The argument principle and Rouché's theorem
How can the zeros of a function inside a loop be counted without finding them, and when do two functions have the same number of zeros there?
Where you are 5 in Series and residues
Why this exists
The The residue theorem turns a loop integral into a sum over the singularities inside. Turned around, it answers a question with no integral in it: how many zeros does f have inside a loop? The quotient \displaystyle \frac{f'}{f} has a simple pole at each zero of f, with residue the order of the zero, so its integral counts zeros. It also counts how often the image of the loop winds around 0, and Rouché’s theorem counts roots of equations no formula solves by comparing two functions on the loop.
Definition
Let G be an open set. A function f is meromorphic on G if it is holomorphic on G except at poles.
The logarithmic derivative of f is \displaystyle \frac{f'}{f}: if L is holomorphic with e^{L} = f, then L' e^{L} = f', so \displaystyle L' = \frac{f'}{f}.
For a loop \gamma with no zero or pole of f on it, Z and P are the numbers of zeros and of poles of f inside \gamma, counted with multiplicity: one of order m counts m times.
Let \sigma : [a, b] \to \mathbb{C} \setminus \{0\} be continuous. A continuous argument of \sigma is a continuous \theta : [a, b] \to \mathbb{R} with \sigma(t) = |\sigma(t)| e^{i\theta(t)} for all t (an argument as in Modulus, argument and polar form, chosen continuously). Result 3 shows one exists. Two differ by a continuous function with values in 2\pi\mathbb{Z}, a constant, so the change of argument \Delta \arg \sigma = \theta(b) - \theta(a) is well defined. For f continuous and nonzero on a path \gamma, \Delta_\gamma \arg f = \Delta \arg (f \circ \gamma).
Intuition
Let \displaystyle f(z) = \left(z - \frac12\right)\left(z + \frac12\right)(z - 2) and let z run once around \gamma = C(0, 1). The arrow \displaystyle z - \frac12 from the zero \displaystyle \frac12 to z turns once, so its argument grows by 2\pi; so does that of \displaystyle z + \frac12. The arrow z - 2 always points left, as \operatorname{Re}(z - 2) \le -1, so its argument returns. Arguments of a product add (Multiplying, powers and De Moivre's theorem), so the argument of f grows by 4\pi: f \circ \gamma winds twice around 0 (Figure 1). A zero of order m inside gives m turns, a pole of order m gives -m, and points outside give none.
Results
1. The logarithmic derivative at a zero or a pole. Let h be holomorphic on D(a, r), and let f be holomorphic on D(a, r) with a zero of order m \ge 1 at a, or holomorphic on D(a, r) \setminus \{a\} with a pole of order m at a. Then for some 0 < \rho \le r, \displaystyle h\frac{f'}{f} is holomorphic on D(a, \rho) \setminus \{a\}, and its residue at a is m\,h(a) at a zero and -m\,h(a) at a pole. Also, where f and g are holomorphic and nonzero, \displaystyle \frac{(fg)'}{fg} = \frac{f'}{f} + \frac{g'}{g}.
2. The argument principle. Let \gamma be a positively oriented simple closed piecewise smooth path, and f meromorphic on an open set G containing \gamma and its inside, not identically 0 on any disk in G, with no zero or pole on \gamma. Then f has finitely many zeros and poles inside \gamma, and
More generally, if the zeros inside \gamma are a_k of orders m_k, the poles are b_j of orders n_j, and h is holomorphic on G, then
3. Counting by winding. Let \gamma : [a, b] \to \mathbb{C} be a closed piecewise smooth path and f holomorphic on an open set containing \gamma, with no zero on \gamma. Then f \circ \gamma is a closed piecewise smooth path not through 0, it has a continuous argument, and
Under the hypotheses of Result 2 all three equal Z - P: the argument of f changes by 2\pi(Z - P) around \gamma, hence the name.
4. Rouché’s theorem. Let \gamma be a positively oriented simple closed piecewise smooth path, and f, g holomorphic on an open set G containing \gamma and its inside, with |g(z)| < |f(z)| for every z on \gamma. Then f and f + g have the same finite number of zeros inside \gamma, counted with multiplicity.
Why it is true
Result 1. Proof. At a zero, Result 6 of Holomorphic functions are power series gives f(z) = (z - a)^m g(z) with g holomorphic on D(a, r) and g(a) \ne 0; at a pole, Result 5 of Laurent series and singularities gives f(z) = (z - a)^{-m} g(z) off a, with g the same. Write k = \pm m. By continuity g \ne 0 on some D(a, \rho), and there, off a, f' = k(z - a)^{k-1}g + (z - a)^{k}g', so
The last term is holomorphic on D(a, \rho), and by Result 2 of The residue theorem the first has residue \lim_{z \to a} k\,h(z) = k\,h(a). The last claim is (fg)' = f'g + fg' divided by fg. \blacksquare
Result 2. Proof. Each point c of G has a punctured disk around it free of zeros and poles. If c is a pole, f is holomorphic on a punctured disk around c and \displaystyle |f(z)| = \frac{|g(z)|}{|z - c|^m} \to \infty (Result 5 of Laurent series and singularities), so there are no zeros near c. If f is holomorphic at c, it is so on a disk, which holds no poles; there are no zeros near c by continuity if f(c) \ne 0, and by Result 6 of Holomorphic functions are power series if f(c) = 0, as f is not 0 on a disk. So the set E of zeros and poles has no limit point in G.
The set K made of \gamma and its inside is closed and bounded, so by Bolzano-Weierstrass it holds only finitely many points of E. The points of E outside K have no limit point in G, so removing them leaves an open set G' \supseteq K. On G', \displaystyle h\frac{f'}{f} is holomorphic except at the zeros and poles inside \gamma, with the residues of Result 1, and the residue theorem (Result 1 of The residue theorem) gives the second formula; h = 1 gives the first. \blacksquare
Result 3. Proof. Put \sigma = f \circ \gamma: closed, never 0, and piecewise smooth with \sigma'(t) = f'(\gamma(t))\gamma'(t). By the definitions,
For the argument, \mu = \min_t |\sigma(t)| > 0 (The extreme value theorem). As \sigma is uniformly continuous, some partition a = t_0 < \cdots < t_N = b has |\sigma(t) - \sigma(t_{j-1})| < \mu on each [t_{j-1}, t_j]. Let \displaystyle c_j = \frac{\sigma(t_{j-1})}{|\sigma(t_{j-1})|}. On that piece \displaystyle \frac{\sigma(t)}{c_j} lies in D(|\sigma(t_{j-1})|, \mu), inside the half-plane \operatorname{Re} w > 0, where \operatorname{Log} is holomorphic with derivative \displaystyle \frac1w (Result 4 of Complex differentiability and the Cauchy-Riemann equations). So \displaystyle L_j(t) = \operatorname{Log}\frac{\sigma(t)}{c_j} is continuous, \displaystyle e^{L_j} = \frac{\sigma}{c_j}, and \displaystyle L_j' = \frac{\sigma'}{\sigma} where \sigma is smooth. Then \operatorname{Im} L_j + \operatorname{Arg} c_j is a continuous argument on the piece; adding a multiple of 2\pi on each piece in turn joins them into a continuous argument \theta on [a, b]. By the The fundamental theorem of calculus, on each piece
Summing, the logarithms telescope to \ln|\sigma(b)| - \ln|\sigma(a)| = 0, leaving i\,\Delta\arg\sigma. \blacksquare
Result 4. Proof. On \gamma, |f| > |g| \ge 0 and |f + g| \ge |f| - |g| > 0. By the Jordan curve theorem (stated in Cauchy's theorem) the inside of \gamma is connected with \gamma in its closure, so \gamma and its inside lie in one component G_0 of G, a region. Neither f nor f + g is 0 on G_0, so by the identity principle (Result 7 of Holomorphic functions are power series) neither is 0 on a disk in G_0, and Result 2 applies to both on G_0.
As f \ne 0 on \gamma, f \ne 0 on an open set U with \gamma \subseteq U \subseteq G_0. On U put \displaystyle \varphi = 1 + \frac{g}{f}, so f + g = f\varphi. On \gamma, \displaystyle |\varphi - 1| = \frac{|g|}{|f|} < 1, so \varphi \ne 0 on \gamma, and near \gamma Result 1 gives
The path \varphi \circ \gamma lies in D(1, 1), inside \operatorname{Re} w > 0, where \operatorname{Log} is an antiderivative of \displaystyle \frac1w. By Result 5 of Contour integrals the integral of \displaystyle \frac1w over it is 0, and by the first display of the proof of Result 3 that integral is \displaystyle \int_\gamma \frac{\varphi'}{\varphi}\,dz. Integrate over \gamma and apply Result 2 to f and f + g: Z(f + g) = Z(f). \blacksquare
Worked examples
Counting without locating
Find \displaystyle \frac{1}{2\pi i}\int_{C(0, 2)} \frac{f'(z)}{f(z)}\,dz and the same over C(0, 4), for \displaystyle f(z) = \frac{(z - 1)^3 e^{z}}{z^2(z^2 + 9)}. Then find \int_{C(0, 5)} \tan z\,dz.
SolutionAs e^{z} \ne 0, f has one zero, at 1, of order 3; it has a pole of order 2 at 0 and simple poles at \pm 3i. None lies on either circle. The product rule of Result 1 shows every residue at once:
Inside C(0, 2) lie 1 and 0, so Result 2 gives Z - P = 3 - 2 = 1. Inside C(0, 4) lie also \pm 3i, and Z - P = 3 - 4 = -1.
Next, \displaystyle \tan z = -\frac{(\cos z)'}{\cos z}. The zeros of \cos z are the real numbers \displaystyle \frac{\pi}{2} + k\pi (shown in Cauchy's theorem), simple as (\cos z)' = -\sin z = \mp 1 there. Inside C(0, 5) lie \displaystyle \pm\frac{\pi}{2} and \displaystyle \pm\frac{3\pi}{2}, since \displaystyle \frac{3\pi}{2} < 5 < \frac{5\pi}{2}, and none is on it. By Result 2, \int_{C(0, 5)} \tan z\,dz = -2\pi i\cdot 4 = -8\pi i.
A polynomial in an annulus
How many zeros of p(z) = z^7 - 5z^3 + z - 1 lie in D(0, 1), and how many in \{z : 1 < |z| < 2\}?
SolutionOn C(0, 1) the term -5z^3 dominates: by the Triangle inequality,
Rouché’s theorem with f(z) = -5z^3 gives p the 3 zeros of -5z^3 in D(0, 1). On C(0, 2) the leading term dominates:
so all 7 zeros lie in D(0, 2). The strict inequalities also give p \ne 0 on both circles, so 7 - 3 = 4 zeros lie in the annulus.
A transcendental equation
Show that 3z^5 = e^{z} has exactly five solutions in D(0, 1), all simple, exactly one of them real.
SolutionLet F(z) = 3z^5 - e^{z}, f(z) = 3z^5 and g(z) = -e^{z}. On C(0, 1), \operatorname{Re} z \le 1, so
By Rouché’s theorem F has 5 zeros in D(0, 1), counted with multiplicity. At a zero e^{z} = 3z^5, so F'(z) = 15z^4 - e^{z} = 3z^4(5 - z), which is not 0: z \ne 0 as F(0) = -1, and z \ne 5. So the five solutions are simple and distinct.
For real x \le 0, 3x^5 \le 0 < e^{x}. For x > 0 the equation reads \ln 3 + 5\ln x - x = 0, whose left side has derivative \displaystyle \frac5x - 1 > 0 on (0, 1); with F(0) < 0 < 3 - e = F(1) there is exactly one real solution, near 0.976. As F(\overline{z}) = \overline{F(z)}, the other four form two conjugate pairs.
A short proof: the local mapping theorem
Let f be holomorphic on an open set G, a \in G, w_0 = f(a), and suppose f is not constant on any disk around a. Prove: (a) f - w_0 has a zero of some finite order m \ge 1 at a, and there are r, \delta > 0 such that for 0 < |w - w_0| < \delta the equation f(z) = w has exactly m solutions in D(a, r), all simple; (b) if G is a region and f is not constant, f(U) is open for every open U \subseteq G (the open mapping theorem).
SolutionProof. (a) By Result 6 of Holomorphic functions are power series, a is a zero of f - w_0 of finite order m \ge 1, with no other zero on some D(a, r_1) \setminus \{a\}. If f' were 0 on a disk around a, f would be constant there (Result 5 of Complex differentiability and the Cauchy-Riemann equations), so likewise f' \ne 0 on some D(a, r_2) \setminus \{a\}. Take 0 < r < \min(r_1, r_2) with \overline{D}(a, r) \subseteq G, and \delta = \min_{|z - a| = r} |f(z) - w_0| > 0. If |w - w_0| < \delta, then f - w = (f - w_0) + (w_0 - w) with |w_0 - w| < |f(z) - w_0| on C(a, r), so by Rouché’s theorem f - w has m zeros in D(a, r). When w \ne w_0 none of them is a, so f' \ne 0 at each: they are simple and distinct.
(b) Let a \in U. If f were constant on a disk around a, the identity principle (Result 7 of Holomorphic functions are power series) would make it constant on G. So (a) applies, with r so small that D(a, r) \subseteq U, and D(f(a), \delta) \subseteq f(U). \blacksquare
Stabilizing a camera gimbal
Hiro and Mei tune the tilt motor of a camera gimbal. Near level, their model makes the tilt error e(t), in radians with t in seconds, satisfy
where K > 0 is the controller gain. As in Constant-coefficient equations, one order higher, every solution is a combination of terms t^j e^{st} over the roots s of p(s) = s^3 + 2s^2 + 3s + K = 0, and |e^{st}| = e^{t\operatorname{Re} s}; so every error dies out exactly when every root has \operatorname{Re} s < 0. How many roots have \operatorname{Re} s > 0 when K = 4, and when K = 10?
SolutionLet \gamma_R be the positively oriented boundary of \{s : |s| < R, \operatorname{Re} s > 0\}: the semicircle s = Re^{it}, \displaystyle -\frac{\pi}{2} \le t \le \frac{\pi}{2}, then the segment from iR down to -iR. On the imaginary axis
The real part vanishes only at \displaystyle y^2 = \frac{K}{2}, the imaginary part only at \displaystyle y = 0, \pm\sqrt3; for K = 4 and 10 these never meet, so no root is on the axis. For R beyond every root’s modulus, Results 2 and 3 give 2\pi Z = \Delta_{\gamma_R}\arg p, summed over the two pieces.
On the semicircle p(s) = s^3(1 + q(s)) with \displaystyle |q(s)| \le \frac2R + \frac{3}{R^2} + \frac{K}{R^3} = \varepsilon. A continuous argument of p there is 3t + \operatorname{Arg}(1 + q(Re^{it})), whose second term stays within \arcsin\varepsilon of 0, so this piece contributes 3\pi in the limit R \to \infty. As \displaystyle \frac{p(iy)}{-iy^3} \to 1 when |y| \to \infty, the segment’s change tends to the net turning \Delta_{\text{axis}} of p(iy) as y runs from \infty down to -\infty. So 2\pi Z = 3\pi + \Delta_{\text{axis}}.
Follow p(iy) (Figure 2): it starts in the third quadrant near -i and ends in the second near i. For K = 4, as y decreases it meets in turn the negative real axis (\displaystyle y = \sqrt3, value -2), the positive imaginary axis (\displaystyle y = \sqrt2), the positive real axis (y = 0, value 4), the negative imaginary axis and the negative real axis, turning clockwise each time. Its argument runs from \displaystyle -\frac{\pi}{2} to \displaystyle -\frac{7\pi}{2}, so \Delta_{\text{axis}} = -3\pi and Z = 0: the gimbal settles.
For K = 10 it crosses the negative imaginary axis at \displaystyle y = \sqrt5 and the positive one at \displaystyle y = -\sqrt5; in between its real part 10 - 2y^2 is positive, so its argument stays in \displaystyle \left(-\frac{\pi}{2}, \frac{\pi}{2}\right). So \Delta_{\text{axis}} = \pi, 2\pi Z = 4\pi, and Z = 2. As p(x) > 0 for real x \ge 0, the two are a conjugate pair: the error oscillates with growing amplitude, and the gimbal shakes. Exercise 10 finds the critical gain.
Boundaries
Zeros or poles on the loop. Then \displaystyle \frac{f'}{f} is not continuous on \gamma and Result 2 says nothing. In Example 5 with K = 6, p(s) = (s + 2)(s^2 + 3) has the roots \displaystyle \pm i\sqrt3 on the axis.
Meromorphic is needed. An essential singularity inside \gamma breaks Result 2 (Exercise 11).
Z - P, not Z. Zeros and poles cancel (Exercise 13); Rouché’s theorem counts zeros because f and g have no poles.
The strict inequality. On C(0, 1), f(z) = z and g(z) = -1 have |g| = |f|, yet f + g = z - 1 has no zero inside and one on the circle. The inequality is sufficient, not necessary, and says nothing about where inside the zeros are.
Loops that are not simple. For a closed path winding several times, each zero and pole is weighted by its winding number n(\gamma, a_k) or n(\gamma, b_j), when for instance G is a disk. This is stated without proof; it follows from the homology form of Cauchy’s theorem, beyond this course.
Where this goes next
The open mapping theorem of Example 4 reproves The maximum modulus principle: the image of a disk around a contains a disk around f(a). Exercise 15 starts the theory of conformal maps, of which Möbius transformations are the first family. Counting roots in a half-plane, as in Example 5, underlies the Nyquist stability criterion of control engineering.
Exercises
Counting with the argument principle
- Find \displaystyle \frac{1}{2\pi i}\int_{C(0, 3)} \frac{f'(z)}{f(z)}\,dz for \displaystyle f(z) = \frac{(z + 1)^2(z - 4)}{(z^2 + 4)^3}.
- Find \displaystyle \int_{C(0, 7)} \frac{e^{z}}{e^{z} - 1}\,dz.
- Find \displaystyle \int_{C(0, 2)} \frac{3z^2 + 2}{z^3 + 2z - 1}\,dz.
- Find \int_{C(0, 4)} z^2\cot z\,dz with the second formula of Result 2.
Rouché’s theorem
- How many zeros of z^6 - 7z^2 + 2 lie in D(0, 1), and how many in \{z : 1 < |z| < 2\}?
- Show that z + e^{-z} = 3 has exactly one solution with \operatorname{Re} z > 0, and that it is real.
- Show that z^5 + 12z + 1 has exactly one zero in \displaystyle D(0, \frac12) and four in \{z : 1 < |z| < 2\}.
- Show that z = 0 is the only solution of \sin z = 2z in D(0, 1).
The gimbal
- The pan axis of the gimbal in Example 5 has characteristic polynomial p(s) = s^3 + s^2 + 2s + 8. How many of its roots have \operatorname{Re} s > 0? Use the image of the imaginary axis.
- For which K > 0 do all roots of s^3 + 2s^2 + 3s + K have \operatorname{Re} s < 0?
Explaining
- Let f(z) = z^2 e^{1/z}. Find \displaystyle \frac{1}{2\pi i}\int_{C(0, 1)} \frac{f'(z)}{f(z)}\,dz, and explain why it is not Z - P.
- The image of C(0, 2) under a polynomial p misses 0 and 5, winds three times around 0 and once around 5. How many solutions do p(z) = 0 and p(z) = 5 have in D(0, 2), counted with multiplicity?
- Find \displaystyle \frac{1}{2\pi i}\int_{C(0, 1)} \frac{f'(z)}{f(z)}\,dz for \displaystyle f(z) = \frac{4z^2 - 1}{4z^2 + 1}, and explain why the image of the circle does not wind around 0 although f has zeros inside it.
Proofs
- (Symmetric Rouché.) Let \gamma be a positively oriented simple closed piecewise smooth path and f, g holomorphic on an open set G containing \gamma and its inside, with |f(z) - g(z)| < |f(z)| + |g(z)| for every z on \gamma. Prove that f and g have the same number of zeros inside \gamma, and that Result 4 is a special case.
- Let f be holomorphic and one-to-one on an open set G. Prove that f'(z) \ne 0 for every z \in G.
- (Hurwitz.) Let \gamma be a positively oriented simple closed piecewise smooth path, and f_n, f holomorphic on an open set G containing \gamma and its inside, with f_n \to f uniformly on \gamma and f \ne 0 on \gamma. Prove that for all large n, f_n has as many zeros inside \gamma as f.
- Prove with Rouché’s theorem that a polynomial of degree d \ge 1 has exactly d zeros in \mathbb{C}, counted with multiplicity (The fundamental theorem of algebra).
Answers
- -4. The zero -1 (order 2) is inside and 4 is outside; the poles \pm 2i, of order 3 each, are inside. So Z - P = 2 - 6.
- 6\pi i. The integrand is \displaystyle \frac{f'}{f} for f(z) = e^{z} - 1, whose zeros 2\pi ik are simple (f' = 1 there). Inside C(0, 7) lie k = 0, \pm 1, as 2\pi < 7 < 4\pi.
- 6\pi i. The integrand is \displaystyle \frac{f'}{f} for f(z) = z^3 + 2z - 1. On C(0, 2), |2z - 1| \le 5 < 8 = |z^3|, so by Rouché’s theorem all 3 zeros lie inside, none on the circle.
- 4\pi^3 i. Here h(z) = z^2 and f = \sin, whose zeros k\pi are real and simple; inside lie 0 and \pm\pi. The formula gives 2\pi i(0 + \pi^2 + \pi^2).
- 2 and 4. On C(0, 1), |z^6 + 2| \le 3 < 7 = |-7z^2|; on C(0, 2), |-7z^2 + 2| \le 30 < 64 = |z^6|. Neither circle holds a zero.
- Let f(z) = z - 3, g(z) = e^{-z}, and \gamma_R as in Example 5 with R > 4. For \operatorname{Re} z \ge 0, |g(z)| \le 1, while |iy - 3| \ge 3 on the axis and |z - 3| \ge R - 3 > 1 on the semicircle. So f + g has one zero inside every such \gamma_R, like f: exactly one solution has \operatorname{Re} z > 0. As \varphi(x) = x - 3 + e^{-x} has \varphi(0) = -2 < 0 < e^{-3} = \varphi(3), it is real, near 2.948.
- On \displaystyle C(0, \frac12), \displaystyle |z^5 + 1| \le 1 + \frac{1}{32} < 6 = |12z|; on C(0, 1), |z^5 + 1| \le 2 < 12; on C(0, 2), |12z + 1| \le 25 < 32 = |z^5|. So one zero lies in \displaystyle D(0, \frac12), still one in D(0, 1), all five in D(0, 2), none on the circles, and four in the annulus.
- On C(0, 1), \displaystyle |\sin z| \le \sum_{k \ge 0} \frac{1}{(2k+1)!} = \sinh 1 < 2 = |2z|, termwise from the series of \sin z. Rouché’s theorem with f(z) = 2z gives 2z - \sin z exactly one zero in D(0, 1), and 0 is one.
- 2. Here p(iy) = (8 - y^2) + iy(2 - y^2). The real part is positive for \displaystyle |y| < 2\sqrt2, which contains the zeros \displaystyle 0, \pm\sqrt2 of the imaginary part, so as for K = 10 in Example 5, \Delta_{\text{axis}} = \pi and 2\pi Z = 4\pi. Indeed p(s) = (s + 2)(s^2 - s + 4), with roots -2 and \displaystyle \frac{1 \pm i\sqrt{15}}{2}.
- 0 < K < 6. For K < 6 the real part of p(iy) vanishes at \displaystyle y^2 = \frac{K}{2} < 3, so p(iy) meets the negative real axis at \displaystyle y = \sqrt3 (value K - 6) and then the axes in the order of K = 4: Z = 0, no root on the axis. For K > 6, \displaystyle \frac{K}{2} > 3 and, as for K = 10, Z = 2. For K = 6 the roots \displaystyle \pm i\sqrt3 lie on the axis.
- 2. By Result 1, \displaystyle \frac{f'}{f} = \frac2z - \frac{1}{z^2}, with integral 2\cdot 2\pi i. But f has no zeros or poles in D(0, 1), and 0 is essential: \displaystyle z^2 e^{1/z} = \sum_{n \ge 0} \frac{z^{2-n}}{n!} has infinitely many negative powers (Result 6 of Laurent series and singularities). So f is not meromorphic there and Result 2 does not apply; the 2 comes from the factor z^2.
- Three and one. By Results 2 and 3 the number of zeros of p in D(0, 2) is the winding number 3. For p - 5, with \sigma = p \circ \gamma, \displaystyle n(\sigma - 5, 0) = \frac{1}{2\pi i}\int \frac{\sigma'(t)}{\sigma(t) - 5}\,dt = n(\sigma, 5) = 1.
- 0. The zeros \displaystyle \pm\frac12 and the poles \displaystyle \pm\frac{i}{2} are simple and inside, so Z - P = 2 - 2, and by Result 3 the image winds 0 times: the zeros add two turns and the poles take two away.
- Proof. If f(z) = 0 on \gamma, the inequality reads |g(z)| < |g(z)|; likewise for g. So both are nonzero on an open U \supseteq \gamma, where \displaystyle \varphi = \frac{f}{g} is holomorphic and nonzero and \displaystyle \frac{f'}{f} = \frac{g'}{g} + \frac{\varphi'}{\varphi} (Result 1). Dividing by |g|, |\varphi - 1| < |\varphi| + 1 on \gamma, which fails for \varphi = -t, t \ge 0. So \varphi \circ \gamma lies in \mathbb{C} \setminus (-\infty, 0], where \operatorname{Log} is an antiderivative of \displaystyle \frac1w, and as in the proof of Result 4, \displaystyle \int_\gamma \frac{\varphi'}{\varphi}\,dz = 0. So \displaystyle \int_\gamma \frac{f'}{f}\,dz = \int_\gamma \frac{g'}{g}\,dz, and Result 2, applied as in that proof, gives Z(f) = Z(g). Result 4 is the case of f and f + g: |f - (f + g)| = |g| < |f| \le |f| + |f + g|. \blacksquare
- Proof. Let a \in G. A one-to-one f is constant on no disk, so by Example 4(a) f - f(a) has a zero of finite order m at a, and some w near f(a) has m distinct preimages near a. As f is one-to-one, m = 1, so f(z) - f(a) = (z - a)g(z) with g(a) = f'(a) \ne 0 (Result 6 of Holomorphic functions are power series). \blacksquare The converse fails: (e^{z})' \ne 0, yet e^{z} = e^{z + 2\pi i}.
- Proof. |f| is continuous and positive on the compact set \gamma, so \mu = \min_{z \in \gamma}|f(z)| > 0. By uniform convergence there is N with |f_n(z) - f(z)| < \mu \le |f(z)| on \gamma for n \ge N. For such n, Rouché’s theorem with g = f_n - f shows that f + g = f_n has as many zeros inside \gamma as f. \blacksquare
- Proof. Write p(z) = a_d z^d + q(z) with a_d \ne 0, q(z) = a_{d-1}z^{d-1} + \cdots + a_0 and S = |a_0| + \cdots + |a_{d-1}|. For |z| = \rho \ge 1, |q(z)| \le S\rho^{d-1}. Fix \displaystyle \rho > \max\left(1, \frac{S}{|a_d|}\right). On C(0, \rho), |q(z)| \le S\rho^{d-1} < |a_d|\rho^d = |a_d z^d|, so by Rouché’s theorem p has as many zeros in D(0, \rho) as a_d z^d, namely d. For |z| = \rho' \ge \rho the same bound gives |p(z)| \ge |a_d|\rho'^d - S\rho'^{d-1} > 0, so no zero lies outside. \blacksquare