that R 6= 0 is crucial for all of this to work. When R = 0, X 2 is the dominant term on the
right in (6.1), and the proposed form of the expansion does not work. It is still possible to eliminate the O(X 3 )
term in (6.1) — as well as any other higher order terms — when R = 0, but a DIFFERENT EXPANSION
IS NEEDED, including logarithmic terms. The first two terms in this expansion are: x = X+a X 2 ln X+. . .
Remark 6.2 What would have happened if we started with a more general form of the transformation relating x
and X, that is: x = X + qX 2 + bX 3 + O(X 4 )? Then, in the second step above the final answer would have taken the
form ẋ = Rx − px2 + kx3 + O(x4 ). Then the next step would have been to select q and b so that p = 1 and k = 0.
This would have given q = 0 and k = −a/2R. That is: the same answer as above. We have simplified the algebra by
taking q = 0 from the very beginning.

8 Obviously we are considering here a situation where X (and x) is small.

9 That is, one can successively eliminate all the higher order terms: O(x3 ), O(x4 ), . . . , etc.
10 We have skipped the quadratic term X 2 , because it is not needed — you should check this later.
18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 14

7 Problem 03.04.11 - Strogatz (An interesting bifurcation diagram)

7.1 Statement for problem 03.04.11
(An interesting bifurcation diagram). Consider the system
dx
= rx − sin(x). (7.1)
dt
A. For the case r = 0, find and classify the fixed points, and sketch the vector field.
B. Show that, when r > 1, there is only one fixed point. What kind of fixed point is it?
C. As r decreases from ∞ to 0, classify all the bifurcations that occur.
D. For 0 < r  1, find an approximate formula for the values of r at which bifurcations occur.
E. Now classify all the bifurcations that occur as r decreases from 0 to −∞.
F. Plot the bifurcation diagram for −∞ < r < ∞, and indicate the stability of the various branches of fixed points.

7.2 Answer for problem 03.04.11

The most efficient way to attack this problem is to, essentially, start by answering the last question (i.e.: F) being
asked. This we do next.
We note that the equilibrium points for equation (7.1), in the (r, x) plane, are given by the curves:
sin(x)
I. x = 0, for all − ∞ < r < ∞. and , for all − ∞ < x < ∞.
II. r =
x
These curves (see figure 7.1) divide the (r, x) plane into four regions, in each of which the flow direction is the same
(away or towards the r-axis). From this, the bifurcation points, and the stability of the various branches of solutions,
follows easily — as shown in figure 7.1. In particular, it is clear that the bifurcation points are given by the local
maximums and minimums of the curve in (II) above. Namely, by the solutions of
0 = sin(x) − x cos(x), (7.2)
with the corresponding values of r given by (II) above.

Bifurcation diagram for dx/dt = rx - sin(x).

10

5
The bifurcation points are shown by black dots, the
stable and unstable branches of equilibrium solutions
by the thick blue and dashed red lines, respectively.
x 0
The flow directions, which are the same in each of the
four colored areas, are shown by the arrows. There is
an infinite number of saddle-node bifurcations, and a
-5 single subcritical (hard) pitchfork bifurcation.

-10

-0.5 0 0.5 1
r
Figure 7.1: (Problem 03.04.11). Bifurcation diagram for ẋ = rx − sin(x).

We are now ready to answer all the questions:

18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 15

A. For r = 0 the equilibrium points are given by xn = n π, with n = 0, ±1, ±2 . . . Furthermore: xn is stable for
n even, and unstable for n odd. The vector field in this case is given by the r = 0 line in figure 7.1.
B. The answer is now obvious. For r > 1, there is only one fixed point, x = 0. This point is unstable.
C. As r decreases from ∞ to 0, the first bifurcation occurs at r = 1 — a subcritical (hard) pitchfork. As r continues
to decrease, a sequence of pairs of saddle node bifurcations occur, where progressively larger values of x appear.
Figure 7.1 should make this process clear.
D. For 0 < r  1, the values of x that correspond to bifurcations are large. Thus, to calculate these values, we
must solve equation (7.2) for |x|  1. Because the equation is invariant under the transformation x → −x, we
will only consider the case x > 0.
It is clear that x  1 solutions must correspond to approximate zeros of the cosine, else the term x cos(x) in
the equation cannot be balanced. Thus we write:

x∗ = cn + ,

where cn = 2n+1 2 π, n  1 is an integer, and  is small. Substituting this expression into equation (7.2),
expanding in Taylor series, using that sin(cn ) = (−1)n , cos(cn ) = 0, ||  1, |cn |  1, and keeping only the
leading order terms, we obtain  ≈ −1/cn . For r we have:

sin(x∗ ) (−1)n 2
r= ≈ = (−1)n , (7.3)
x∗ cn (2n + 1)π

where n must be large (and even, so r > 0).

E. As r increases from −∞ to 0, a sequence of pairs of saddle node bifurcations occur, where progressively larger
values of x appear. Figure 7.1 should make this process clear.
F. The bifurcation diagram for −∞ < r < ∞, with the stability of the various branches of fixed points indicated,
is in figure 7.1.

8 Problem 16.09.20 (growing yeast)

8.1 Statement for problem 16.09.20
A (late) medieval 11 baker grows yeast in a vat in his basement. Every day he collects the amount he needs to make
bread, which is more or less constant. He is very careful to keep the vat at constant temperature (warm, but not
 
hot), and feeds the yeast regularly with water and flour. dN N
An extremely simple model for the yeast is in (8.1). = r N 1 − − H. (8.1)
dt K
In this model, in the absence harvesting, the yeast the population is assumed to grow logistically. The effects of the
harvesting are modeled by the (constant) term −H < 0.
Remark 8.3 The fact that the yeast is collected at a constant rate H > 0, independent of the amount present N , is
silly. However, the yeast collection is done by the baker’s apprentice, which is not too bright. He simply goes down to
the basement (every day) with the same jar, which he fills to the brim. And, if he happens that there is not enough
to fill the jar, that day he takes as much as there is. 12
Questions and tasks.
dx
a. Show that the system can be rewritten in dimensionless form as = x (1 − x) − h, (8.2)
dτ
for suitable defined dimensionless quantities x, τ , and h.
11 Late medieval because the use of yeast to make bread did not become common till the Renaissance, at least in Europe.
12 Next day the baker gets another apprentice.
18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 16

b. Plot the bifurcation diagram (as a function of the parameter h) for equation (8.2), and discuss the various
behaviors (in time) that the solutions exhibit in the “physical” regime (x, h ≥ 0).
c. Show that a bifurcation occurs at a certain value h = hc , and classify this bifurcation.
d. Discuss the long-term behavior of the yeast in the vat for h < hc and h > hc , and give the biological interpre-
tation in each case.
e. The fact that the harvesting rate is constant leads to silly behavior of the model solutions: the amount of yeast
can become negative! Fix the model so as to incorporate into it the exact behavior by the apprentice described
in remark 8.3. Revise your analysis above in view of this change.

Note: a smart apprentice would adjust his collection strategy to depend on N , at the very least. In item e you are
not being asked to write a model for an improved strategy. The question is: how do you use (8.1) so that it mimics
the actual behavior described in remark 8.3?

Remark 8.4 Details about the continuum limit. In order to arrive at an ode, as in (8.1), in a situation where
there is a clear discrete event (collecting the yeast once a day, as in remark 8.3) you need the changes between
discrete events to be small, so you can approximate the process by a continuous function. But yeast can grow very
fast, as much as duplicate its mass in as little as 90 minutes. So, the stuff above contradicts this, and makes the
funny story I was trying the embed the problem into fail.13 Here are a couple of ways the science can be fixed:
(1) The apprentice collects the yeast much more frequently, in smaller batches.
(2) The vat is not in as good shape as stated earlier, so the yeast grows a lot slower.
At any rate, even if a late medieval bakery is not the best example giving rise to (8.1), there are many situations
where something is being produced by either a biological or a chemical process, and that something is being collected
more-or-less continuously at some rate. Equation (8.1) models the simplest set-up of this kind.

8.2 Answer for problem 16.09.20

a. Let x = N/K, τ = rt, & h = H/(rK). Equivalently: N = Kx, t = τ /r, & H = rKh.
This reduces equation (8.1) to (8.2). We can also write the equation in the form

dX 1 1 − 4h
= R − X 2, where X =x− and R = . (8.3)
dτ 2 4
b. Since the equation can be written in the form ẋ = −(x − 1/2)2 − (h − 1/4), it is easy to see that it yields the
bifurcation diagram in figure 8.1. From this diagram we see that there are several cases to consider in terms of
Bifurcation diagram: dx/dt = x(1ïx)ïh
1

0.75
Figure 8.1: (Problem 16.09.20). Bifurcation
diagram for equation (8.2). A saddle node bi-
furcation occurs for h = 1/4 and x = 1/2. The
x
0.5 first quadrant only is displayed, as only x ≥ 0,
h ≥ 0 make sense. The stable (unstable, resp.)
branches of critical points are plotted by solid blue
0.25
(dashed red, resp.) lines. The arrows indicate the
flow direction.

0
0 0.25 0.5
h
the behavior in time of the solutions.
13 If you wish, the same problem any movie involving science and technology runs into. If you put all the details, the story flops.
18.353 MIT, (Rosales) Problems motivated by Strogatz’s book. 17

b1. h > 0.25.

The solutions are strictly monotone decreasing, since ẋ ≤ 0.25 − h < 0. They start at x = −∞ at some
finite time, and reach x = ∞ at a later finite time — note that they become negative at some point. They
have an inflection point at x = 0.5, and are convex (ẍ > 0) before, and concave (ẍ < 0) thereafter.
Proof: The only issue is why do the solutions go from −∞ to ∞ in a finite time. This is because, for |x| large
the equation reduces to ẋ = −x2 , whose solutions reach infinity in a finite time.
√
Note that, in fact, exact solutions are available: x = 12 − µ tan(µ(τ − τ0 )), where µ = h − 0.25. But we do
not need this information for the arguments here.
√
b2. 0 ≤ h < 0.25 and x is below the unstable critical point (x < xL = 0.5 − µ, where µ = 0.25 − h).
The solutions are strictly monotone decreasing, concave, and defined for −∞ < τ < τ0 . They approach
the critical point xL as τ → −∞, and x = −∞ as τ → τ0 . They become negative at some point in time.
1
Exact solutions: x = 2 + µ coth(µ(τ − τ0 )), with τ < τ0 .
√
b3. 0 ≤ h < 0.25 and x is between the critical points (xL < x < xR = 0.5 + µ, where µ = 0.25 − h).
The solutions are sigmoids, going from x = xL (at τ = −∞) to x = xR (at τ = ∞).
1
Exact solutions: x = 2 + µ tanh(µ(τ − τ0 )).
√
b4. 0 ≤ h < 0.25 and x is above the stable critical point (x > xR = 0.5 + µ, where µ = 0.25 − h).
The solutions are strictly monotone decreasing, convex, and defined for τ0 < τ < ∞. They approach the
critical point xR as τ → ∞, and x = ∞ as τ → τ0 .
b5. h = 0.25.
The behavior is the “limit” of either b2 or b4, except that the approach to the critical point (as τ → ∞
if above the critical point, or τ → −∞ if below it) is algebraic, instead of exponential.
1 1
Exact solutions: x = 2
+ τ −τ0
.
c. From equation (8.3), it should be obvious that a saddle-node bifurcation occurs for h = hc = 0.25, and that
there is no other bifurcation.
d. When the solution becomes negative, the model ceases to make sense. The problem is caused by the
hypothesis of a constant harvest rate — which is non-sense once the amount of yeast available becomes small
enough. To make sense of the model in these cases, we will “cut-off” the differential equation when N
becomes zero, and assume that an “extinction event” has occurred — i.e.: the apprentice took all the yeast
from the vat at some point in time, see remark 8.3. With the issue of meaning for the negative solutions settled,
we are now ready to give biological interpretations.
Case h > hc = 0.25. The harvest rate is too large, above any rate at which the yeast can ever grow. The
yeast vat is depleted, always.
Case 0 < h < hc = 0.25. The harvest rate is small enough that the yeast growth can compensate for it.
There are two critical populations at which the harvest rate exactly balances the growth rate:
q q
xL = 21 − 1−4h 4 and xU = 12 + 1−4h4 .

xL is unstable: small decreases (increases) in population decrease (increase) the growth rate — resulting in either
an extinction event, or growth towards xU (which is stable).
e. See the answer to d.

THE END.