Argument Principle & Rouché's Theorem

Derivation

Argument Principle & Rouché's Theorem

D-185 Home PU-205 Threads Depends on The Residue Theorem

Statement

Let \(f\) be meromorphic on a domain containing a positively-oriented simple closed contour \(C\) and its interior, with no zeros or poles lying on \(C\). Then the contour integral of the logarithmic derivative counts the enclosed zeros minus the enclosed poles, each with multiplicity: \(\dfrac{1}{2\pi i}\oint_C \dfrac{f'(z)}{f(z)}\,dz = Z - P\). As a corollary (Rouché), if \(f\) and \(g\) are holomorphic on and inside \(C\) with \(|f(z)-g(z)| < |f(z)|\) everywhere on \(C\), then \(f\) and \(g\) enclose the same number of zeros.

Why it matters

The argument principle turns a hard question — "how many roots does this equation have in this region?" — into a single contour integral, and identifies that integer with the winding number of the image curve \(f(C)\) about the origin. It is the analytic engine behind root-counting, the Nyquist stability criterion in control theory, and the spectral counting of resonances and poles of scattering amplitudes.

Rouché's theorem is its most-used consequence: it lets you locate roots of a complicated function by comparing it with a simpler dominant term, and it gives a two-line proof of the Fundamental Theorem of Algebra. In physics it underpins mode-counting for waveguides, level counting for quantum wells via the Jost function, and stability boundaries of dynamical systems.

Assumptions

\(f\) is meromorphic on an open set containing \(\overline{\text{int}(C)}\).If \(f\) had an essential singularity inside \(C\), the residue calculation of \(f'/f\) breaks down and the integral need not be an integer multiple of \(2\pi i\).

\(f\) has no zero and no pole on the contour \(C\) itself.A zero or pole on \(C\) makes \(f'/f\) non-integrable there (the integrand blows up on the path), so the integral is undefined rather than merely shifted.

\(C\) is a simple, closed, rectifiable, positively-oriented (counter-clockwise) Jordan curve.Reversing orientation flips the sign to \(P-Z\); a self-intersecting or multiply-wound curve replaces the count by a sum weighted by the winding number \(n(C,z_0)\) about each singular point.

Zeros and poles are counted with multiplicity/order.Drop this and the identity fails at every repeated root: a double zero contributes \(2\), not \(1\), because its residue of \(f'/f\) equals its order.

(Rouché) The strict inequality \(|f-g|<|f|\) holds at every point of \(C\).If equality \(|f-g|=|f|\) is attained anywhere, \(g\) may vanish on \(C\) or the homotopy \(f+t(g-f)\) can pass through \(0\), and the two functions may enclose different numbers of zeros.

Derivation

\[ f(z) = (z-a)^{m}\,g(z), \qquad g \text{ holomorphic near } a,\ g(a)\neq 0 \]

A zero of \(f\) of order \(m\) at \(a\) factors out its vanishing; \(g\) collects the non-vanishing remainder. Legal because a holomorphic function has an isolated zero of definite finite order. A

\[ \frac{f'(z)}{f(z)} = \frac{m}{z-a} + \frac{g'(z)}{g(z)}, \qquad \operatorname{Res}_{z=a}\frac{f'}{f} = m \]

Differentiate the product \(f=(z-a)^m g\) and divide by \(f\); \(g'/g\) is holomorphic at \(a\) since \(g(a)\neq 0\), so the only singular part is the simple pole with residue \(m\). B

\[ f(z) = (z-b)^{-p}\,h(z),\quad h(b)\neq 0 \;\Longrightarrow\; \frac{f'}{f} = \frac{-p}{z-b} + \frac{h'}{h},\quad \operatorname{Res}_{z=b}\frac{f'}{f} = -p \]

Same manipulation for a pole of order \(p\): the local exponent is \(-p\), so the residue of the logarithmic derivative is \(-p\). This is the only place poles enter. B

\[ \frac{1}{2\pi i}\oint_C \frac{f'}{f}\,dz = \sum_{\text{zeros } a_j} m_j \;+\; \sum_{\text{poles } b_k} (-p_k) \]

Apply the residue theorem (prior result): the integral equals \(2\pi i\) times the sum of residues of \(f'/f\) at all its singularities inside \(C\), which are exactly the zeros and poles of \(f\). B

\[ \boxed{\;\frac{1}{2\pi i}\oint_C \frac{f'(z)}{f(z)}\,dz = Z - P\;} \]

Collect the sums: \(Z=\sum_j m_j\) is the total zero count and \(P=\sum_k p_k\) the total pole count, each with multiplicity. This is the argument principle. A

\[ \frac{f'}{f} = \frac{d}{dz}\log f \;\Longrightarrow\; \frac{1}{2\pi i}\oint_C \frac{f'}{f}\,dz = \frac{1}{2\pi}\,\Delta_C\!\arg f = n\big(f(C),0\big) \]

The integrand is the derivative of \(\log f = \ln|f| + i\arg f\); around a closed loop \(\ln|f|\) returns to its value while \(\arg f\) changes by \(2\pi\) times the winding number of the image curve \(f(C)\) about the origin. Hence "argument principle." C

\[ \text{(Rouché) } |f-g| < |f| \text{ on } C \;\Longrightarrow\; F := \frac{g}{f},\qquad |F-1| = \frac{|g-f|}{|f|} < 1 \text{ on } C \]

Assume \(f,g\) holomorphic with no zeros of \(f\) on \(C\) (guaranteed since \(|f|>|f-g|\ge 0\)). Divide the hypothesis by \(|f|\); the ratio \(F\) is well-defined and meromorphic near \(C\). B

\[ F(C) \subset \{\,w : |w-1|<1\,\} \;\Rightarrow\; 0 \notin F(C),\quad n\big(F(C),0\big)=0 \]

The image of \(C\) lies in the open disk of radius \(1\) centered at \(w=1\), which does not contain the origin. On this simply-connected disk a holomorphic branch of \(\log w\) exists, so \(F'/F=(\log F)'\) is an exact differential and its loop integral vanishes: the image cannot wind around \(0\). C

\[ 0 = \frac{1}{2\pi i}\oint_C \frac{F'}{F}\,dz = \frac{1}{2\pi i}\oint_C\!\left(\frac{g'}{g}-\frac{f'}{f}\right)dz = Z_g - Z_f \]

Since \(F=g/f\), the logarithmic derivative splits as \(F'/F=g'/g-f'/f\). Each term is an argument-principle integral; \(f,g\) are holomorphic so they have no poles (\(P=0\)), leaving the zero counts. B

\[ \boxed{\,Z_f = Z_g\,}\qquad\text{(number of zeros inside } C \text{, with multiplicity)} \]

Rearrange step 9. Two functions that stay closer to each other than to zero on the boundary enclose the same number of roots inside. A

Result

\[ \frac{1}{2\pi i}\oint_C \frac{f'(z)}{f(z)}\,dz = Z - P, \qquad |f-g|<|f| \text{ on } C \;\Rightarrow\; Z_f = Z_g \]

Reading. The contour integral of the logarithmic derivative of \(f\) equals the net number of zeros minus poles trapped inside the loop — equivalently, the number of times the image curve \(f(C)\) circles the origin. Rouché says that if a "perturbation" \(g-f\) is uniformly smaller in magnitude than \(f\) itself all along the boundary, the perturbation cannot create or destroy any enclosed root, so the simpler comparison function \(f\) already tells you the count.

Units check. The problem is dimensionless: \(z\) is a pure complex number, \(f'/f\) has dimension \(1/[\text{length in the }z\text{-plane}]\), and \(dz\) restores it, so the integrand-times-\(dz\) is dimensionless. Dividing by \(2\pi i\) yields the integer \(Z-P\), a pure count with no units — as it must, since counting objects can never carry a physical dimension.

Limiting cases

Holomorphic \(f\) (no poles, \(P=0\)): the integral counts zeros alone, \(\frac{1}{2\pi i}\oint_C f'/f\,dz = Z\).
\(f\) constant and non-zero: \(f'=0\), integrand vanishes, \(Z-P=0\) — no zeros or poles anywhere.
Single simple zero enclosed: integral \(=1\); \(f(C)\) winds once positively about the origin.
\(g\) dominates on a large circle \(|z|=R\to\infty\) for a degree-\(n\) polynomial: comparison with the leading term \(a_n z^n\) gives exactly \(n\) roots — the Fundamental Theorem of Algebra.
Rouché with \(|g|\ll|f|\): \(g\) is a negligible correction; the root count is fixed entirely by \(f\).

Breaks when

A zero or pole sits on \(C\). The integrand \(f'/f\) has a non-integrable singularity on the path; the integral diverges and neither \(Z-P\) nor Rouché is defined. Any counting statement requires deforming \(C\) off the singularity first.
The Rouché inequality degrades to equality \(|f-g|=|f|\) somewhere on \(C\). Then \(F=g/f\) can reach the boundary circle \(|w-1|=1\) and touch \(w=0\), i.e. \(g\) may vanish on \(C\); the winding argument fails and the two counts can differ.
\(f\) has an essential singularity inside \(C\). By the Casorati–Weierstrass / Picard behavior there is no finite-order residue for \(f'/f\), infinitely many zeros may accumulate, and the integral need not equal any integer.
The curve is not simple / winds multiple times. The count generalizes to \(\sum_j n(C,a_j)\,m_j - \sum_k n(C,b_k)\,p_k\); the bare "inside/outside" reading of \(Z-P\) no longer applies.

Failure modes

Forgetting multiplicity. Counting a triple root as one zero instead of three; the residue of \(f'/f\) at an order-\(m\) zero is \(m\), not \(1\).
Sign error on poles. Adding pole orders to \(Z\) instead of subtracting them; the result is \(Z-P\), and each pole of order \(p\) contributes \(-p\).
Wrong Rouché split. Choosing \(f\) as the small term and \(g\) as the dominant term so that \(|f-g|<|f|\) is violated; you must let \(f\) be the term that is strictly largest on \(C\).
Checking the inequality at only one point of \(C\). \(|f-g|<|f|\) must hold at every boundary point; verifying it at a single \(z\) proves nothing.
Applying it across a pole on the contour. Using Rouché or the argument principle on a circle that passes through a singularity, instead of shrinking or enlarging the radius to clear it.
Confusing the two Rouché forms. Mixing up \(|f-g|<|f|\) (symmetric, modern) with \(|h|<|f|\Rightarrow f\) and \(f+h\) agree (additive); both are correct but the comparison function differs.

Discussion

The deepest content of the argument principle is the identification of an analytic integral with a topological integer. The left-hand side is a smooth contour integral that could, a priori, take any value; the right-hand side is forced to be a whole number because it is the degree of the map \(f:C\to\mathbb{C}\setminus\{0\}\), i.e. the winding number of \(f(C)\) about the origin. This is the one-dimensional ancestor of the index theorems that pervade modern physics — the same "count = integral of a density" logic that gives the Chern number of a Berry curvature, the number of Landau levels, and topological invariants of band insulators.

Rouché follows because winding number is a homotopy invariant. The straight-line homotopy \(f_t = f + t(g-f)\), \(t\in[0,1]\), never vanishes on \(C\) precisely when \(|g-f|<|f|\), so \(f_0=f\) and \(f_1=g\) have image curves that can be continuously deformed into one another without crossing the origin. Since the winding number cannot jump during a continuous deformation, the enclosed zero counts of \(f\) and \(g\) coincide. Read this way, Rouché is nothing but "the degree of a map is a homotopy invariant."

The physical incarnations are direct. In control theory the Nyquist criterion counts unstable closed-loop poles as the number of times the frequency response \(G(i\omega)\) encircles the critical point — literally the argument principle applied to \(1+G\). In scattering theory, Levinson's theorem relates the total phase change of the \(S\)-matrix (an argument-principle integral of \(\det S\)) to the number of bound states, and resonance-counting for the Jost function uses the same identity. In numerical analysis, the argument principle is the basis of contour-integral eigenvalue solvers (Beyn's method, FEAST) that count and locate eigenvalues inside a region without ever forming a characteristic polynomial.

A subtle rigour point deserves emphasis: step 8 relies on the image lying in a simply-connected region avoiding the origin, which guarantees a single-valued holomorphic logarithm and hence an exact \(F'/F\). This is where the strictness of \(|f-g|<|f|\) is load-bearing — it is exactly the condition that keeps \(F=g/f\) inside the punctured-plane region where \(\log\) is well-defined. Relaxing to non-strict inequality re-admits the boundary point \(w=0\) into the closure of the image, and the homotopy \(f_t\) can then pass through a zero on \(C\), invalidating the invariance. This is the same phenomenon that makes degree theory delicate at the boundary of the admissible class of maps.

Common misconceptions. The integral does not count zeros and poles separately — only their signed difference \(Z-P\); a function with equally many zeros and poles inside \(C\) gives integral zero even though both are present. And Rouché does not tell you where the roots are, only how many lie inside \(C\); localization requires choosing nested contours and differencing the counts.

Worked examples

Roots of \(p(z)=z^5+3z+1\) inside the unit disk \(|z|<1\). Choose the comparison function as the term that dominates on \(|z|=1\). A

\[ f(z) = 3z, \qquad g(z) = z^5+3z+1, \qquad g-f = z^5+1 \]

Bound the difference on the boundary \(|z|=1\) and compare with \(|f|\). B

\[ |g-f| = |z^5+1| \le |z|^5 + 1 = 2, \qquad |f| = |3z| = 3, \qquad 2 < 3 \checkmark \]

Rouché applies: \(g\) and \(f\) have equal zero counts inside \(|z|<1\). The comparison \(f=3z\) has a single simple zero at \(z=0\). A

\[ Z_g = Z_f = 1 \]

\[ z^5+3z+1 \text{ has exactly } \mathbf{1} \text{ root in } |z|<1. \]

Reading. Along the unit circle the linear term \(3z\) beats everything else, so the quintic behaves topologically like \(3z\): its image winds once about the origin, trapping one root. The other four roots lie in \(|z|\ge 1\).

Units check. Dimensionless throughout; the answer is the pure integer \(1\).

Net count for the meromorphic \(f(z)=\dfrac{(z-1)^2}{z+2}\) on \(|z|=3\). Identify the enclosed zeros and poles with multiplicity. A

\[ \text{zero: } z=1 \ (\text{order } 2), \quad |1|=1<3; \qquad \text{pole: } z=-2 \ (\text{order } 1),\quad |-2|=2<3 \]

Both singular points lie inside the circle of radius \(3\). Apply the argument principle. B

\[ Z = 2, \qquad P = 1, \qquad \frac{1}{2\pi i}\oint_{|z|=3}\frac{f'(z)}{f(z)}\,dz = Z-P = 2-1 = 1 \]

Interpret as a winding number: the image \(f(|z|=3)\) encircles the origin once, net. C

\[ \oint_{|z|=3}\frac{f'}{f}\,dz = 2\pi i\,(2-1) = 2\pi i \]

\[ \frac{1}{2\pi i}\oint_{|z|=3}\frac{f'(z)}{f(z)}\,dz = 1 \]

Reading. The double zero contributes \(+2\) and the simple pole contributes \(-1\), for a net winding number of \(+1\). The double zero counts twice precisely because the residue of \(f'/f\) at an order-\(m\) zero equals \(m\).

Units check. Pure numbers: \(Z=2\), \(P=1\), and \(Z-P=1\) is a dimensionless integer.

Problems

Evaluate \(\dfrac{1}{2\pi i}\displaystyle\oint_{|z|=2}\dfrac{f'(z)}{f(z)}\,dz\) for \(f(z)=\dfrac{(z-1)(z-i)}{z-5}\).
Solution
Zeros of \(f\): \(z=1\) (\(|1|=1<2\)) and \(z=i\) (\(|i|=1<2\)), both inside \(|z|=2\), each order \(1\), so \(Z=2\). Pole: \(z=5\), \(|5|=5>2\), outside, so \(P=0\). By the argument principle the integral \(=Z-P = 2-0 = \boxed{2}\).
How many roots does \(z^4+z^3-4z+1\) have in \(|z|<1\)? Use Rouché.
Solution
Take \(f(z)=-4z\) (dominant on \(|z|=1\)) and \(g(z)=z^4+z^3-4z+1\). Then \(g-f=z^4+z^3+1\), so on \(|z|=1\), \(|g-f|\le |z|^4+|z|^3+1=3\), while \(|f|=|{-4z}|=4\). Since \(3<4\), Rouché gives \(Z_g=Z_f\). The comparison \(-4z\) has one simple zero (at \(0\)) inside, so \(\boxed{1}\) root of the quartic lies in \(|z|<1\).
Show that every zero of \(p(z)=z^7-5z^3+12\) lies in the annulus \(1\le |z|<2\).
Solution
No zeros in \(|z|<1\): on \(|z|=1\), \(|z^7-5z^3|\le |z|^7+5|z|^3 = 6 < 12 = |12|\), so comparing \(p\) with the constant \(12\) (which has no zeros) gives \(0\) zeros inside \(|z|<1\). All seven zeros in \(|z|<2\): on \(|z|=2\), \(|{-5z^3+12}| \le 5\cdot 8 + 12 = 52 < 128 = |z^7|\), so comparing \(p\) with \(z^7\) (seven zeros at the origin, with multiplicity) gives \(7\) zeros inside \(|z|<2\). Since none are in \(|z|<1\) and all \(7=\deg p\) are in \(|z|<2\), all seven lie in the annulus \(1\le|z|<2\). \(\blacksquare\)
Count the zeros of \(z^5+15z+1\) in the annulus \(\tfrac32<|z|<2\).
Solution
On \(|z|=2\): compare with \(f=z^5\): \(|15z+1|\le 30+1=31 < 32 = |z^5|\), so \(5\) zeros lie in \(|z|<2\). On \(|z|=\tfrac32\): compare with \(f=15z\): \(|z^5+1|\le (\tfrac32)^5+1 = 7.59\ldots+1 = 8.59\ldots < 22.5 = |15z|\), so \(1\) zero lies in \(|z|<\tfrac32\). The annulus therefore contains \(5-1=\boxed{4}\) zeros.
Prove the Fundamental Theorem of Algebra: a polynomial \(p(z)=a_nz^n+\cdots+a_0\) with \(a_n\neq 0\) and \(n\ge 1\) has exactly \(n\) roots in \(\mathbb{C}\), counted with multiplicity.
Solution
Let \(f(z)=a_nz^n\) and \(g(z)=p(z)\), so \(g-f=a_{n-1}z^{n-1}+\cdots+a_0\). On the circle \(|z|=R\), \[ |g-f| \le \sum_{k=0}^{n-1}|a_k|R^k, \qquad |f| = |a_n|R^n. \] Dividing, \(\dfrac{|g-f|}{|f|} \le \dfrac{1}{|a_n|}\sum_{k=0}^{n-1}\dfrac{|a_k|}{R^{\,n-k}} \to 0\) as \(R\to\infty\). Hence for all sufficiently large \(R\), \(|g-f|<|f|\) on \(|z|=R\), and Rouché gives \(Z_p = Z_f\) inside \(|z|<R\). The comparison \(f=a_nz^n\) has a zero of order \(n\) at the origin, so \(Z_f=n\). Every root of \(p\) has modulus bounded (e.g. by \(1+\max_k|a_k/a_n|\)), so taking \(R\) beyond this bound captures all of them: \(p\) has exactly \(n\) roots, counted with multiplicity. \(\blacksquare\)