Topic 11 Notes: Jeremy Orloff
Topic 11 Notes: Jeremy Orloff
Jeremy Orloff
11 Argument Principle
11.1 Introduction
The argument principle (or principle of the argument) is a consequence of the residue
theorem. It connects the winding number of a curve with the number of zeros and poles
inside the curve. This is useful for applications (mathematical and otherwise) where we
want to know the location of zeros and poles.
Setup.
γ a simple closed curve, oriented in a counterclockwise direction.
f (z) analytic on and inside γ, except for (possibly) some finite poles inside (not on) γ and
some zeros inside (not on) γ.
Let p1 , . . . , pm be the poles of f inside γ.
Let z1 , . . . , zn be the zeros of f inside γ.
Write mult(zk ) = the multiplicity of the zero at zk . Likewise write mult(pk ) = the order of
the pole at pk .
We start with a theorem that will lead to the argument principle.
Theorem 11.1. With the above setup
Z 0
f (z) X X
dz = 2πi mult(zk ) − mult(pk ) .
γ f (z)
Proof. To prove this theorem we need to understand the poles and residues of f 0 (z)/f (z).
With this in mind, suppose f (z) has a zero of order m at z0 . The Taylor series for f (z)
near z0 is
f (z) = (z − z0 )m g(z)
where g(z) is analytic and never 0 on a small neighborhood of z0 . This implies
f 0 (z) m(z − z0 )m−1 g(z) + (z − z0 )m g 0 (z)
=
f (z) (z − z0 )m g(z)
m g 0 (z)
= +
z − z0 g(z)
Since g(z) is never 0, g 0 (z)/g(z) is analytic near z0 . This implies that z0 is a simple pole of
f 0 (z)/f (z) and 0
f (z)
Res , z0 = m = mult(z0 ).
f (z)
Likewise, if z0 is a pole of order m then the Laurent series for f (z) near z0 is
f (z) = (z − z0 )−m g(z)
1
11 ARGUMENT PRINCIPLE 2
Definition. We write Zf,γ for the sum of multiplicities of the zeros of f inside γ. Likewise
for Pf,γ . So the Theorem 11.1 says,
Z 0
f
dz = 2πi (Zf,γ − Pf,γ ) . (1)
γ f
Definition. Winding number. We have an intuition for what this means. We define it
formally via Cauchy’s formula. If γ is a closed curve then its winding number (or index)
about z0 is defined as
1 1
Z
Ind(γ, z0 ) = dz.
2πi γ z − z0
(In class I’ll draw some pictures. You should draw a few now.)
One of the key notions in this topic is mapping one curve to another. That is, if z = γ(t) is
a curve and w = f (z) is a function, then w = f ◦ γ(t) = f (γ(t)) is another curve. We say
f maps γ to f ◦ γ. We have done this frequently in the past, but it is important enough
to us now, so that we will stop here and give a few examples. This is a key concept in the
argument principle and you should make sure you are very comfortable with it.
Example 11.2. Let γ(t) = eit with 0 ≤ t ≤ 2π (the unit circle). Let f (z) = z 2 . Describe
the curve f ◦ γ.
Solution: Clearly f ◦ γ(t) = e2it traverses the unit circle twice as t goes from 0 to 2π.
Example 11.3. Let γ(t) = it with −∞ < t < ∞ (the y-axis). Let f (z) = 1/(z + 1).
Describe the curve f ◦ γ(t).
Solution: f (z) is a fractional linear transformation and maps the line given by γ to the
circle through the origin centered at 1/2. By checking at a few points:
1 1+i 1 1−i
f (−i) = = , f (0) = 1, f (i) = = , f (∞) = 0.
−i + 1 2 i+1 2
11 ARGUMENT PRINCIPLE 3
−i = γ(−1)
f (i)
You will also see this called the principle of the argument.
Theorem 11.4. Argument principle. For f and γ with the same setup as above
Z 0
f (z)
dz = 2πi Ind(f ◦ γ, 0) = 2πi(Zf,γ − Pf,γ ) (2)
γ f (z)
So we need to show is that the integral also equals the winding number given. This is
simply the change of variables w = f (z). With this change of variables the countour
z = γ(t) becomes w = f ◦ γ(t) and dw = f 0 (z) dz so
Z 0
f (z) dw
Z
dz = = 2πi Ind(f ◦ γ, 0)
γ f (z) f ◦γ w
The last equality in the above equation comes from the definition of winding number.
Note that by assumption γ does not go through any zeros of f , so w = f (γ(t)) is never zero
and 1/w in the integral is not a problem.
Here is an easy corollary to the argument principle that will be useful to us later.
Corollary. Assuming that f ◦ γ does not go through −1, i.e. there are no zeros of 1 + f (z)
on γ then
f0
Z
= 2πi Ind(f ◦ γ, −1) = 2πi(Z1+f,γ − Pf,γ ). (3)
γ f +1
Proof. Applying the argument principle in Equation 2 to the function 1 + f (z), we get
(1 + f )0 (z)
Z
dz = 2πi Ind(1 + f ◦ γ, 0) = 2πi(Z1+f,γ − P1+f,γ )
γ 1 + f (z)
11 ARGUMENT PRINCIPLE 4
Now, we can compare each of the terms in this equation to those in Equation 3:
(1 + f )0 (z) f 0 (z)
Z Z
dz = dz (because (1 + f )0 = f 0 )
γ 1 + f (z) γ 1 + f (z)
Ind(1 + f ◦ γ, 0) = Ind(f ◦ γ, −1) (1 + f winds around 0 ⇔ f winds around -1)
Z1+f,γ = Z1+f,γ (same in both equations)
P1+f,γ = Pf,γ (poles of f = poles of 1 + f )
Example 11.5. Let f (z) = z 2 + z Find the winding number of f ◦ γ around 0 for each of
the following curves.
1. γ1 = circle of radius 2.
2. γ2 = circle of radius 1/2.
3. γ3 = circle of radius 1.
answers. f (z) has zeros at 0, −1. It has no poles.
So, f has no poles and two zeros inside γ1 . The argument principle says Ind(f ◦ γ1 , 0) =
Zf,γ1 − Pf,γ = 2
Likewise f has no poles and one zero inside γ2 , so Ind(f ◦ γ2 , 0) = 1 − 0 = 1
For γ3 a zero of f is on the curve, i.e. f (−1) = 0, so the argument principle doesn’t apply.
The image of γ3 is shown in the figure below – it goes through 0.
Im(w)
Im(z)
γ1
w = f (z) = z 2 + z
γ3
γ2
Re(z) Re(w)
1 1 2
2
Then
Ind(f ◦ γ, 0) = Ind((f + h) ◦ γ, 0).
That is,
Zf,γ − Pf,γ = Zf +h,γ − Pf +h,γ (4)
Proof. In class we gave a heuristic proof involving a person walking a dog around f ◦ γ on
a leash of length h ◦ γ. Here is the analytic proof.
The argument principle requires the function to have no zeros or poles on γ. So we first
show that this is true of f , f + h, (f + h)/f . The argument is goes as follows.
Zeros: The fact that 0 ≤ |h| < |f | on γ implies f has no zeros on γ. It also implies f + h
has no zeros on γ, since the value of h is never big enough to cancel that of f . Since f and
f + h have no zeros, neither does (f + h)/f .
Poles: By assumption f and h have no poles on γ, so f + h has no poles there. Since f has
no zeros on γ, (f + h)/f has no poles there.
Now we can apply the argument principle to f and f + h
Z 0
1 f
dz = Ind(f ◦ γ, 0) = Zf,γ − Pf,γ . (5)
2πi γ f
1 (f + h)0
Z
dz = Ind((f + h) ◦ γ, 0) = Zf +h,γ − Pf +h,γ . (6)
2πi γ f + h
h h
Next, by assumption < 1, so
◦ γ is inside the unit circle. . This means that
f f
h f +h
1+ = maps γ to the inside of the unit disk centered at 1. (You should draw a
f f
figure for this.) This implies that
f +h
Ind ◦ γ, 0 = 0.
f
Z 0
f +h g
Let g = . The above says Ind(g ◦ γ, 0) = 0. So, dz = 0. (We showed above that
f γ g
g has no zeros or poles on γ.)
g0 (f + h)0 f 0
Now, it’s easy to compute that = − . So, using
g f +h f
Z 0
(f + h)0
Z 0
g f
Z
Ind(g ◦ γ, 0) = dz = dz − dz = 0 ⇒ Ind((f + h) ◦ γ, 0) = Ind(f ◦ γ, 0).
γ g γ f +h γ f
Now equations 5 and 6 tell us Zf,γ − Pf,γ = Zf +h,γ − Pf +h,γ , i.e. we have proved Rouchés
theorem.
Corollary. Under the same hypotheses, If h and f are analytic (no poles) then
Zf,γ = Zf +h,γ .
Proof. Since the functions are analytic Pf,γ and Pf +h,γ are both 0. So Equation 4 shows
Zf = Zf +h . QED.
11 ARGUMENT PRINCIPLE 6
CR iR
C1
Re(z)
−iR
To apply the corollary to Rouchés theorem we need to check that (for R large) |h| < |f | on
C1 + CR . On C1 , z = iy, so
So |h| < |f | on C1 .
On CR , z = x + iy with x < 0 and |z| = R. So,
|f (z)| > R − 3 for R large, |h(z)| = 2|ex+iy | = 2ex < 2 (since x < 0).
So |h| < |f | on CR .
The only zero of f is at z = −3, which lies inside the contour.
Therefore, by the Corollary to Rouchés theorem, f + h has the same number of roots as f
inside the contour, that is 1. Now let R go to infinity and we see that f + h has only one
root in the entire half-plane.
Theorem. Fundamental theorem of algebra.
Rouchés theorem can be used to prove the fundamental theorem of algebra as follows.
Proof. Let
P (z) = z n + an−1 z n−1 + . . . + a0
be an nth order polynomial. Let f (z) = z n and h = P − f . Choose an R such that
R > max(1, n|an−1 |, . . . , n|a0 |). Then on |z| = R we have
R n−1 R n−2 R
|h| ≤ |an−1 |Rn−1 + |an−2 |Rn−2 + . . . + |a0 | ≤ R + R + . . . + < Rn .
n n n
11 ARGUMENT PRINCIPLE 7
On |z| = R we have |f (z)| = Rn , so we have shown |h| < |f | on the curve. Thus, the
corollary to Rouchés theorem says f + h and f have the same number of zeros inside
|z| = R. Since we know f has exactly n zeros inside the curve the same is true for the
polynomial f + h. Now let R go to infinity, we’ve shown that f + h has exactly n zeros in
the entire plane.
Note. The proof gives a simple bound on the size of the zeros: they are all have magnitude
less than or equal to max(1, n|an−1 |, . . . , n|a0 |).
The Nyquist criterion is a graphical technique for telling whether an unstable linear time
invariant system can be stabilized using a negative feedback loop. We will look a little more
closely at such systems when we study the Laplace transform in the next topic. For this
topic we will content ourselves with a statement of the problem with only the tiniest bit of
physical context.
Note. You have already encountered linear time invariant systems in 18.03 (or its equiva-
lent) when you solved constant coefficient linear differential equations.
A linear time invariant system has a system function which is a function of a complex
variable. Typically, the complex variable is denoted by s and a capital letter is used for the
system function.
Let G(s) be such a system function. We will make a standard assumption that G(s)
is meromorphic with a finite number of (finite) poles. This assumption holds in many
interesting cases. For example, quite often G(s) is a rational function Q(s)/P (s) (Q and P
are polynomials).
We will be concerned with the stability of the system.
Definition. The system with system function G(s) is called stable if all the poles of G are
in the left half-plane. That is, if all the poles of G have negative real part.
The system is called unstable if any poles are in the right half-plane, i.e. have positive real
part.
For the edge case where no poles have positive real part, but some are pure imaginary we
will call the system marginally stable. This case can be analyzed using our techniques. For
our purposes it would require and an indented contour along the imaginary axis. If we have
time we will do the analysis.
s
Example 11.9. Is the system with system function G(s) = stable?
(s + 2)(s2 + 4s + 5)
Solution: The poles are −2, −2 ± i. Since they are all in the left half-plane, the system is
stable.
s
Example 11.10. Is the system with system function G(s) = 2 stable?
(s − 4)(s2 + 4s + 5)
Solution: The poles are ±2, −2 ± i. Since one pole is in the right half-plane, the system is
11 ARGUMENT PRINCIPLE 8
unstable.
s
Example 11.11. Is the system with system function G(s) = stable?
(s + 2)(s2 + 4)
Solution: The poles are −2, ±2i. There are no poles in the right half-plane. Since there
are poles on the imaginary axis, the system is marginally stable.
Terminology. So far, we have been careful to say ‘the system with system function G(s)’.
From now on we will allow ourselves to be a little more casual and say ‘the system G(s)’.
It is perfectly clear and rolls off the tongue a little easier!
We can visualize G(s) using a pole-zero diagram. This is a diagram in the s-plane where
we put a small cross at each pole and a small circle at each zero.
Example 11.12. Give zero-pole diagrams for each of the systems
s s s
G1 (s) = , G2 (s) = , G3 (s) =
(s + 2)(s2+ 4s + 5) (s2 − 4)(s2 + 4s + 5) (s + 2)(s2 + 4)
Solution: These are the same systems as in the examples just above. We first note that
they all have a single zero at the origin. So we put a circle at the origin and a cross at each
pole.
Im(s) Im(s) Im(s)
G1 (s) G2 (s) G3 (s) x
x i x i i
x Re(s) x x Re(s) x Re(s)
x 1 x 1 1
x
This is just to give you a little physical orientation. Given our definition of stability above,
we could, in principle, discuss stability without the slightest idea what it means for physical
systems.
The poles of G(s) correspond to what are called modes of the system. A simple pole at s1
corresponds to a mode y1 (t) = es1 t . The system is stable if the modes all decay to 0, i.e. if
the poles are all in the left half-plane.
Physically the modes tell us the behavior of the system when the input signal is 0, but there
are initial conditions. A pole with positive real part would correspond to a mode that goes
to infinity as t grows. It is certainly reasonable to call a system that does this in response
to a zero signal (often called ‘no input’) unstable.
11 ARGUMENT PRINCIPLE 9
To connect this to 18.03: if the system is modeled by a differential equation, the modes
correspond to the homogeneous solutions y(t) = est , where s is a root of the characteristic
equation. In 18.03 we called the system stable if every homogeneous solution decayed to 0.
That is, if the unforced system always settled down to equilibrium.
If the system with system function G(s) is unstable it can sometimes be stabilized by what
is called a negative feedback loop. The new system is called a closed loop system. Its system
function is given by Black’s formula
G(s)
GCL (s) = , (7)
1 + kG(s)
where k is called the feedback factor. We will just accept this formula. Any class or book
on control theory will derive it for you.
In this context G(s) is called the open loop system function.
Since GCL is a system function, we can ask if the system is stable.
Theorem. The poles of the closed loop system function GCL (s) given in Equation 7 are
the zeros of 1 + kG(s).
Proof. Looking at Equation 7, there are two possible sources of poles for GCL .
1. The zeros of the denominator 1 + kG. The theorem recognizes these.
2. The poles of G. Since G is in both the numerator and denominator of GCL it should be
clear that the poles cancel. We can show this formally using Laurent series. If G has a pole
of order n at s0 then
1
bn + bn−1 (s − s0 ) + . . . a0 (s − s0 )n + a1 (s − s0 )n+1 + . . . ,
G(s) = n
(s − s0 )
where bn 6= 0. So,
1 n
(s−s0 )n (bn + bn−1 (s − s0 ) + . . . a0 (s − s0 ) + . . .)
GCL (s) =
1 + (s−sk 0 )n (bn + bn−1 (s − s0 ) + . . . a0 (s − s0 )n + . . .)
(bn + bn−1 (s − s0 ) + . . . a0 (s − s0 )n + . . .)
= ,
(s − s0 )n + k (bn + bn−1 (s − s0 ) + . . . a0 (s − s0 )n + . . .)
For the Nyquist plot and criterion the curve γ will always be the imaginary s-axis. That is
For a system G(s) and a feedback factor k, the Nyquist plot is the plot of the curve
w = kG ◦ γ(ω) = kG(iω).
That is, the Nyquist plot is the image of the imaginary axis under the map w = kG(s).
Note. In γ(ω) the variable is a greek omega and in w = G ◦ γ we have a double-u.
1
Example 11.16. Let G(s) = . Draw the Nyquist plot with k = 1.
s+1
Solution: In this case G(s) is a fractional linear transformation, so we know it maps the
imaginary axis to a circle. It is easy to check it is the circle through the origin with center
w = 1/2. You can also check that it is traversed clockwise.
11 ARGUMENT PRINCIPLE 11
Im(z) Im(w)
s = γ(ω) = iω
1
2i = γ(2) w = G ◦ γ(ω) = iω+1
G(−2i)
Re(z) Re(w)
G(∞) 1 G(0)
2
G(2i)
−2i = γ(−2)
The Nyquist criterion gives a graphical method for checking the stability of the closed loop
system.
Theorem 11.18. Nyquist criterion. Suppose that G(s) has a finite number of zeros and
poles in the right half-plane. Also suppose that G(s) decays to 0 as s goes to infinity. Then
the closed loop system with feedback factor k is stable if and only if the winding number of
the Nyquist plot around w = −1 equals the number of poles of G(s) in the right half-plane.
More briefly,
GCL (s) is stable ⇔ Ind(kG ◦ γ, −1) = PG,RHP
Here, γ is the imaginary s-axis and PG,RHP is the number of poles of the original open
loop system function G(s) in the right half-plane.
Proof. GCL is stable exactly when all its poles are in the left half-plane. Now, recall that the
poles of GCL are exactly the zeros of 1 + kG. So, stability of GCL is exactly the condition
that the number of zeros of 1 + kG in the right half-plane is 0.
Let’s work with a familiar contour.
11 ARGUMENT PRINCIPLE 12
Im(w)
iR
CR
x
C1 z1
Re(w)
x
z2
−iR
Finally, we can let R go to infinity. The assumption that G(s) decays 0 to as s goes to ∞
implies that in the limit, the entire curve kG ◦ CR becomes a single point at the origin. So
in the limit kG ◦ γR becomes kG ◦ γ. QED
The Nyquist criterion is a visual method which requires some way of producing the Nyquist
plot. For this we will use one of the MIT Mathlets (slightly modified for our purposes).
Open the Nyquist Plot applet at
http://web.mit.edu/jorloff/www/jmoapplets/nyquist/nyquistCrit.html
Play with the applet, read the help.
Now refresh the browser to restore the applet to its original state. Check the Formula box.
The formula is an easy way to read off the values of the poles and zeros of G(s). In its
original state, applet should have a zero at s = 1 and poles at s = 0.33 ± 1.75 i.
The left hand graph is the pole-zero diagram. The right hand graph is the Nyquist plot.
Example 11.19. To get a feel for the Nyquist plot. Look at the pole diagram and use the
mouse to drag the yellow point up and down the imaginary axis. Its image under kG(s)
will trace out the Nyquis plot.
Notice that when the yellow dot is at either end of the axis its image on the Nyquist plot
is close to 0.
Example 11.20. Refresh the page, to put the zero and poles back to their original state.
There are two poles in the right half-plane, so the open loop system G(s) is unstable. With
k = 1, what is the winding number of the Nyquist plot around -1? Is the closed loop system
stable?
11 ARGUMENT PRINCIPLE 13
Solution: The curve winds twice around -1 in the counterclockwise direction, so the winding
number Ind(kG ◦ γ, −1) = 2. Since the number of poles of G in the right half-plane is the
same as this winding number, the closed loop system is stable.
Example 11.21. With the same poles and zeros, move the k slider and determine what
range of k makes the closed loop system stable.
Solution: When k is small the Nyquist plot has winding number 0 around -1. For these
values of k, GCL is unstable. As k increases, somewhere between k = 0.65 and k = 0.7
the winding number jumps from 0 to 2 and the closed loop system becomes stable. This
continues until k is between 3.10 and 3.20, at which point the winding number becomes 1
and GCL becomes unstable.
Answer: The closed loop system is stable for k (roughly) between 0.7 and 3.10.
Example 11.22. In the previous problem could you determine analytically the range of k
where GCL (s) is stable?
Solution: Yes! This is possible for small systems. It is more challenging for higher order
systems, but there are methods that don’t require computing the poles.
In this case, we have
s−1
G(s) (s−0.33)2 +1.752 s−1
GCL (s) = = =
1 + kG(s) k(s−1)
1 + (s−0.33) 2 +1.752
(s − 0.33) + 1.752 + k(s − 1)
2
Given Equation 7, in 18.04 we can ask if there are any poles in the right half-plane without
needing any underlying physical model. Still, it’s nice to have some sense of where this fits
into science and engineering.
In a negative feedback loop the output of the system is looped back and subtracted from
the input.
Example 11.25. The heating system in my house is an example of a system stabilized
by feedback. The thermostat is the feedback mechanism. When the temperature outside
(input signal) goes down the heat turns on. Without the thermostat it would stay on and
overheat my house. The thermostat turns the heat up or down depending on whether the
inside temperature (the output signal) is too low or too high (negative feedback).
Example 11.26. Walking or balancing on one foot are examples negative feedback systems.
If you feel yourself falling you compensate by shifting your weight or tensing your muscles
to counteract the unwanted acceleration.