CHAPTER 27 / FREQUENCY DOMAIN

The Nyquist stability criterion: counting encirclements

Draw L(jω) once, count how many times it circles −1, and you know how many closed-loop poles lie in the right half-plane.

THE SHORT ANSWER

The Nyquist stability criterion decides closed-loop stability from the open-loop frequency response. Count N, the net clockwise encirclements of −1 by the complete Nyquist curve of L(s), and P, the open-loop poles in the right half-plane. The closed loop then has Z = N + P poles in the right half-plane, so it is stable exactly when N = −P.

Z=N+P,Z=0  ⟺  N=−PZ=N+P,\qquad Z=0\iff N=-P
WHAT YOU WILL LEARN

Sketch the polar plot of L(jω), count the encirclements of −1, and use Z = N + P to find the gain range for closed-loop stability.

The previous chapter showed why −1 is the critical point and read a margin at one crossover. That is enough for simple loops. Here you use the whole curve, which also settles the cases a single margin cannot: several crossings, integrators, and plants that are unstable before feedback is applied.

Review the previous lesson: Why −180° matters: Nyquist and margins
START WITH AN EXAMPLE

First, picture it

Drive a peg into the complex plane at the point −1 and tie a pencil to it with a string. Now let s walk once around the whole right half-plane: up the imaginary axis from −j∞ to +j∞, then back along a huge semicircle. At every s, hold the pencil at L(s). When s returns to its start, the pencil has drawn a closed curve and the string is wrapped around the peg a whole number of times. That number is all you need. Each closed-loop pole in the right half-plane winds the string once clockwise, and each open-loop pole there winds it once counter-clockwise. Count the wraps, add the number of unstable open-loop poles, and you have the number of unstable closed-loop poles without solving the characteristic equation.

THE IDEA

Build the idea step by step

The procedure has five steps. First, count P, the poles of L(s) in the right half-plane. Second, sketch the polar plot of L(jω) for ω from 0 to ∞: find where it starts, where it ends, and where it crosses the real axis by solving Im L(jω) = 0. Third, complete the curve: reflect it in the real axis for negative frequencies, and if L has poles at the origin, join the two ends with a large arc that turns clockwise by half a turn per pole. Fourth, count N, the number of times the complete curve encircles −1 clockwise; one counter-clockwise encirclement counts as −1. Fifth, compute Z = N + P. The closed loop is stable exactly when Z = 0, that is, when N = −P. This page counts clockwise as positive from start to finish; books that count counter-clockwise as positive write the same theorem as Z = P − N.

Read the equations

F(s)=1+L(s)=DL(s)+NL(s)DL(s),L(s)=NL(s)DL(s)F(s)=1+L(s)=\frac{D_L(s)+N_L(s)}{D_L(s)},\qquad L(s)=\frac{N_L(s)}{D_L(s)}

The zeros of F are the closed-loop poles and the poles of F are the open-loop poles. The Nyquist criterion counts both inside the right half-plane.

Z=N+PZ=N+P

Z is the number of closed-loop poles in the right half-plane and P the number of open-loop poles there. N is the number of clockwise encirclements of −1 by the complete Nyquist curve; a counter-clockwise encirclement counts as −1.

closed loop stable  ⟺  Z=0  ⟺  N=−P\text{closed loop stable}\iff Z=0\iff N=-P

With a stable open loop (P = 0) the curve must not encircle −1. With P unstable open-loop poles it must encircle −1 exactly P times counter-clockwise.

∠L(j0+)=−90∘⋅k,lim⁡ω→∞∠L(jω)=−90∘ (n−m)\angle L(j0^+)=-90^\circ\cdot k,\qquad \lim_{\omega\to\infty}\angle L(j\omega)=-90^\circ\,(n-m)

These give the two ends of the polar plot for a loop with k poles at the origin, n poles and m zeros in total. They hold for a positive gain with all other poles and zeros in the left half-plane.

Im⁡L(jωpc)=0,Re⁡L(jωpc)<0⇒crossing at L(jωpc)\operatorname{Im}L(j\omega_{pc})=0,\quad \operatorname{Re}L(j\omega_{pc})<0\qquad\Rightarrow\qquad\text{crossing at }L(j\omega_{pc})

Solve Im L(jω) = 0 for ω, then substitute into Re L. A negative result is a crossing of the negative real axis, and its frequency is a phase-crossover frequency.

L(s)=K L0(s),L0(jωpc)=−a⇒Kcrit=1aL(s)=K\,L_0(s),\quad L_0(j\omega_{pc})=-a\qquad\Rightarrow\qquad K_{\mathrm{crit}}=\frac1a

Changing K scales the whole curve about the origin, so the crossing sits at −Ka. It reaches −1 at the critical gain 1/a, where the encirclement count changes.

s=εejθ,  θ:−90∘→+90∘⇒L(s)≈cεk e−jkθs=\varepsilon e^{j\theta},\ \ \theta:-90^\circ\to+90^\circ\qquad\Rightarrow\qquad L(s)\approx\frac{c}{\varepsilon^{k}}\,e^{-jk\theta}

This is the indentation around k poles at the origin. Its image is an arc of very large radius that turns clockwise through k half-turns and joins the ω = 0⁻ end of the plot to the ω = 0⁺ end.

GM=1∣L(jωpc)∣,GMdB=−20log⁡10∣L(jωpc)∣\mathrm{GM}=\frac{1}{|L(j\omega_{pc})|},\qquad \mathrm{GM_{dB}}=-20\log_{10}|L(j\omega_{pc})|

The gain margin is read where the curve crosses the negative real axis. It is the factor by which the gain can rise before that crossing reaches −1, for a loop already shown to be stable.

PM=180∘+∠L(jωgc),∣L(jωgc)∣=1\mathrm{PM}=180^\circ+\angle L(j\omega_{gc}),\qquad |L(j\omega_{gc})|=1

The phase margin is read where the curve crosses the unit circle. It is the extra phase lag that would rotate that point onto −1.

What the symbols mean

L(s)L(s)
open-loop (loop) transfer function, dimensionless
F(s)F(s)
1 + L(s), the closed-loop denominator function
NL, DLN_L,\ D_L
numerator and denominator polynomials of L(s)
ZZ
number of closed-loop poles in the right half-plane
PP
number of poles of L(s) in the right half-plane
NN
net clockwise encirclements of −1 by the complete Nyquist curve; counter-clockwise counts negative
KK
adjustable positive loop gain
L0(s)L_0(s)
the loop transfer function with K = 1
ω\omega
angular frequency, rad/s
ωpc\omega_{pc}
phase-crossover frequency, where the curve crosses the negative real axis, rad/s
ωgc\omega_{gc}
gain-crossover frequency, where |L| = 1, rad/s
aa
distance from the origin to the negative-real-axis crossing of L₀
KcritK_{\mathrm{crit}}
gain at which the curve passes through −1
kk
number of poles of L(s) at the origin (system type)
n, mn,\ m
total numbers of poles and zeros of L(s)
ε, θ\varepsilon,\ \theta
radius and angle of the small indentation around the origin
cc
value at s = 0 of L(s) with its origin poles removed
GM, PM\mathrm{GM},\ \mathrm{PM}
gain margin (a ratio, or dB) and phase margin (degrees)

The count assumes that L(s) has more poles than zeros, so the large semicircle of the contour maps to the origin and only s = jω matters. Poles of L at the origin are not counted in P, because the contour is indented to pass them on the right. If the curve passes exactly through −1 the count is undefined: the closed loop then has a pole on the imaginary axis, which is the boundary between stable and unstable.

WORKED EXAMPLE

Three loops, one counting rule

  1. Case A, a stable open loop. The poles are at −1, −2 and −3, so P = 0. Expand the denominator before substituting s = jω.

    L(s)=K(s+1)(s+2)(s+3)=Ks3+6s2+11s+6,P=0L(s)=\frac{K}{(s+1)(s+2)(s+3)}=\frac{K}{s^3+6s^2+11s+6},\qquad P=0
  2. Collect the real and imaginary parts of the denominator. The plot starts at K/6 on the positive real axis and, with three more poles than zeros, enters the origin at −270°.

    L(jω)=K(6−6ω2)+jω(11−ω2),L(j0)=K6L(j\omega)=\frac{K}{(6-6\omega^2)+j\omega(11-\omega^2)},\qquad L(j0)=\frac K6
  3. The curve crosses the real axis where the imaginary part of the denominator is zero. Substitute that frequency into the real part.

    11−ω2=0 ⇒ ωpc=11≈3.32 rad/s,L(j11)=K6−66=−K6011-\omega^2=0\ \Rightarrow\ \omega_{pc}=\sqrt{11}\approx3.32\ \mathrm{rad/s},\qquad L(j\sqrt{11})=\frac{K}{6-66}=-\frac{K}{60}
  4. Count. For K < 60 the crossing lies to the right of −1 and the curve does not encircle it. For K > 60 the curve and its mirror image together go around −1 twice clockwise.

    0<K<60:  N=0, Z=0+0=0K>60:  N=2, Z=2+0=20<K<60:\ \ N=0,\ Z=0+0=0\qquad\qquad K>60:\ \ N=2,\ Z=2+0=2
  5. Check with the characteristic polynomial. The Routh condition gives the same limit. At K = 120 the roots are 0.50 ± j4.21 and −7.00: two in the right half-plane, as counted.

    s3+6s2+11s+(6+K)=0:6⋅11>6+K  ⟺  K<60s^3+6s^2+11s+(6+K)=0:\qquad 6\cdot11>6+K\iff K<60
  6. Case B, an integrator in the loop. The contour is indented around the pole at the origin, so that pole is outside and P = 0.

    L(s)=Ks(s+1)(s+2),L(jω)=K−3ω2+jω(2−ω2),P=0L(s)=\frac{K}{s(s+1)(s+2)},\qquad L(j\omega)=\frac{K}{-3\omega^2+j\omega(2-\omega^2)},\qquad P=0
  7. As ω → 0⁺ the magnitude grows without limit at −90°, and the curve comes down along a vertical asymptote. Find the asymptote from the real part.

    Re⁡L(jω)=−3K9ω2+(2−ω2)2 ⟶ −3K4(ω→0+)\operatorname{Re}L(j\omega)=\frac{-3K}{9\omega^2+(2-\omega^2)^2}\ \longrightarrow\ -\frac{3K}{4}\quad(\omega\to0^+)
  8. The real-axis crossing again comes from the imaginary part of the denominator.

    2−ω2=0 ⇒ ωpc=2 rad/s,L(j2)=K−6=−K62-\omega^2=0\ \Rightarrow\ \omega_{pc}=\sqrt2\ \mathrm{rad/s},\qquad L(j\sqrt2)=\frac{K}{-6}=-\frac K6
  9. Close the curve. The indentation maps to an arc at infinity from +90° through 0° to −90°, clockwise through the right half-plane. With that arc in place, −1 is outside for K < 6 and is encircled twice clockwise for K > 6. The limit K = 6 is the same value that the Routh table and the imaginary-axis crossing of the root locus give.

    0<K<6:  N=0, Z=0K>6:  N=2, Z=20<K<6:\ \ N=0,\ Z=0\qquad\qquad K>6:\ \ N=2,\ Z=2
  10. Case C, an unstable open loop. The pole at +1 gives P = 1. Make the denominator real to read off the real and imaginary parts.

    L(s)=Ks−1,L(jω)=Kjω−1=−K(1+jω)1+ω2,P=1L(s)=\frac{K}{s-1},\qquad L(j\omega)=\frac{K}{j\omega-1}=\frac{-K(1+j\omega)}{1+\omega^2},\qquad P=1
  11. The plot starts at −K, stays in the lower half-plane for ω > 0 and ends at the origin: it is a circle of diameter K centred at −K/2. As ω runs from −∞ to +∞ the circle is traced counter-clockwise.

    ∣L(jω)+K2∣=K2,L(j0)=−K,L(j∞)=0\left|L(j\omega)+\frac K2\right|=\frac K2,\qquad L(j0)=-K,\qquad L(j\infty)=0
  12. Count. For K > 1 the point −1 is inside the circle: one counter-clockwise encirclement, which is exactly what P = 1 requires. For K < 1 it is outside and the unstable pole remains.

    K>1:  N=−1, Z=−1+1=00<K<1:  N=0, Z=0+1=1K>1:\ \ N=-1,\ Z=-1+1=0\qquad\qquad 0<K<1:\ \ N=0,\ Z=0+1=1
  13. Check with the characteristic equation: the single closed-loop pole is at 1 − K. Here raising the gain stabilises the loop. At K = 2 the crossing is at −2, and the gain may fall by a factor of 2 before stability is lost.

    1+Ks−1=0 ⇒ s=1−K,s<0  ⟺  K>11+\frac{K}{s-1}=0\ \Rightarrow\ s=1-K,\qquad s<0\iff K>1
More detail and model assumptions
A CLOSER LOOK

Where Z = N + P comes from

Write F(s) = 1 + L(s). If L is the ratio of a numerator polynomial to a denominator polynomial, then F is (denominator + numerator) divided by the denominator. The zeros of F are the roots of the closed-loop characteristic equation, and the poles of F are the open-loop poles. So Z counts the zeros of F in the right half-plane and P counts its poles there.

Factor F into terms (s − zᵢ) on top and (s − pᵢ) underneath, and move s once clockwise around a closed path. A factor whose root lies inside the path turns through 360° clockwise; a factor whose root lies outside swings back and forth and returns with no net turn. Factors on top add their angle to F, and factors underneath subtract theirs. So F(s) circles the origin Z − P times clockwise. This is the argument principle.

Choose the path that encloses the entire right half-plane: the imaginary axis plus a semicircle of radius R → ∞. Then N = Z − P with Z and P counted in the right half-plane. F circling the origin is the same as L = F − 1 circling −1, so you plot L itself, and one plot serves every gain because multiplying L by K only rescales the curve. Only L(jω) enters the test, so it also works on measured frequency-response data and on loops with a time delay, where a Routh table cannot be written.

A CLOSER LOOK

Sketching the polar plot by hand

Three facts fix the shape. The start: as ω → 0, a type-0 loop begins at its DC gain on the real axis, a type-1 loop begins infinitely far away at −90°, and a type-2 loop begins infinitely far away at −180°. The end: as ω → ∞ the curve enters the origin at an angle of −90° times (number of poles − number of zeros). The crossings: wherever Im L(jω) = 0, the curve crosses the real axis at Re L(jω). These angles assume a positive gain and no right-half-plane poles or zeros; each such factor changes the angles, as case C of the worked example shows.

For L(s) = K/D(s) the algebra is short. Put s = jω in D and collect real and imaginary parts: D(jω) = A(ω) + jB(ω). The real-axis crossings are the roots of B(ω) = 0, and at each of them L = K/A. For a type-1 loop, also find the vertical asymptote: Re L(jω) tends to a finite constant as ω → 0 while Im L goes to −∞. In case B of the worked example, Re L = −3K/(9ω² + (2 − ω²)²), which tends to −3K/4.

When L has a numerator, first multiply top and bottom by the complex conjugate of the denominator so that the denominator is real, then read off Re L and Im L. Always mark the direction of increasing ω with an arrow. Without the arrow you cannot tell a clockwise encirclement from a counter-clockwise one.

A CLOSER LOOK

Poles at the origin: indent the contour and close the curve

A pole of L at s = 0 lies on the contour, and L is infinite there. Step around it on a semicircle of small radius ε in the right half-plane: s = εe^{jθ}, with θ running from −90° to +90°. The pole is now outside the contour, which is why it is not counted in P.

On that semicircle a type-k loop behaves like L ≈ c/sᵏ = (c/εᵏ)e^{−jkθ}, where c is what remains of L at s = 0 after the origin poles are removed. The radius is enormous, and the angle falls by k × 180° while θ rises by 180°. So the image is an arc at infinity that turns clockwise through k half-turns, from the ω = 0⁻ end of the plot to the ω = 0⁺ end. For K/(s(s+1)(s+2)), c = K/2 is positive and the arc runs from +90° through 0° to −90°: it closes through the right half-plane.

The sign of c decides on which side the arc passes. Take L = K(s+1)/(s(s−1)), which has P = 1. Here c = −K, so the arc runs from −90° through 180° to +90°: it closes on the left, far beyond −1. The finite part of the plot crosses the real axis at −K when ω = 1. For K > 1 the point −1 is encircled once counter-clockwise, so N = −1 and Z = 0. For K < 1 only the far arc goes around it, so N = +1 and Z = 2. The closed-loop polynomial s² + (K − 1)s + K agrees: it is stable only for K > 1.

A CLOSER LOOK

Margins on the polar plot, and when they mislead

Both margins are distances from the curve to −1, each measured in one direction. If the curve crosses the negative real axis at −a, the gain can grow by the factor 1/a before the crossing reaches −1: GM = 1/a, or −20 log₁₀ a in decibels. The phase margin is the angle, seen from the origin, between the negative real axis and the point where the curve crosses the unit circle. In case A with K = 30 the crossing is at −0.5, so GM = 2 (6 dB), and the unit-circle crossing at ω ≈ 2.35 rad/s gives PM ≈ 25°.

A loop is conditionally stable when it is stable only for gains inside an interval, so that lowering the gain can destabilise it. For example, L = K(s+1)²/s³ crosses the real axis at −2K when ω = 1 and is stable only for K > 0.5. In the stable range the crossing lies to the left of −1, and the formula GM = 1/a returns a number below 1. The correct statement is that the gain may fall by the factor 2K before stability is lost.

So use margins in this order: establish stability with the encirclement count first, then read the margins, and report every real-axis crossing when there are several. Both margins can also look healthy while the curve bulges close to −1 between the two crossings. The shortest distance from −1 to the curve catches that case, and the sensitivity chapter returns to it.

INTERACTIVE / L(s) = K/(s(s+1)(s+2))Move the K slider, then count how often the curve circles −1
−1ReIm|L|→∞K = 2.00
  • ω > 0
  • ω < 0 (mirror image)
  • detour around s = 0
  • phase-margin arc and gain-margin gap
  • dashed circle |L| = 1.5: true scale inside, compressed outside
N (clockwise)
0
P (open-loop RHP)
0
Z = N + P
0
RHP roots of 1 + L
0
Gain factor to reach −1
×3.00 (9.5 dB)
Phase margin
32.6°
closed loop stable

N = 0, P = 0, so Z = N + P = 0: closed loop stable. Solving 1 + L(s) = 0 directly gives −0.24 ± 0.86j, −2.52, with 0 in the right half-plane. Multiplying K by 3.00 (9.5 dB) would move the real-axis crossing at ω = 1.41 rad/s onto −1. Phase margin 32.6° at the gain crossover ω = 0.75 rad/s. A pole at s = 0: the contour detours around it, and its image is the large clockwise arc at |L| → ∞.

Interactive Nyquist explorer: choose a loop, change the gain and compare how often the curve encircles −1 with the closed-loop poles and the gain and phase margins. Try it online at eigenroom.com/theory/nyquist-criterion.

TRY IT

Connect the example to the lab

Open the motor lab's frequency workbench and choose the Nyquist tab. The motor loop has no unstable open-loop pole, so P = 0 and the nominal closed loop is stable only if the plotted curve, together with its mirror image, leaves −1 outside. With the preset gains the workbench reports a gain margin of about 25 dB, so the curve crosses the negative real axis near −0.05. The lab plots the sampled loop L(e^{jωT}); the counting rule is the same, with 'outside the unit circle' in place of 'right half-plane'.

Open the Nyquist plot in the motor lab
YOUR EXPERIMENT

Make a prediction before moving a control.

  1. PREDICTThe motor loop is open-loop stable, so P = 0. If you double Kp, Ki and Kd together, where does each point of the Nyquist curve move, and by how many decibels does the gain margin change?
  2. CHANGE ONE THINGIn this lesson open the frequency workbench, choose the Nyquist tab and note both margins. Then double all three gains: Kp 0.4 → 0.8, Ki 0.6 → 1.2, Kd 0.008 → 0.016. Change nothing else.
  3. OBSERVECheck that the curve keeps its shape while every point moves twice as far from the origin, and that −1 is still outside. The gain margin should drop from about 25.3 dB to about 19.3 dB, with the phase crossover staying near 42 rad/s; the phase margin falls from about 68° to about 57°.
  4. EXPLAINExplain why doubling L moves the real-axis crossing from −a to −2a without changing its frequency, and why that costs 20 log₁₀ 2 ≈ 6 dB of gain margin. Then state how many more doublings the nominal loop could take before the curve encircles −1, and what Z would be at that point.
BE CAREFUL

'The curve must not encircle −1' is the stability condition only when P = 0. For an open-loop unstable plant, zero encirclements means the unstable pole is still there: stability needs exactly P counter-clockwise encirclements. Count on the complete curve, including the mirror image and any closing arc, never on the positive-frequency half alone.

Check your understanding+

L(s) has one pole in the right half-plane. Its complete Nyquist curve encircles −1 once clockwise. How many closed-loop poles are in the right half-plane?

Two. N = +1 and P = 1, so Z = N + P = 2. Stability would have required one counter-clockwise encirclement, N = −1.
PRACTICE PROBLEMS

Solve these before opening the solutions

Work each problem on paper first. Open the answer to check yours, then the full solution if you need the steps, and mark how it went.

  1. A loop has three equal lags. Find where its polar plot crosses the negative real axis, the range of K > 0 for closed-loop stability, and the gain margin when K = 16.

    L(s)=K(s+2)3L(s)=\frac{K}{(s+2)^3}
    Show answer

    AnswerThe crossing is at −K/64 (ω = 2√3 rad/s). The loop is stable for 0 < K < 64, and at K = 16 the gain margin is 4, about 12 dB.

    L(j23)=−K64,0<K<64,GM∣K=16=4≈12.0 dBL(j2\sqrt3)=-\frac{K}{64},\qquad 0<K<64,\qquad \mathrm{GM}\big|_{K=16}=4\approx12.0\ \mathrm{dB}
    Show full solution

    Solution

    1. All three poles are at −2, so P = 0. Expand the denominator at s = jω.

      (jω+2)3=(8−6ω2)+jω(12−ω2)(j\omega+2)^3=(8-6\omega^2)+j\omega(12-\omega^2)
    2. Set the imaginary part to zero and evaluate the real part there.

      ωpc=12=23≈3.46 rad/s,L(jωpc)=K8−72=−K64\omega_{pc}=\sqrt{12}=2\sqrt3\approx3.46\ \mathrm{rad/s},\qquad L(j\omega_{pc})=\frac{K}{8-72}=-\frac{K}{64}
    3. With P = 0 the loop is stable when −1 is not encircled, that is, when the crossing is to the right of −1.

      K64<1  ⟺  K<64(K>64: N=2, Z=2)\frac{K}{64}<1\iff K<64\qquad(K>64:\ N=2,\ Z=2)
    4. At K = 16 the crossing is at −0.25. The gain margin is the reciprocal of that distance.

      GM=10.25=4,20log⁡104≈12.0 dB\mathrm{GM}=\frac{1}{0.25}=4,\qquad 20\log_{10}4\approx12.0\ \mathrm{dB}
  2. For this type-1 loop find the vertical asymptote of the polar plot, the real-axis crossing and the stable range of K > 0. Then state N and Z for K = 96.

    L(s)=Ks(s+2)(s+4)L(s)=\frac{K}{s(s+2)(s+4)}
    Show answer

    AnswerAsymptote Re L = −3K/32; crossing at −K/48 when ω = 2√2 rad/s; stable for 0 < K < 48. At K = 96, N = 2 and Z = 2.

    Re⁡L→−3K32,L(j22)=−K48,0<K<48,K=96: N=2, Z=2\operatorname{Re}L\to-\frac{3K}{32},\qquad L(j2\sqrt2)=-\frac{K}{48},\qquad 0<K<48,\qquad K=96:\ N=2,\ Z=2
    Show full solution

    Solution

    1. P = 0: the poles at −2 and −4 are stable and the origin pole is passed on the right. Write the denominator at s = jω.

      L(jω)=K−6ω2+jω(8−ω2)L(j\omega)=\frac{K}{-6\omega^2+j\omega(8-\omega^2)}
    2. Multiply by the conjugate to get the real part, then let ω → 0.

      Re⁡L(jω)=−6K36ω2+(8−ω2)2 ⟶ −6K64=−3K32\operatorname{Re}L(j\omega)=\frac{-6K}{36\omega^2+(8-\omega^2)^2}\ \longrightarrow\ -\frac{6K}{64}=-\frac{3K}{32}
    3. The crossing is where the imaginary part of the denominator vanishes.

      ωpc=8=22 rad/s,L(jωpc)=K−48\omega_{pc}=\sqrt8=2\sqrt2\ \mathrm{rad/s},\qquad L(j\omega_{pc})=\frac{K}{-48}
    4. The closing arc runs clockwise through the right half-plane, so −1 is outside when K/48 < 1.

      0<K<48: N=0, Z=00<K<48:\ N=0,\ Z=0
    5. At K = 96 the crossing is at −2, to the left of −1, and the complete curve encircles −1 twice clockwise. The roots confirm it.

      s3+6s2+8s+96=0 ⇒ s≈0.43±j3.71, −6.87s^3+6s^2+8s+96=0\ \Rightarrow\ s\approx0.43\pm j3.71,\ -6.87
  3. Read the gain and phase margins of this loop from its polar plot: find where the curve meets the unit circle and whether it ever meets the negative real axis.

    L(s)=4s(s+2)L(s)=\frac{4}{s(s+2)}
    Show answer

    AnswerThe curve meets the unit circle at ω ≈ 1.57 rad/s with phase −128.2°, so PM ≈ 51.8°. It never reaches the negative real axis at a finite frequency, so the gain margin is infinite.

    ωgc≈1.57 rad/s,PM≈51.8∘,GM=∞\omega_{gc}\approx1.57\ \mathrm{rad/s},\qquad \mathrm{PM}\approx51.8^\circ,\qquad \mathrm{GM}=\infty
    Show full solution

    Solution

    1. Set the magnitude to 1 and solve the resulting quadratic in ω².

      4ωω2+4=1 ⇒ ω4+4ω2−16=0 ⇒ ω2=−2+25≈2.472\frac{4}{\omega\sqrt{\omega^2+4}}=1\ \Rightarrow\ \omega^4+4\omega^2-16=0\ \Rightarrow\ \omega^2=-2+2\sqrt5\approx2.472
    2. Evaluate the phase at that frequency: −90° from the integrator plus the lag of the pole at −2.

      ωgc≈1.572,∠L=−90∘−tan⁡−11.5722≈−128.2∘\omega_{gc}\approx1.572,\qquad \angle L=-90^\circ-\tan^{-1}\frac{1.572}{2}\approx-128.2^\circ
    3. The phase margin is the angle left before −180°.

      PM=180∘−128.2∘≈51.8∘\mathrm{PM}=180^\circ-128.2^\circ\approx51.8^\circ
    4. The imaginary part is negative for every finite ω > 0, so there is no phase crossover and no finite gain makes the curve reach −1.

      Im⁡L(jω)=−8ω(ω2+4)<0 ⇒ GM=∞\operatorname{Im}L(j\omega)=\frac{-8}{\omega(\omega^2+4)}<0\ \Rightarrow\ \mathrm{GM}=\infty
  4. This loop has an integrator and an unstable pole. Sketch the complete Nyquist curve, including the closing arc, and find the range of K > 0 for which the closed loop is stable. Verify with the closed-loop characteristic polynomial.

    L(s)=K(s+4)s(s−2)L(s)=\frac{K(s+4)}{s(s-2)}
    Show answer

    AnswerP = 1. The curve crosses the real axis at −K/2 (ω = 2√2 rad/s) and the closing arc passes on the far left. Stable for K > 2, where N = −1; for 0 < K < 2, N = +1 and Z = 2.

    K>2: N=−1, Z=00<K<2: N=+1, Z=2K>2:\ N=-1,\ Z=0\qquad\qquad 0<K<2:\ N=+1,\ Z=2
    Show full solution

    Solution

    1. P = 1 from the pole at +2. Multiply top and bottom by the conjugate of the denominator.

      L(jω)=K(4+jω)−ω2−2jω ⇒ Re⁡L=−6Kω2+4,Im⁡L=K(8−ω2)ω(ω2+4)L(j\omega)=\frac{K(4+j\omega)}{-\omega^2-2j\omega}\ \Rightarrow\ \operatorname{Re}L=\frac{-6K}{\omega^2+4},\quad \operatorname{Im}L=\frac{K(8-\omega^2)}{\omega(\omega^2+4)}
    2. Ends and crossing: the positive-frequency branch comes down from +j∞ along Re L = −1.5K, crosses the real axis at ω = 2√2 and enters the origin from below.

      ω→0+: L→−3K2+j∞,L(j22)=−K2,ω→∞: L→0\omega\to0^+:\ L\to-\frac{3K}{2}+j\infty,\qquad L(j2\sqrt2)=-\frac K2,\qquad \omega\to\infty:\ L\to0
    3. Closing arc: near the origin L ≈ −2K/s. On the indentation its angle runs from −90° through 180° to +90°, a clockwise half-turn through the far left.

      s=εejθ:L≈2Kε ej(180∘−θ)s=\varepsilon e^{j\theta}:\qquad L\approx\frac{2K}{\varepsilon}\,e^{j(180^\circ-\theta)}
    4. Count along the ray from −1 to the left. For K > 2 it meets the two branches through −K/2, each a counter-clockwise crossing, and the far arc, a clockwise crossing: net one counter-clockwise encirclement. For K < 2 it meets only the far arc.

      K>2: N=−2+1=−1, Z=0K<2: N=+1, Z=2K>2:\ N=-2+1=-1,\ Z=0\qquad\qquad K<2:\ N=+1,\ Z=2
    5. Check: the characteristic polynomial is stable exactly when both coefficients are positive. At K = 1 its roots are 0.50 ± j1.94, two in the right half-plane.

      s2+(K−2)s+4K=0:K−2>0 and 4K>0  ⟺  K>2s^2+(K-2)s+4K=0:\qquad K-2>0\ \text{and}\ 4K>0\iff K>2
  5. Exam-level. Use the Nyquist criterion to find every K > 0 for which the closed loop is stable, and give Z in each remaining range of K. Then confirm the result with a Routh table.

    L(s)=K(s−1)(s+2)(s+3)L(s)=\frac{K}{(s-1)(s+2)(s+3)}
    Show answer

    AnswerStable only for 6 < K < 10. For 0 < K < 6, Z = 1; for K > 10, Z = 2.

    0<K<6: Z=16<K<10: Z=0K>10: Z=20<K<6:\ Z=1\qquad 6<K<10:\ Z=0\qquad K>10:\ Z=2
    Show full solution

    Solution

    1. P = 1. Expand the denominator and split it at s = jω.

      D(s)=s3+4s2+s−6,D(jω)=−(4ω2+6)+jω(1−ω2)D(s)=s^3+4s^2+s-6,\qquad D(j\omega)=-(4\omega^2+6)+j\omega(1-\omega^2)
    2. The imaginary part vanishes at ω = 0 and at ω = 1, so the curve touches the real axis twice before ending at the origin.

      L(j0)=−K6,L(j1)=K−10=−K10L(j0)=-\frac K6,\qquad L(j1)=\frac{K}{-10}=-\frac{K}{10}
    3. Direction: for 0 < ω < 1 the curve is below the axis, and for ω > 1 above it. With the mirror image this makes two loops: a small one between −K/6 and −K/10 traced counter-clockwise, and a large one between −K/10 and 0 traced clockwise.

      Im⁡L(jω)=−K ω(1−ω2)∣D(jω)∣2\operatorname{Im}L(j\omega)=\frac{-K\,\omega(1-\omega^2)}{|D(j\omega)|^2}
    4. Place −1 in each region and apply Z = N + P.

      0<K<6:  N=0,Z=16<K<10:  N=−1,Z=0K>10:  N=+1,Z=2\begin{aligned}0<K<6:&\ \ N=0,&Z&=1\\6<K<10:&\ \ N=-1,&Z&=0\\K>10:&\ \ N=+1,&Z&=2\end{aligned}
    5. Routh check: the s⁰ entry needs K > 6 and the s¹ entry needs K < 10. At K = 8 the crossings are at −1.33 and −0.8, so the gain may rise by a factor of 1.25 or fall to 0.75 of its value.

      s3+4s2+s+(K−6)=0:K−6>0,4⋅1>K−6  ⟺  6<K<10s^3+4s^2+s+(K-6)=0:\qquad K-6>0,\quad 4\cdot1>K-6\iff 6<K<10

Mark each exercise after you check your answer. Marks stay in this browser.

KEY TERMS

Terms this lecture teaches

All terms in the glossary →
QUESTIONS STUDENTS ASK

Frequently asked questions

Why do some textbooks write Z = P − N instead of Z = N + P?

They count counter-clockwise encirclements as positive. With clockwise counted positive the theorem reads Z = N + P; with counter-clockwise positive the same theorem reads Z = P − N. Pick one convention and use it from start to finish.

Do open-loop poles at the origin count in P?

No. The Nyquist contour is indented around poles at s = 0 so that they lie outside it, and P counts only poles inside the right half-plane. The indentation maps to a large arc that turns clockwise by half a turn per origin pole and closes the curve.

Is 'the Nyquist curve must not encircle −1' always the stability condition?

Only when the open loop is stable, P = 0. If L has P unstable poles, the complete curve must encircle −1 exactly P times counter-clockwise; zero encirclements then leaves P closed-loop poles in the right half-plane. Count on the complete curve, including the mirror image and any closing arc.

Why use the Nyquist criterion instead of a Routh table?

It needs only L(jω), so it works with measured frequency-response data and with time delays, where no polynomial exists for a Routh table. Changing K only rescales the curve, so the stable gain range is visible at once, and the plot also shows how close the loop comes to −1.

Why the next lesson follows

The count tells you whether a loop is stable and how far its curve stays from −1. When that distance is too small you do not have to give up gain. A lead or lag compensator reshapes L(jω) near the crossover and bends the curve away from −1; that is the next chapter.

YOUR PROGRESS

Finished the reading and the exercises? Mark the lecture as done to tick it on the syllabus. Progress stays in this browser.

EIGENROOM · https://eigenroom.com/theory/nyquist-criterion · Updated 2026-11-15