Answers to P-Set # 05, 18.385j/2.

036j
MIT (Fall 2020)
Rodolfo R. Rosales (MIT, Math. Dept., room 2-337, Cambridge, MA 02139)
November 5, 2020

Contents
1 Large µ limit for Liénard system #03 2
1.1 Statement: Large µ limit for Liénard system #03 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Answer: Large µ limit for Liénard system #03 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Phase Plane Surgery #01 4

2.1 Statement: Phase Plane Surgery #01 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Answer: Phase Plane Surgery #01 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 Simple Poincaré Map for a limit cycle #02 7

Stability of a limit cycle in a 2-D system by (explicitly) computing the Poincaré map . . . . . . . . . . . . 7
3.1 Statement: Simple Poincaré Map for a limit cycle #02 . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.2 Answer: Simple Poincaré Map for a limit cycle #02 . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4 Problem 07.02.x1 (area evolution) 9

4.1 Problem 07.02.x1 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2 Problem 07.02.x1 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

5 Problem 07.05.06 - Strogatz (Biased van der Pol) 11

5.1 Problem 07.05.06 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5.2 Problem 07.05.06 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

6 Problem 08.02.05 - Strogatz (Hopf bifurcation using a computer) 17

6.1 Problem 08.02.05 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
6.2 Problem 08.02.05 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

7 Problem 08.02.07 - Strogatz (Hopf & homoclinic bifurcations via computer) 19

7.1 Problem 08.02.07 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
7.2 Problem 08.02.07 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

8 Problem 08.04.03 - Strogatz (Homoclinic bifurcation via computer) 21

8.1 Problem 08.04.03 statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
8.2 Problem 08.04.03 answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
8.2.1 Analysis for the Hopf bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

List of Figures
1.1 Liénard system #03 with large µ. Phase portrait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.1 Phase portrait with three limit cycles and three critical points . . . . . . . . . . . . . . . . . . . . . . . 7
5.1 Problem 07.05.06. Biased van der Pol (a2 > 1). Nullclines and flow field . . . . . . . . . . . . . . . . . 12
5.2 Problem 07.05.06. Biased van der Pol (a2 < 1). Nullclines and flow field . . . . . . . . . . . . . . . . . 12
5.3 Problem 07.05.06. Biased van der Pol (µ  1); phase portrait . . . . . . . . . . . . . . . . . . . . . . . 15

1
18.385 MIT, (Rosales) Large µ limit for Lienard system #03. 2

5.4 Problem 07.05.06. Biased van der Pol (0 < 1 − a  1); limit cycles . . . . . . . . . . . . . . . . . . . . 16
6.1 Problem 08.02.05. Stable spiral point before a Hopf bifurcation . . . . . . . . . . . . . . . . . . . . . . 18
6.2 Problem 08.02.05. Soft (supercritical) Hopf bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
7.1 Problem 08.02.07. Stable spiral point before a Hopf bifurcation . . . . . . . . . . . . . . . . . . . . . . 19
7.2 Problem 08.02.07. Soft (supercritical) Hopf bifurcation . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
7.3 Problem 08.02.07. Limit cycle about to undergo a homoclinic bifurcation . . . . . . . . . . . . . . . . 21
8.1 Problem 08.04.03. Phase portraits slightly before an homoclinic bifurcation . . . . . . . . . . . . . . . 23
8.2 Problem 08.04.03. Phase portrait slightly after an homoclinic bifurcation . . . . . . . . . . . . . . . . 23

1 Large µ limit for Liénard system #03

1.1 Statement: Large µ limit for Liénard system #03
A Liénard equation has the form ẍ + µ f 0 (x) ẋ + g(x) = 0, (1.1)
for some functions f and g. Here µ > 0 is a parameter.
d
This can be re-written in the form (ẋ + µ f (x)) + g(x) = 0. (1.2)
dt
1
Introduce y = µ ẋ + f (x), to get the system
1
ẋ = µ (y − f (x)) and ẏ = − g(x). (1.3)
µ
1 3 1 5
In this problem we will consider the case f (x) = −x + x − x and g(x) = x, (1.4)
3 60
with µ  1.
Analyze the large µ limit for this system. In particular:
1. Are there any limit cycles? Are they stable, unstable, semi-stable?
2. Are there any critical points? Are they attractors, repellers?
3. Does the system have any global attractor?
4. Sketch the phase plane portrait.

1.2 Answer: Large µ limit for Liénard system #03

The behavior in the µ  1 limit is controlled by the following facts:
f01. g(x) > 0 for x > xg = 0, and g(x) < 0 for x < xg .
Thus ẏ < 0 for x > xg and ẏ > 0 for x < xg .
Let yg = f (xg ) = 0.
f02. The curve y = f (x) (see the dashed black line in figure 1.1) has four special points {xn }n=1 to 4 , with
corresponding values yn , given by its local maximums/minimums. We have
1) x1 < x2 < xg < x3 < x4 .
2) x1 and x3 are local minimums, with y1 < y3 .
3) x2 and x4 are local maximums, with y2 < y4 .
4) The curve is (strictly) monotonic in each of the intervals open intervals In = (xn , xn+1 ),
n=0:4, where x0 = −∞ and x5 = ∞ — decreasing, increasing, decreasing, . . .
5) y → ±∞ as x → ∓∞.
Notation: The branch In means the portion of the curve given by y = f (x), x ∈ In .
Notation, for points: Pn = (xn , yn ) and C = (xg , yg ) = critical point.
C is the only critical point.
18.385 MIT, (Rosales) Large µ limit for Lienard system #03. 3

f03. Since µ  1, y changes very slowly, while x changes very fast (except when y − f (x) is small).

It follows that:
f04. For y − f (x) > 0 (but not too small), the phase plane trajectories are nearly horizontal lines, with the flow
left to right.
f05. For y − f (x) < 0 (but not too small), the phase plane trajectories are nearly horizontal lines, with the flow
right to left.
f06. In the t growing direction the trajectories get trapped in a narrow band near the branches I1 and I3 — which
they can exit only near the points P2 and P3 , see item f08.
These are the stable/attracting branches.
f07. In the t decreasing direction the trajectories get trapped in a narrow band near the branches I0 , I2 , and I4
— which they can “exit” only near the points P1 for I0 , P2 for the left half of I2 , P3 for the right half of I2 ,
and P4 for I4 , see item f08.
These are the unstable/repelling branches.
Hint: the easy way to figure out the phase portrait near an unstable branch, is to draw it for the reversed flow
(t decreasing), and then reverse the arrows.
f08. In the trapping narrow band near each branch, the solutions slowly move up or down, depending on the sign
of g. That is:
Up for I0 , I1 , and the left side (x < xg ) of I2 .
Down for I4 , I3 , and the right side (x > xg ) of I2 .
Eventually (forward/backwards in time for the stable/unstable branches) the solutions get pushed past one of
the points Pn . At this point the trajectories “exit” the narrow trapping band, and the trajectories switch to
the behavior in either item f04 or item f05.
Task #1 left to the reader. How do the solutions behave in the trapping bands near the branches? In particular: Do
the solutions cross the branches? If so, which solutions, and how is the crossing? Is the narrow trapping band on one
side of the branch (which side?), or does it enclose the branch?
Task #2 left to the reader. How do the solutions behave near the points Pn , n=1:4? Sketch a detail of the phase portrait
near these points.
f09. From the facts above, it follows that all the orbits leave the critical point C along the narrow band near I2 , up
to the left, or down to the right. This with two exceptions: the nearly horizontal orbits that leave C to the left,
or to the right (these two orbits are the transition from the orbits leaving up to the left, and the ones leaving
down to the right). It follows that C is an unstable node.
Task #3 left to the reader. How do the solutions behave near the critical point C? Sketch a detail of the phase portrait
near this point, including the branch I2 .

Using the facts above, it is now easy to figure out what the orbits do — see figure 1.1. The conclusion is that:
c1. There are two limit cycles, both clockwise. The larger one is unstable, and encloses the smaller one, which is
stable.
The larger one consists (approximately) of two horizontal segments (from the branch I0 to P4 , and from the
branch I4 to P1 ), plus the complimentary pieces from the branches I0 and I4 .
The smaller one consists (approximately) of two horizontal segments (from P2 to the branch I3 , and from P3
to the branch I1 ), plus the complimentary pieces from the branches I1 and I3 .
c2. The orbits that leave the larger limit cycle either zoom to infinity, or converge towards the smaller limit cycle.
The orbits that leave the unstable node C converge towards the inner limit cycle.
18.385 MIT, (Rosales) Phase Plane Surgery #01. 4

Large µ limit in Lienard #03, µ = 50

3

Liénard system large µ limit #03.

1
The dashed blue (resp. black) line in-
y 0 dicates the zero of g (resp. the curve
y = f (x)). The critical point C is an
−1 unstable node. The two limit cycles can
be seen in thicker red than the other or-
−2
bits.
−3
−5 −4 −3 −2 −1 0 1 2 3 4 5

Figure 1.1: Phase portrait for the system in (1.3 – 1.4).

Task #4 left to the reader. Consider the solutions trapped near the branch I0 . Eventually these solutions leave the trapping
region, either to the right or the left. There must be an orbit that is the transition between these two behaviors, and stays close
to the branch all the way to infinity. Find an approximate expression for this orbit.

Task #5 left to the reader. There are two orbits that leave C (one up to the left, the other down to the right) and stay close
to the branch I2 all the way to either P2 or P3. Find an approximate expression for these orbits.

2 Phase Plane Surgery #01

2.1 Statement: Phase Plane Surgery #01
Can a smooth vector field exist in the plane such that:
— The critical points are P1 = (−2, 0), P2 = (0, 0) and P3 = (2, 0).
— All the critical points are spirals.
— The circles with radii: R1 = 1 centered at P1 , R2 = 4 centered at P2 , and R3 = 1 centered at P3 , are orbits.

Would your answer change if P2 is a saddle?

In either case, if your answer is yes, sketch the way the orbits might look in an example satisfying the criteria above.
Challenge question: In either case, if your answer is yes, can you give an actual example (i.e.: write the vector field
explicitly) that gives you a phase portrait with the same qualitative features (the closed orbits need not be circles
for this).

2.2 Answer: Phase Plane Surgery #01

The answer to the first question in this problem is: NO.
The reason is that the circle of radius R2 centered at P2 would be an orbit enclosing all three critical points. Thus
the sum of the indexes of the critical points would have to be one, which would contradict the fact that the critical
points are all supposed to be spirals.
18.385 MIT, (Rosales) Phase Plane Surgery #01. 5

If, on the other hand, P2 is a saddle, then a vector field with the given properties exists.
We will show this by actually constructing one. In our construction the closed orbits doing the job are not be circles,
but this is a minor point: The method used to construct the example can be adapted (at the price of much more algebra
in the answer) to obtain closed orbits that are circles — or ellipses or any other closed curves one may desire.
1
Let V be the potential defined by V = −2 x2 + x4 . For this potential
4
— x = 0 is a local maximum — with V (0) = 0.
— x = ±2 are local minimums — with V (−2) = V (2) = −4.
dV
— Everywhere else 6= 0.
dx
1 2
Define E = E(x, y) by E= y + V (x). (2.1)
2
Then
 
dV
1. ∇E = (Ex , Ey ) = ,y vanishes at the points P1 , P2 , and P3 , only.
dx
2. P1 and P3 are local minimums of the surface z = E(x, y), while P2 is a saddle.
3. The level curves E = c = constant (with −4 < c < 0) have two components: one enclosing (only) P1 and the
other enclosing (only) P3 .
4. The level curves E = c = constant (with 0 < c) have only one component, which encloses all the Pj ’s and all
the level curves with −4 < E < 0.

Consider now the phase plane system given by:

dx
= −Ey − f (E) Ex , !
dt d x
i.e.: = (∇E)⊥ − f (E) ∇E, (2.2)
dy dt y
= Ex − f (E) Ey ,
dt
where f = f (E) is some function to be defined later, and the superscript ⊥ indicates rotation by π/2 of a vector in
the plane.

Remark 2.1 (Motivation). The flow provided by the system of equations in (2.2) has two components. The first
one, given by (∇E)⊥ , induces a flow along the level lines of the surface z = E(x, y) — for this component alone
the orbits would be the level curves of E. The second component, given by f ∇E, induces a flow normal to the level
lines of E — for this component alone the orbits would correspond to the path of water flowing down/up 1 the gravity
gradient on the surface z = E.
Thus, for the system in (2.2), the level curves of E corresponding to zeros of f (E) are orbits. On the other hand,
when f is small, the orbits will (approximately) track the level curves of E — slowly drifting up or down the gradient
of E, depending on the sign of f . Thus, given the properties of E — itemized below equation (2.1), it should be clear
that: by judiciously choosing f = f (E), we can obtain a system with the desired properties.
A clarification regarding orbits and level lines: when above we say “the orbits would be level lines of E”, or similar,
it should be clear that this applies for each component of E = constant, if E = constant is made up by more than
one curve — e.g.: when −4 < E < 0. The case E = 0 (saddle level) has three components, one where x > 0, another
where x < 0, and the critical point P2 . Finally, the case E = −4 has two components: the critical points P1 and P3 .
♣
1 Down if f > 0; up if f < 0.
18.385 MIT, (Rosales) Phase Plane Surgery #01. 6

Below we fill in the details of the intuitive arguments in the remark.

(
f ẋ − ẏ = −(1 + f 2 ) Ex ,
For the system in (2.2), we have (2.3)
f ẏ + ẋ = −(1 + f 2 ) Ey ,

from which it follows that the critical points of (2.2) are exactly the Pj ’s — since these are the only points where
∇E vanishes. It is also clear that, along orbits:
dE
= ẋ Ex + ẏ Ey = − Ex2 + Ey2 f (E).


(2.4)
dt
Hence: Any level curve E = E0 = constant, such that f (E0 ) = 0, is an orbit. (2.5)

Now take f =  (E 2 − 4), (2.6)

where  > 0 is an arbitrary constant. With this choice for f = f (E), note that:
√
5. The critical points P1 = (−2, 0) and P3 = (2, 0) are stable spirals if  < 2/49 ≈ 0.0673 . . ., or stable improper
√ √
nodes if  = 2/49, or stable nodes if 2/49 < .
" #
This follows from the 2 × 2 matrix of the linearized −96  −1
A= .
equations near the critical points, both given by: 8 −12 
6. The critical point P2 = (0, 0) is always a saddle.
" #
This follows from the 2 × 2 matrix of the linearized −16  −1
A= .
equations near the critical point, given by: −4 4 
7. For any orbit that is not a critical point, (2.4) yields

As t→ ∞ E→ 2 or E → −4.
(2.7)
As t → −∞ E → −2 or E → ∞.

A sketch of a proof is at the end of this problem answer.

Proviso: If an orbit has E → ∞ in the direction of increasing t, then the limit is achieved for a finite value of t.
The other values (E = ±2 and E = −4) require either t → ∞ or t → −∞.
√
Thus, provided that 0 <  < 2/49 ≈ 0.0673 . . ., we have:
8. The two level curves E = −2 are unstable limit cycles enclosing the critical points P1 and P3 .
9. The level curve E = 2 is a stable limit cycle enclosing all the critical points.
10. The critical points P1 and P3 are stable spiral points, while P2 is a saddle point.

Figure 2.1 shows the phase portrait of the system for the choice  = 1/40. The degree of stability of the limit cycles,
and of the critical points P1 and P3 , is controlled by  — as should be obvious from (2.4). As  grows, the limit
√
cycles become more stable (resp. unstable), and at  = 2/49 the critical points P1 and P3 switch from spirals to
nodes. Unless  is fairly small, the orbits move away (towards) the limit cycles so fast that they appear as the spokes
of bicycle tires near them.

It should be obvious that the shape and location of the limit cycles can be changed — by replacing E by some other
function with different level curves. It is possible to get the limit cycles to have almost any desired shape (at the
price of having to manufacture a complicated 2 function E).
2 An explicit expression for E might not even be possible. But the surface z = E(x, y) need not be “geometrically” more complicated than

the one given by (2.1).

18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 7

dx/dt = - E y - f E x and dy/dt = Ex - f E y.

4

1
Figure 2.1: Phase portrait for the system
y 0 in (2.2), with  = 0.025 in (2.6). The two
unstable limit cycles are given by the com-
-1 ponents of the level curve E = −2, and the
single stable limit cycle is given by the level
-2 curve E = 2.

-3

-4
-4 -3 -2 -1 0 1 2 3 4
x

Sketch of the proof of (2.7). Consider an orbit that is not a critical point. Then Ex2 + Ey2 > 0 everywhere on the
orbit — although Ex2 + Ey2 may approach zero if the orbit approaches a critical point as either t → ∞, or t → −∞.
Then (2.4) shows that E is increasing for −2 < E < 2, and decreasing for |E| > 2. Note that E is restricted to the
range −4 ≤ E < ∞.

3 Simple Poincaré Map for a limit cycle #02

3.1 Statement: Simple Poincaré Map for a limit cycle #02
Consider the following autonomous phase plane system
dx ν


= (x2 + y 4 ) ν x − 4 x3 − x2 y − ν x y 2 − 4 y 3 , 

dt 
where ν > 0. (3.1)
dy 2 4 1 3 ν 2 2 3


= (x + y ) ν y + x − x y + xy − ν y , 

dt 4 4

This system has a periodic solution (show this), which can be written in the form

dΦ
x = 2 cos Φ, y = sin Φ, where = 2 (x2 + y 4 ) = 2 (1 + cos2 Φ)2 . (3.2)
dt
This solution produces an orbit going through the point x = 0, y = 1 in the phase plane. The orbit is an ellipse, as
(3.2) shows. 3
Construct (either numerically 4 or analytically) a Poincaré map near this orbit, and use it to show that the orbit
is a stable limit cycle. Define the Poincaré map z → u = P (z) as follows:
3 Note that Φ is a strictly increasing function of time.
4 If you do it numerically, keep ν as a variable and check your answers for several values — say: ν = 0.1, 0.5, 1, 2, 5.
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 8

• For every sufficiently small z, let x = X(t, z) and y = Y (t, z) be the solution of (3.1) defined by X(0, z) =
0 and Y (0, z) = 1 + z.
• For this solution the polar angle θ in the phase plane is an increasing function of time, starting at θ = 12 π
for t = 0. Thus, there is a time t = tz at which the
solution reaches θ = 25 π (note that tz is a function of z). Then take u = Y (tz , z) − 1.
Hint. Because tz is a function of z, unknown a priori, the definition of the Poincaré map above is a bit awkward to
implement. To avoid having to calculate tz for each solution, it is a good idea to use a parameter other than time
to describe the orbits. For example, if the equations are written in terms of a parameter such as the polar angle —
namely dxdθ
= F (x, y) and dy dθ
= G(x, y), then the Poincaré map is easier to describe, as θ varies from θ = 21 π
to θ = 52 π in every one of the orbits needed to compute u = P (z). Note that this is just a “for example”, using the
polar angle is not the best choice. Scale the variables first, so that the limit circle is a circle, not an ellipse. ♣
Small challenge: You should be able to write P analytically. The formula is not even messy.

3.2 Answer: Simple Poincaré Map for a limit cycle #02

Since the questions of interest here have to do with the topology of the orbits in phase space, it does not matter if we
parameterize the orbits with time, or some other parameter, as long as they both yield the same direction of travel
along the orbits (so stability is not affected). Thus, introduce a “new” time along the orbits, as follows:
dx

= ν x − ν4 x3 − x2 y − ν x y 2 − 4 y 3 , 

dτ 2 4 dτ 
= x + y , =⇒ (3.3)
dt dy 1 3 ν 2

2 3 
= ν y + 4 x − 4 x y + xy − ν y . 
dτ
Except for the critical point at the origin, dτ /dt > 0, so that τ is acceptable as an orbit parameter. Now we notice
that we can write the equations in the form
dx

= ν x (1 − 14 x2 − y 2 ) − 4 y ( 41 x2 + y 2 ), 

dτ 
(3.4)
dy 1 2 2 1 2 2

= ν y (1 − 4 x − y ) + x ( 4 x + y ).


dτ
1
The combination x2 + y 2 in these equations suggests that we introduce elliptic-polar coordinates
4
1 y
x = 2 R cos Φ and y = R sin Φ ⇐⇒ R2 = x2 + y 2 and tan Φ = 2 . (3.5)
4 x
dR dΦ
Hence 5 = ν R (1 − R2 ) and = 2 R2 , so that
dτ dτ
dR2
= ν 1 − R2 =⇒ R2 = 1 + ((R0 )2 − 1) exp ν π2 − Φ ,




(3.6)
dΦ
where R0 is the value of R for Φ = π/2. The Poincaré map now follows from this last equation. We want to find
the value R = 1 + u, at Φ = 5 π/2, of the orbit for which R0 = 1 + z. Hence
q
(1 + u)2 = 1 + (z 2 + 2 z) e−2 π ν =⇒ u = P (z) = −1 + 1 + (z 2 + 2 z) e−2 π ν . (3.7)
(
dP Thus the limit cycle
In particular: P (0) = 0 and 0 < (0) = e−2 π ν < 1.
dz in (3.2) is stable.

5 Note that, for R = 1, these equations provide the periodic solution in equation (3.2).
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 9

4 Problem 07.02.x1 (area evolution)

4.1 Statement for problem 07.02.x1
Consider some arbitrary orbit, Γ, for the phase plane system
d~r
= F~ (~r) where ~r = (x, y)T , F~ = (f (x, y), g(x, y))T , (4.1)
dt
and F~ has continuous partial derivatives up to (at least) second order. That is: Γ is the curve in the plane given by
some solution ~r = ~rγ (t) to (4.1). Then
A. Let Ω = Ω(t) be an “infinitesimal” region that is being advected, along Γ, by the flow given by (4.1). For
example:
A1. Let Ω(0) be a disk of “infinitesimal” radius dr, centered at ~rγ (0).
A2. For every point ~rp0 ∈ Ω(0), let ~r = ~rp (t) be the solution to (4.1) defined by the initial data ~rp (0) = ~rp0 .
A3. Then, at any time t∗ , the set Ω(t∗ ) is given by all the points ~rp (t∗ ), where ~rp0 runs over all the points in
Ω(0).

Note that Ω(0) need not be a disk. Any infinitesimal region containing ~rγ (0) will do. All we need is that the
notion of area applies to it — see item B.
B. Let A = A(t) be the area of Ω(t).
Find a differential equation for the time evolution of A. The equation that you will find is trivially extended to
higher dimensions — e.g. to characterize the evolution of the volume in a 3-D phase space.
Hints.
h1. First, introduce the vector δ~r = δ~r(t) = ~rp − ~rγ for every point in Ω(t). This vector characterizes the evolution
of the “shape” of Ω as the set moves along Γ. In order to calculate how A(t) evolves, you only need to know
how the δ~r vectors evolve.
h2. For every vector δ~r, write an equation giving δ~r(t + dt) in terms of δ~r(t) and the partial derivatives of F~
along Γ. Since you are dealing with infinitesimal terms, you can neglect higher order terms, so as to obtain a
relationship from δ~r(t) to δ~r(t + dt) given by a linear transformation. Make sure that this linear transformation
correctly includes the O(dt) terms, which you will need to calculate time derivatives.
h3. From the transformation in item h2 derive a relationship between A(t + dt) and A(t) — use the fact that, for
linear transformations, areas are related by the absolute value of the determinant. You need to calculate the
determinant only up to O(dt).
h4. Use the result in item h3 to calculate the time derivative of A, and obtain the differential equation.

4.2 Answer for problem 07.02.x1

At any time t, we can write

~rγ (t + dt) = ~rγ (t) + F~γ (t) dt, (4.2)

~rp (t + dt) = ~rp (t) + F~p (t) dt, (4.3)

where we have neglected O (dt)2 contributions, F~γ = F (~rγ ), F~p = F (~rp ), and ~rp = ~rp (t) tracks an arbitrary point

in Ω(t) — as in item A2. Hence we can write

 
δ~r (t + dt) = δ~r (t) + F~p (t) − F~γ (t) dt


= I + Mγ (t) dt δ~r (t), (4.4)
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 10

where δ~r = ~rp − ~rγ , I is the identity matrix, M is the matrix of partial derivatives of F~
!
fx fy
M= ,
gx gy

Mγ = M (~rγ ), and we have neglected O (dr)2 dt terms to arrive at the second line in (4.4). It follows that

 
A(t + dt) = det I + Mγ (t) dt A(t)
 


= 1 + Trace Mγ (t) dt A(t), (4.5)

where (again) we have neglected O (dt)2 terms when computing the determinant. From this last equation we obtain

d  
~ A,
A = div F (4.6)
dt
 
where we have used that Trace Mγ (t) = div F~ , with the divergence evaluated along Γ.

A second, more “mathematical”, derivation follows:

~ = R(t,
Define R ~ ~
r ) by
∂ ~  
R = F~ R
~ ~ (0, ~r ) = ~r.
with R (4.7)
∂t
Thus R~ characterizes the whole flow given by the system in (4.1).
 
In particular, ~ ~ (t, ~
rγ (t) = R rγ (0)) and ~ ~ t, ~
rp (t) = R rp0 rp0 ∈ Ω(0).
for every ~

Let now S = S(t, ~ ~ with respect to ~r, with Sγ = S(t, ~

r ) be the matrix of the partial derivatives of R rγ (0)). Then

A(t) = J(t) A(0), where J = det(Sγ ) (4.8)

~ Now we have
is the Jacobian — along Γ — of the transformation from Ω(0) to Ω(t) defined by R.
1. ∂
∂t
S = M S, where M is the matrix of partial derivatives of F~ .
~ in (4.7).
This follows by taking partial derivatives, with respect to ~r, of the equation for R
2. ln(J ) = ln (det(Sγ )) = Trace (ln(Sγ )) . . . . . . . . . . . . . . . . . . . . . . . . . . . Why do we need this? See remark 4.2.
 
dS
3. J1 dJ
dt
= Trace dtγ Sγ−1 = Trace (Mγ ).
This follows by taking the time derivative of the equation in item 2, and then using the result in item 1.

It should be now clear that, from (4.8) and the result in item 3, the same equation (4.6) obtained earlier for the time
evolution of A follows.

Remark 4.2 In the calculation above we need to calculate the time derivative of J, which is defined as the deter-
minant of a matrix. Directly calculating the derivatives of a determinant is, generally, rather messy. On the other
hand, the formula in item 2 transforms this into the calculation of the derivative of the trace of the logarithm of a
matrix. While the derivative of the logarithm of a matrix is, itself, hard to compute (because matrix multiplication
is not commutative), the trace fixes this problem — since Trace(AB) = Trace(BA) for any square matrices A and B
— and one can then take the derivative as if scalars, instead of matrices, were involved.
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 11

5 Problem 07.05.06 - Strogatz (Biased van der Pol)

5.1 Statement for problem 07.05.06
Suppose the van der Pol oscillator is biased by a constant force:

d2 x dx
+ µ (x2 − 1) + x = a, (5.1)
dt2 dt
where a can be positive, negative, or zero. (Assume µ > 0 as usual).
a) Find and classify all the fixed points.
b) Plot the nullclines in the Liénard plane. Show that if they intersect on the middle branch of the cubic nullcline,
the corresponding fixed point is unstable.
c) For µ  1, show that the system has a stable limit cycle if and only if |a| < ac , where ac is to be determined.
(Hint. Use the Liénard plane.)
d) Sketch the phase portrait for a slightly greater than ac . Show that the system is excitable — it has a globally
attracting fixed point, but some (small, but not infinitesimal) disturbances can send the system on a long
excursion through phase space before returning to the fixed point; compare with Exercise 4.5.3.
This system is closely related to the Fitzhugh-Nagumo model of neural activity; for an introduction see Murray,
J. (1989) Mathematical Biology (Springer, New York) or Edelstein-Keshet, L. (1988) Mathematical Models in
Biology (Random House, New York).

5.2 Answer for problem 07.05.06

We begin by writing the Liénard plane form of the equations. The system in (5.1) can be written in the form:
 
d dx 1 3
+ µ( x − x) + x = a.
dt dt 3

Thus, in terms of  
1 dx 1
y= + µ( x3 − x) ,
µ dt 3
we can write the equations in the Liénard plane as:
 
dx 1 3 dy 1
=µ y− x +x and = (a − x). (5.2)
dt 3 dt µ

The equations are invariant under the transformation x → −x, y → −y, and a → −a. Thus:

There is no loss of generality in assuming that a > 0. (5.3)

a) Fixed points.
The system in (5.2) has only one fixed point, namely P = (x0 , y0 ) = (a, 31 a3 − a). The linearized system near
this critical point is:
dX dY 1
= µ(1 − a2 )X + µY and = − X, (5.4)
dt dt µ
with eigenvalues
1 p 
λ= µ(1 − a2 ) ± µ2 (1 − a2 )2 − 4 . (5.5)
2
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 12

Thus:
2
a1) For 1+ < a2 P is a stable node. Figure 5.1-A.
µ
2
a2) For 1 < a2 < 1+ P is a stable spiral. Figure 5.1-B.
µ
2
a3) For 1− < a2 < 1 P is an unstable spiral. Figure 5.2-A.
µ
2
a4) For a2 < 1− P is an unstable node. Figure 5.2-B.
µ

Biased van der Pol: µ = 3.00 and a = 1.60 Biased van der Pol: µ = 1.00 and a = 1.60
Nullclines and flow. Fixed point: stable node Nullclines and flow. Fixed point: stable spiral

1 1

y 0 y 0

-1 -1
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
x x
Figure 5.1: (Problem 07.05.06). Biased van der Pol; equation (5.2). Nullclines and flow field.
A (left). Fixed point: stable node, 1 + µ2 < a2 . B (right). Fixed point: stable spiral, 1 < a2 < 1 + µ2 .

Biased van der Pol: µ = 1.00 and a = 0.50 Biased van der Pol: µ = 3.00 and a = 0.50
Nullclines and flow. Fixed point: unstable spiral Nullclines and flow. Fixed point: unstable node

1 1

y 0 y 0

-1 -1
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
x x
Figure 5.2: (Problem 07.05.06). Biased van der Pol; equation (5.2). Nullclines and flow field.
A (left). Fixed point: unstable spiral, 1 − µ2 < a2 < 1. B (right). Fixed point: unstable node, a2 < 1 − µ2 .

The borderline values a2 = 1 + 2/µ, a2 = 1, and a2 = 1 − 2/µ correspond (in the linearized regime) to a stable
improper node (double eigenvalue with geometric multiplicity one), a center and an unstable improper node,
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 13

respectively. In order to ascertain the detailed behavior of the orbits near the critical point in these cases, a
nonlinear analysis is required. However, the only situation we actually have to worry about is the one that
occurs for a = 1 and the critical point is a linear center — because this is the only case where the nonlinear
terms decide the stability properties of the fixed point. In the other cases the nonlinear terms may turn the
improper node into a weak spiral, but they cannot change the stability properties of the critical point. In what
follows we will do a nonlinear analysis near the center, for the case a = 1. We will use a method that is an
alternative to the two-timing asymptotic techniques introduced in the lectures.
Nonlinear analysis near the linear center.
Assume a = 1, and write x = 1 + X and y = − 32 + 1
µ Y . The equations are then
 
dX 1 dY
= Y − µ X2 + X3 and = −X. (5.6)
dt 3 dt

As expected, at leading order for X 2 + Y 2  1, these are the harmonic oscillator equations. Consider now the
harmonic oscillator energy: E = 12 X 2 + Y 2 . This quantity is conserved by the linear part of the equations

above in (5.6), but the nonlinear terms change the conservation equation Ė = 0 to:
 
dE 3 1 4
= −µ X + X . (5.7)
dt 3

We claim now that it is possible to correct E, by higher order cubic terms in X and Y , in such a way that the
leading order cubic terms on the right hand side in (5.7) are eliminated. To do this we must find a combination
of terms of the form X n Y 3−n (with n = 0, 1, 2, 3), whose time derivative — as given by equation (5.6) — is, to
leading order, equal to µX 3 . Adding then this combination to E will produce the desired result. This turns
out to be not too difficult a task: a few simple manipulations using equation (5.6) shows that the following
four equalities apply:   
d 2
−X 2 Y − Y 3 = X3 

dt  3





d 1 3


2

X = X Y  
dt  3

+ O (X 2 + Y 2 )2 .


 (5.8)
d 1
− Y3 = XY 2 


dt  3 




d 2 2 3 3


XY + X = Y 

dt 3


Therefore any possible combination of cubic terms on the right in (5.7) can be eliminated. Thus consider the
corrected energy  
1 2 2

2 2 3
Ec = X +Y −µ X Y + Y , (5.9)
2 3
which satisfies:
dEc 1 2 2 4
= − µX 4 + 2µ2 X 3 Y + µ X Y. (5.10)
dt 3 3
Consider now an arbitrary solution near the center (say, within a distance 0 <   1 of the center). For any
finite period of time, we know that we can write

X =  cos(t − t0 ) + O(2 ) and Y = − sin(t − t0 ) + O(2 ), (5.11)

where t0 is some constant. Substituting this into the right hand side in (5.10), and integrating from t = 0 to
t = 2π, we can compute the leading order change in Ec as the solution goes once around the center. This yields:
µ
∆Ec = − 4 + O(5 ). (5.12)
8
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 14

Thus E decreases each time the solution goes around the fixed point. Since the surface z = E(x, y) has a local
minimum at the critical point, we conclude that:

For a = 1 the critical point is a nonlinear spiral. (5.13)

Remark 5.3 As we pointed out earlier, the approach used above to arrive at the conclusion in (5.13) is
something that one can always do near a critical point that is a center, no matter what the
nonlinearity. This follows because the equations near a center can always be manipulated into a form such
that the linear part looks as in (5.6). Thus E will satisfy an equation similar to (5.7), with some nonlinearity
on the right hand side. Then (5.8) shows that E can be modified to some Ec (by adding cubic terms to E) so
that the leading order time derivative of Ec is quartic, as in (5.10). Then the argument above in (5.11 – 5.12)
will — generally — give a nonzero value to ∆Ec , meaning that we will know if the point becomes a stable or
an unstable spiral point due to the nonlinear effects.
The approach will fail when the calculation yields ∆Ec = O(5 ) above in (5.12). In this case one must add
further corrections 6 to Ec , so as to compute ∆Ec with higher accuracy. This will require the elimination
of both the fourth and fifth order nonlinearities on the right hand side in (5.10). We further note that this
failure will occur, precisely, under the same circumstances where the alternative two-timing approach will require
computation of the asymptotic approximation to an order higher than the normal cubic.
The advantage of the approach here over the two-timing asymptotic approximation introduced in the lectures is
that it requires quite a bit less computation.. The disadvantage is that it gives less detailed information
— something that, many times is not too important. It is also somewhat less general and in problems involving
many dimensions it is not necessarily as simple to implement as in 2-D. At any rate: it is worth keeping
it in mind whenever one has to elucidate stability issues in situations where linearized theory is
not good enough.
Another “advantage” of this approach, is that it is often easy to recast the argument in a form that is mathe-
matically rigorous — something that is much harder, or impossible, with two-timing calculations. For example,
in our case here, the argument from (5.10) to (5.12) is not rigorous, because it involves the more-or-less hand-
waving step (5.11). However, note that, of the two quartic terms on the right in (5.10), one is a perfect
derivative at leading order. Thus we can write:
 
dE d 1 2 4 1  
= Ec − µ X = − µX 4 + O (X 2 + Y 2 )5/4 .
dt dt 2 3
But this means that, in a small enough neighborhood of the critical point, E is decreasing. This shows that the
point must be a stable spiral — without any need to invoke (5.11).

b) Nullclines.
Figures 5.1 and 5.2 show plots of the nullclines (and the flow field) for the four typical cases (as outlined in
(a1) through (a4) at the beginning of part (a)) that can arise for a > 0.
It is clear that instability of the critical point (figure 5.2) corresponds to the nullclines intersecting on the
middle branch of the cubic nullcline.

c) Case µ  1.
When µ  1, the flow is nearly horizontal 7 to the right, for y > (1/3) x3 − x; and nearly horizontal to the
left, for y < (1/3) x3 − x. It is only in a neighborhood of size O(1/µ) near the nullcline y = (1/3) x3 − x that
significant motion in the vertical direction takes place. In this neighborhood the orbits move down for x > a
and up for x < a. Two distinct situations can then arise:
6 This is possible: the vanishing of ∆Ec means that the O(4 ) terms on the right hand side in (5.10) are exact derivatives.
7 Parallel to the x-axis.
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 15

Biased van der Pol: µ >> 1 and a = 0.15 Biased van der Pol: µ >> 1 and a = 1.60
Phase portrait. Fixed point: unstable node Phase portrait. Fixed point: stable node

1 1

y 0 y 0

-1 -1
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
x x
Figure 5.3: (Problem 07.05.06). Biased van der Pol. Phase portrait for equation (5.2) with µ  1.
A (left). Case with a limit cycle: a2 < 1. B (right). Case without a limit cycle: 1 < a2 .

c1) Limit cycle: a < ac = 1.

A typical phase portrait for this case is shown in figure 5.3-A. Because the dividing point 8 between y
increasing and y decreasing (near the nullcline y = 31 x3 − x) is in the middle branch of the cubic; a
situation arises where all the orbits approach a limit cycle. The argument is exactly the same we used
for the relaxation oscillations in the un-biased van der Pol equation.
Figure 5.3-A shows some typical orbits, the limit cycle, the nullcline y = 13 x3 − x, and the principal curves
for the node at the critical point. Note that the middle branch of the nullcline approximates the principal
curve along which all orbits but two leave the node.
You should think carefully about what is going on with the orbits as they approach the nullcline. They
do not merge with it, but stay so close than in the figure they end up looking as the same curve. In
particular, sketch a blown up picture of the local orbit geometry near the points (x, y) = ±(1, −2/3) (the
local maximum and minimum of the nullcline), and along the middle branch of the nullcline.
c2) Global attracting fixed point a > ac = 1.
A typical phase portrait for this case is shown in figure 5.3-B. Now the dividing point between y increasing
and y decreasing is not in the middle branch of the cubic. It is then clear that this leads to the situation
shown in figure 5.3-B, without a limit cycle, and with all solutions approaching the critical point as
t → ∞.
Again: think carefully about the behavior of the orbits near the nullcline; in particular of the local orbit
geometry near (x, y) = ±(1, −2/3) and the middle branch of the nullcline.
A final note: figure 5.3-B corresponds to the case a < 2. For a > 2 the critical point moves beyond the S
shaped part of the nullcline and some minor (and obvious) changes are needed in the phase portrait.

An interesting question arises now:

What happens near the critical threshold, i.e.: a ≈ ac = 1?

From (a2) and (a3) we see that 9 as a decreases through a = 1, the critical point switches from a stable spiral
to an unstable spiral. On the other hand, (5.13) shows that the nonlinearity is stabilizing near the critical
8 That is: the critical point for the equations.
9 For µ > 0 fixed.
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 16

point. Thus:
A supercritical Hopf bifurcation occurs at a = 1 (µ > 0 fixed).
This fact creates a slight puzzle when viewed in the light of the the result above in (c1) and the phase portrait
√
in figure 5.3-A. Namely: for a slightly less than 1, the limit cycle is supposed to have size O( 1 − a), while
figure 5.3-A indicates a limit cycle of size O(1). The puzzle is resolved by realizing that the situations considered
in (c1) and (c2) correspond to a fixed and µ → ∞. Then:

c3) This explains why the critical point is a node in figure 5.3-A. In the situation considered in (c1) the critical
point is always a node, as follows from (a4) for a < 1 fixed and µ → ∞.
√
c4) The Hopf bifurcation theory assures us that, for a < 1 “close” to 1, the limit cycle will have size O( 1 − a).
But it says nothing about how close is close. In this particular case with the large parameter µ in the
equation, close means very close. Clearly, much closer that O(1/µ), since the Hopf bifurcation theory
assumes a situation where the departure of the critical point from a center is small — if 1 − a = O(1/µ),
(5.5) shows that the departures will be anything but small. Thus, in the regime considered in (c1) and in
figure 5.3-A, the limit cycle need not be small at all.

Biased van der Pol: a = 0.999 limit cycles.

Biased van der Pol: a = 0.990 limit cycles.
Fixed point: unstable spiral Fixed point: unstable spiral
1.2
-0.52
µ = 11.21411...
µ = 3.6506
0.8 µ = 3.650553
-0.56
µ = 3.65053
0.4
y y
-0.6
µ = 3.65
0
µ = 11.2
µ = 3.655
-0.64 µ = 10
-0.4

-0.68 -0.8
0.7 0.9 1.1 1.3 -2 -1 0 1 2
x x
Figure 5.4: (Problem 07.05.06). Biased van der Pol; equation (5.2) for 0 < 1 − a  1 fixed. The limit cycles grow in
size as µ grows. A (left). Case a = 0.999. B (right). Case a = 0.990.

The behavior of the limit cycle for 0 < 1 − a  1 fixed, as µ changes, is illustrated in figure 5.4 — where the
nullclines and limit cycles for various values of µ are plotted. When 1 − a  1/µ the limit cycle is in the
regime where the asymptotic Hopf bifurcation theory applies, but it is very small.10 As µ grows, the limit
cycle also grows in size. The cycle grows at first in such a way that it approximates the right dip in the ẋ = 0
nullcline, with the loop completed by a nearly horizontal jump from the middle unstable branch to the right
stable branch of the nullcline. Once it reaches the maximum size it can achieve in this fashion, it continues
growing by having a jump across to the left (instead of the right) stable branch from the middle branch. This
switch is done when the jumping place to the right reaches the top of the left “mountain”. At first the jump
to the left is small — cutting across the top. But, as µ continues to grow, it travels down. Eventually the limit
cycle ends up enclosing the whole S shaped part of the nullcline. Then a situation like the one described in
figure 5.3 is reached.
10 For µ large it is smaller than the O(1/µ) region near the nullcline — which in the figures it is approximated by a just line.
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 17

The behavior described in the prior paragraph is true in only a very rough sense when (1 − a) is moderately
small. However, when (1 − a) is very small, the limit cycle reaches a finite size only for µ fairly large. Then the
description in the paragraph above is quite accurate. Computation of the limit cycle as it follows the described
behavior is very hard, for two reasons:

A. The jumps across from the middle branch to the side branches of the nullcline are very sensitive to
numerical errors, since the middle branch is extremely unstable for µ  1. Thus, very small errors can
dramatically alter the jumping place — thus the overall result of the calculation.
B. The limit cycle shape is very sensitive to the value of µ (this is, actually, a consequence of A.) This can
be seen in figure 5.3-A, where very small changes in µ cause large changes in the limit cycle size.

It is because of this difficulty that the limit cycles shown in figure 5.3 are for regimes where µ is not very large.
Thus the jumps across from the middle to the side branches of the nullcline are not very sharp. Had we been
able to compute the evolution (as µ grows) of the limit cycle for some real small value of (1 − a), the departures
of the limit cycles from curves made up from parts of the nullcline and horizontal segments would have been
indistinguishable. This is one of those examples where theory and analytical reasoning are much better that a
numerical calculation.

d) Excitable system: 0 < a − 1  1.

Consider the situation depicted in figure 5.3-B. If the fixed point is perturbed in such a way that the solution
is sent to a point on the phase plane with y < −2/3, then the solution will undergo a whole swing around the
nullcline,11 before returning to the critical point. If a − 1 is small, then the perturbation needed to do this will
be small. Since the critical point is a global attractor in this case, it follows that the system is excitable.
In problem 4.5.3 the threshold of excitability was given by the presence of a second critical point. Here there
is no other critical point and the role of the second critical point is taken by the the point (x, y) = (1, −2/3)
— notice that, in the limit µ → ∞ this point behaves very much like a critical point of some strange sort, as
can be seen in figure 5.3-B.

6 Problem 08.02.05 - Strogatz (Hopf bifurcation using a computer)

6.1 Statement for problem 08.02.05
For the following system
dx dy
= y + µx and = −x + µ y − x2 y, (6.1)
dt dt
a Hopf bifurcation occurs at the origin when µ = 0. Using a computer, plot the phase portrait and determine whether
the bifurcation is subcritical or supercritical. For small values of µ, verify that the limit cycle is nearly circular. Then
measure the period and radius of the limit cycle, and show that the radius R scales with µ as predicted by theory.

6.2 Answer for problem 08.02.05

See figure 6.1 for the phase portrait with µ = −0.1. It indicates a stabilizing nonlinearity, thus a supercritical (soft)
Hopf bifurcation.
Figure 6.2 shows a picture of the phase portrait for µ = 0.1 on the left and the limit cycles for various values of
0 < µ  1 on the right. A stable limit cycle appears around the critical point for 0 < µ  1. This confirms that a
11 Going first to the left branch of the nullcline, then up to the local maximum, etc.
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 18

2
dx/dt=µx+y & dy/dt=-x+ µy-x y, for: µ = -0.1
2

1 Problem 08.02.05.
System (6.1) for µ = −0.1 < 0.
All the orbits spiral towards the origin, even those O(1)
y0
away. Since µ is fairly small, this is a good hint that the
nonlinearity is stabilizing, which should lead to a super-
critical (soft) Hopf bifurcation.
-1

-2
-1 0 1 2
x
Figure 6.1: (Problem 08.02.05). Phase portrait for the system in (6.1) when µ = −0.1 < 0

2
dx/dt=µx+y & dy/dt=-x+ µy-x y, for: µ = 0.1 Limit cycles for: xt=µx+y & yt=-x+ µy-x 2y.
2 0.4
µ = 0.016
µ = 0.01

1 0.2
µ = 0.0025

y0 y 0

-1 -0.2

-2 -0.4
-1 0 1 2 -0.2 0 0.2 0.4
x x

Figure 6.2: (Problem 08.02.05). Left: phase portrait for the system in (6.1) when µ = 0.1 > 0. The picture on the
right shows the limit cycles for various values of 0 < µ  1.

supercritical (soft) Hopf bifurcation occurs.

For 0 < µ  1 it is easy to see (in figure 6.2) that the Limit Cycle Parameters, 0 < µ  1.
limit cycles are nearly circular. The table on the right R Period
shows a listing of various parameters for these cycles. µ R =radius. √ .
µ π
It should be clear that the theoretical predictions (e.g.:
R 0.0160 0.35775 2.8283 2.0014
√ ∼ constant, and period ∼ linear period) are satis- 0.0100 0.28285 2.8285 2.0005
µ
fied. 0.0025 0.14142 2.8284 2.0000
Furthermore, the limit cycle for µ = 0.1 is not very circular, but if we interpret its radius as the value of x when y = 0,
R
we obtain R ≈ 0.89235, which yields √ = 2.8219 (quite close to the values in the table). In this case the period is
µ
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 19

P ≈ 2.0579π.

7 Problem 08.02.07 - Strogatz

(Hopf and homoclinic bifurcations using a computer)
7.1 Statement for problem 08.02.07
For the following system
dx dy
= µ x + y − x2 and = −x + µ y + 2x2 , (7.1)
dt dt
a Hopf bifurcation occurs at the origin when µ = 0. Using a computer, plot the phase portrait and determine
whether the bifurcation is subcritical or supercritical. For small values of µ, verify that the limit cycle is nearly
circular. Then measure the period and radius of the limit cycle, and show that the radius R scales with µ as
predicted by theory.
In addition to a Hopf bifurcation, this system also exhibits an homoclinic bifurcation of the limit cycle. FIND IT.

7.2 Answer for problem 08.02.07

See figure 7.1 for the phase portrait with µ = −0.04. It indicates a stabilizing nonlinearity, thus a supercritical (soft)
Hopf bifurcation.

2 2
x =µx+y-x & y = -x + µ y + 2x ,
t t
for µ = -0.04

0.45

0.3
Problem 08.02.07.
0.15 System (7.1) for µ = −0.04 < 0, when the critical point
y is a stable spiral.
0 All the orbits spiral towards the origin, even those O(1) away.
Since µ is fairly small, this is a good hint that the nonlinearity
-0.15 is stabilizing, which should lead to a super-critical (soft)
Hopf bifurcation. See remark 7.4.
-0.3

-0.3 -0.15 0 0.15 0.3 0.45

x
Figure 7.1: (Problem 08.02.07). Phase portrait for the system in (7.1) when µ = −0.04 < 0.

Figure 7.2 shows a picture of the phase portrait for µ = 0.04 on the left and the (stable) limit cycles for various small
positive values of 0 < µ  1 on the right. A stable limit cycle appears around the critical point for 0 < µ  1. Thus
a super-critical (soft) Hopf bifurcation occurs.
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 20

xt = µ x + y - x 2 & yt = -x + µ y + 2x2, xt = µ x + y - x 2 & yt = -x + µ y + 2x2,

for µ = 0.04 Limit cycles for: 1000µ = 10, 5 & 3.

0.45 0.15

0.3 0.1

0.05
0.15
y y 0
0
-0.05
-0.15
-0.1
-0.3
-0.15
-0.3 -0.15 0 0.15 0.3 0.45 -0.15 -0.1 -0.05 0 0.05 0.1 0.15

x x
Figure 7.2: (Problem 08.02.07). Left: phase portrait for the system in (7.1) for µ = 0.04 < 0. Right: the limit cycles
for three values of 0 < µ  1 — larger values of µ correspond to larger limit cycles.

For 0 < µ  1 it is easy to see (in figure 7.2) that the Limit Cycle Parameters, 0 < µ  1.
limit cycles are nearly circular. The table on the right R Period
shows a listing of various parameters for these cycles. µ R =radius. √ .
µ 2π
It should be clear that the theoretical predictions (e.g.:
R 0.010 0.1259 1.2590 1.0396
√ ∼ constant, and period ∼ linear period) are satis- 0.005 0.0926 1.3096 1.0318
µ
fied. 0.003 0.0731 1.3346 1.0113

Furthermore, for this system an homoclinic bifurcation of the limit cycle also happens, as follows:
2
As µ grows from µ = 0, the limit cycle grows in size, till it eventually reaches the critical point at x = 1+µ
2+µ and
y = x2 − µx (which is a saddle). At this point an homoclinic bifurcation of the limit cycle happens — see figure 7.3
— with the period of the limit cycle diverging to ∞, and the limit cycle becoming an homoclinic connection for the
saddle.

Remark 7.4 The distinction between a sub and super critical Hopf bifurcation arises solely from the stability of
the involved limit cycle. If the limit cycle is stable, then the bifurcation is super-critical (or soft). If the limit cycle
is unstable, then the bifurcation is sub-critical (or hard). In turn, this is determined by the role of the nonlinearity
at the bifurcation point (where the linear terms are neutrally stable). If the nonlinear terms are stabilizing, then
the bifurcation is super-critical; if the nonlinear terms are destabilizing, then the bifurcation is sub-critical; if the
nonlinear terms are neutrally stable, then the situation is singular and the Hopf bifurcation scenario does not apply.
In the particular case of the system in (7.1), the fact that the limit cycle appears for µ > 0 is completely irrelevant to
the classification of the type of Hopf bifurcation that occurs. What matters is that the limit cycle that is born is stable.
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 21

xt = µ x + y - x 2 & yt = -x + µ y + 2x2,
for µ = 0.065
0.6

0.45

0.3 Problem 08.02.07.

Limit cycle for the system in (7.1) for µ = 0.065, when
0.15 the limit cycle has grown and it is about to become an
y 2
homoclinic orbit for the saddle at x = 1+µ
2+µ
, y = x2 −
0
µx. The limit cycle period is about 52 for this value of µ.
-0.15 Note that an homoclinic bifurcation of this limit cycle is
about to happen.
-0.3

-0.3 -0.15 0 0.15 0.3 0.45 0.6

x
Figure 7.3: (Problem 08.02.07). Limit cycle for the system in (7.1) when µ = 0.65.

8 Problem 08.04.03 - Strogatz (Homoclinic bifurcation via computer)

8.1 Statement for problem 08.04.03
Using numerical integration, find the value of µ at which the system
dx dy
= µ x + y − x2 and = −x + µ y + 2 x2 , (8.1)
dt dt
undergoes a homoclinic bifurcation. Sketch the phase portrait just above and below the bifurcation. In fact:
1. Find and classify all the critical points for all values of µ.
2. For µ = 0 the origin is a center for the linearized equations. What happens for the nonlinear equations? Are the
nonlinear terms stabilizing or destabilizing? What sort of critical point is the origin for the full equations: stable
spiral, unstable spiral, or center? You should be able to do this analytically — See hint 8.1.
3. What happens at µ crosses 0? (Justify your answer). The result in item 2 should help here!
4. Increase µ from µ = 0, and find the homoclinic bifurcation (this is where you’ll need a computer).
5. Optional: Compute the period of the limit cycle as the homoclinic bifurcation is approached, and verify the
theoretical prediction: period ∼ − log |µ − µc |.

Remark 8.5 This problem is very similar (same system of equations) to Strogatz problem 8.2.7. However: 8.2.7 is
purely computational, while here you are being asked to do the analysis behind the problem.

Hint 8.1 To do the analysis in item 2, you have two alternatives:

A. Do a “two-times expansion” for orbits near the critical point. Namely: write the equations in terms of x =  X
and y =  Y (where 0 <   1). Then expand.
B. Find a “local Liapunov function”, E = (x2 + y 2 )+ higher order terms, such that dE
dt
< 0 near the origin. In
dE dE
fact dt ≤ 0 is O.K., as long as dt = 0 only for curves the orbits cross — e.g. the axis.
The first alternative is a straightforward application of the methods in the “Weakly Nonlinear Things” notes. The
second actually provides a rigorous proof of the result. However, it turns out that getting E is not completely
trivial! The naive approach to searching for E is
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 22

0. Define E0 = x2 + y 2 and compute its time derivative. This yields

dE0
= (3rd-order terms) + (4th-order terms).
dt
Of course, this is not good enough: the 3rd-order terms can have any sign. Hence:
1. Add 3rd-order term “corrections” to E0 , to eliminate the 3rd-order terms in E˙0 . That is, define E1 = E0 + 3rd-order
terms, so that
dE1
= (4th-order terms) + (5th-order terms).
dt
There is only one way to do this. Unfortunately, some of the 4th-order terms are positive. Hence:
2. Add 4th-order terms “corrections” to E1 , to eliminate the bad 4th-order terms in E˙1 . That is, define E2 = E1 + 4th-order
terms, so that
dE2
= (negative 4th-order terms) + (5th-order terms) + (6th-order terms).
dt
Again: there is only one way to do this. Unfortunately, this still does not work. Some of the higher order terms here
are always smaller than the negative 4-th order terms, but some are not. For example, if −x2 y 2 is a negative 4-th order
term, then: (i) −x2 y 2 + x3 y 2 is always negative for x2 + y 2  1, so x3 y 2 is not a problem, but (ii) −x2 y 2 + x4 y can
switch sign (if 0 < y < x2  1), so x4 y is a “bad” term. Hence:
3. Add 5th-order terms “corrections” to E2 , to eliminate the bad 5th order terms ... Unfortunately, you then end up with
“bad” 6th order terms!
This never ends! Fortunately: if you do the process above correctly, you will notice that: while the terms in En
involve ever higher powers of y, there is only a very small set of powers of x that appear. Hence, look for a Liapunov
function of the form E = g(y) + x2 f (y) + . . ., where g, f , etc., are to be determined. This will work: there is a
finite (and small) numbers of terms involved. After you have obtained E in this fashion, you will see that it can be
expanded as in item B above.

8.2 Answer for problem 08.04.03

Before we start, let us find the critical points for the system in (8.1). Multiplying the first equation by µ, and
subtracting from the result to the second equation, yields 0 = (1 + µ2 ) x − (2 + µ) x2 at the critical points. From
this, and y = −µ x + x2 (also valid at the critical points), we obtain:
• P0 = (x, y) = (0, 0) is a stable spiral for µ < 0, and an unstable spiral for µ > 0. It can be shown (see
§ 8.2.1) that a supercritical (soft) bifurcation occurs for µ = 0, and a stable limit cycle appears enclosing
the origin for µ > 0.
2
 
• Ps = (x, y) = 1+µ 2+µ
1, 1−2 µ
2+µ
is a critical point for µ 6= −2. Always a saddle.
The homoclinic bifurcation occurs as the limit cycle created by the Hopf bifurcation at µ = 0 grows in size with µ,
till at a critical µ = µc ≈ 0.0661 it collides with the saddle at Ps , and it is destroyed. This process is illustrated
by figures 8.1 and 8.2.
On the left in figure 8.1, for µ = 0.055, we see how the limit cycle has grown from its original position close to the
origin, and it is now large enough to get very close to the saddle at Ps . Notice how the stable manifold for the
saddle approaches the limit cycle, while the unstable one “hugs” it closely (going backwards in time) and then leaves
towards infinity.
On the right in figure 8.1, for µ = 0.066, only the limit cycle is shown. This µ is very close to the critical value. The
limit cycle is very close to the saddle, and is very hard to distinguish it from an homoclinic connection. Precisely
at µ = µc the limit cycle becomes an homoclinic connection, with the period going to infinity. As µ → µc , the limit
cycle period behaves like | log(µc − µ)| — see the table below.
Finally, figure 8.2 shows the phase portrait for µ = 0.077, above critical. The limit cycle has disappeared, the left
un-stable manifold from the saddle approaches the spiral point as t → −∞, while the stable one approaches infinity
as t → ∞.
 2 2 2 2
x =µx+y-x & yt = -x + µ y + 2x , xt = µ x + y - x & yt = -x + µ y + 2x ,
18.385 MIT,t (Rosales) Simple Poincaré Map for a limit cycle #02. 23
for µ = 0.055. for µ = 0.066.

0.6 0.6

0.4 0.4

0.2 0.2
y y
0 0

-0.2 -0.2

-0.4 -0.4
-0.4 -0.2 0 0.2 0.4 0.6 -0.4 -0.2 0 0.2 0.4 0.6
x x
Figure 8.1: (Problem 8.4.3).
2 Phase portraits for the 2system (8.1), ẋ = µ x + y − x2 & ẏ = −x + µ y + 2 x2 . Left:
xt = µ x + y - x & yt = -x + µ y + 2x ,
µ = 0.055, slightly below µc ≈ 0.0661. Right: limit cycle for µ = 0.066, below but very close to µc .
for µ = 0.077.

0.6

0.4
Phase plane portrait for the system in (8.1), that
is ẋ = µ x + y − x2 and
0.2
y ẏ = −x + µ y + 2 x2 ,
with µ = 0.077. This value is slightly above
0
the value µc ≈ 0.061, at which an homoclinic
bifurcation of a limit cycle occurs.
-0.2
Horizontal axis = x. Vertical axis = y.

-0.4
-0.4 -0.2 0 0.2 0.4 0.6
x
Figure 8.2: (Problem 8.4.3). Phase portrait for (8.1) with µ = 0.077, slightly above µc ≈ 0.0661.

In general, homoclinic bifurcations are hard to find by means other than numerical integration of the equations — this
example is no exception to this rule. Perhaps there is an argument one can make to find (and approximate) the
critical value at which the bifurcation occurs, but I was unable to find one. I was not even able to produce an
argument indicating that a bifurcation should occur, other than the following (very gross) one: Once the Hopf
√
bifurcation occurs, the size of the limit cycle grows like µ. Since the saddle starts fairly close to the origin, even
√
rather small values of µ give values of µ that make the limit cycle large enough to “reach” the saddle. Hence, it
√
is not unreasonable to expect a bifurcation to occur for some small µ. This is what happens, with µc ≈ 0.26 and
Ps (µc ) at a distance ≈ 0.53 from P0 .
18.385 MIT, (Rosales) Simple Poincaré Map for a limit cycle #02. 24

For 0 < |µ − µc |  1 the period of a limit cycle about to disappear (due µ Period α
to an homoclinic bifurcation) behaves like − log(|µ − µc |). The table on 0.055 8.969 1.993
Period
the right illustrates this, with α = − log(µ c −µ)
= constant at leading 0.057 9.265 1.971
order. 0.059 9.636 1.948
Note that the calculation of µc is not very reliable, so getting very close 0.061 10.130 1.919
to it is not quite possible. Nevertheless, the table shows reasonable agree- 0.063 10.877 1.883
ment with the theoretical expectation. 0.065 12.458 1.829

8.2.1 Analysis for the Hopf bifurcation

The origin switches from a stable to an unstable spiral point as µ crosses µ = 0. Hence, in order to show that a
supercritical Hopf bifurcation occurs for µ = 0, all we need to do is to show that the origin is (nonlinear) stable spiral
for µ = 0. Namely, for the system
dx dy
= y − x2 and = −x + 2 x2 . (8.2)
dt dt
Following the hint (I will not display here the calculations described in the hint that motivate this form), we search
for a Liapunov function of the form
E = g(y) + x2 f (y) + x3 h(y). (8.3)
It is then easy to check that
d
(g) = −x g 0 + 2 x2 g 0 , (8.4)
dt
d 2

x f = 2 x y f − 2 x3 f − x3 f 0 + 2 x4 f 0 , (8.5)
dt
and
d 3
x h(y) = 3 x2 y h − 3 x4 h − x4 (1 − 2 x) h0 .


(8.6)
dt
The terms linear in x cancel if g = 2 y f , and the cubic ones cancel if f 0 = −2 f . Thus
0

1
f = e−2 y > 0 and g = 1 − e−2 y − y e−2 y


(8.7)
2
yield
dE
= 4 x2 y f − 4 x4 f + 3 x2 y h − 3 x4 h − x4 (1 − 2 x) h0 . (8.8)
dt
4
Now take h = − f , to obtain
3
dE 8
=− x4 (1 − 2 x) f . (8.9)
dt 3
This is precisely what we were looking for. For x small, Ė < 0, except along the coordinate line x = 0 where Ė = 0.
Since non-trivial orbits cross x = 0 (i.e.: ẋ 6= 0 for x = 0 and y 6= 0) E is a decreasing function along them. Further:

E = x2 + y 2 + O((x2 + y 2 )3/2 ), (8.10)

so that the origin is a local minimum for E. In addition, it is easy to see that θ̇ = −1 + O((x2 + y 2 )1/2 ), where θ is
the polar angle. Hence the origin is a stable spiral point.

THE END.