The Argument Principle and Rouche's Theorem

The Nyquist stability criterion in control engineering is a direct application of the argument principle: the number of unstable closed-loop poles is read off from the winding number of a transfer function's image curve around $-1$, a translation of exactly the topological counting idea here into engineering practice. In numerical analysis, root-isolation algorithms (e.g., for finding all zeros of a polynomial in a given disk) use Rouché-type bounds to certify how many roots lie in a candidate region before running expensive iterative solvers. In the study of dynamical systems and complex dynamics, counting preimages and periodic points of holomorphic maps (e.g., in the theory of the Mandelbrot set and Julia sets) relies on the argument principle to count solutions of $f(z) = w$ locally, connecting directly to the local behavior exploited in Conformal Mappings, where the argument principle underlies the proof that non-constant holomorphic maps are locally $z\mapsto z^n$ up to conformal change of coordinates. In algebraic geometry and topology, the argument principle is the one-complex-variable shadow of the general theory of degree of a map and intersection numbers. It also gives the cleanest possible proof of the Fundamental Theorem of Algebra: compare $p(z) = a_nz^n + \cdots$ with $f(z) = a_n z^n$ on a huge circle, Rouché forces exactly $n$ roots inside, no more, no less.

Ask the tutor: How do I extend Rouché's theorem to count zeros when $f$ and $g$ have common poles inside the contour?

Think of $f$ as a machine that takes the boundary curve $\gamma$ and outputs another curve $f(\gamma)$ in the target plane. As $z$ traverses $\gamma$ once counterclockwise, $w = f(z)$ traces some closed loop. The argument principle says: count the zeros of $f$ inside $\gamma$ (minus poles), and that number equals how many times $f(\gamma)$ wraps around $0$. Why should this be true? Near a simple zero $a$, $f(z) \approx c(z-a)$ for constant $c\neq0$, so as $z$ makes one small loop around $a$, the argument of $f(z)$ increases by $2\pi$ — one extra wind around $0$ contributed locally. A zero of order $m$ contributes $m$ winds (locally $f(z)\approx c(z-a)^m$, and the map $\zeta \mapsto \zeta^m$ wraps $m$ times as $\zeta$ winds once). Poles contribute windings of the opposite sign, because $1/(z-b)^m$ winds $m$ times clockwise as $z$ circles $b$ once. Globally, deforming $\gamma$ to small circles around each zero/pole (justified by contour deformation, since $f$ has no other singularities in between) adds up all these local contributions — this is exactly the residue-sum structure. Rouché's inequality $|g| < |f|$ on $\gamma$ guarantees $f(\gamma)$ and $(f+g)(\gamma)$ stay "on the same side" of the origin at every point — literally, $|f + g - f| < |f|$ means $f+g$ never crosses through $0$ as we linearly interpolate from $f$ to $f+g$, so the winding number cannot jump.

Now we promote the argument principle into a comparison tool.

Example 1 (asymmetric polynomial, Rouché on a circle). How many roots does

have inside the disk $|z| < 2$? Take $\gamma = \{|z| = 2\}$. Split $p = f + g$ with $f(z) = z^7$ (the dominant term for large $|z|$) and $g(z) = -5z^3 + 12$. On $|z| = 2$:

Since $52 < 128$, Rouché's hypothesis $|g| < |f|$ holds on $|z|=2$. So $p = f+g$ has the same number of zeros in $|z|<2$ as $f(z) = z^7$, which has a single zero of order $7$ at the origin. Hence $p$ has exactly $7$ roots (with multiplicity) in $|z|<2$ — i.e. all its roots lie in this disk (consistent with $p$ having degree $7$). Now refine: how many roots lie in $|z| < 1$? On $|z|=1$, try $f(z) = -5z^3$ (guess the dominant term on this smaller circle) and $g(z) = z^7 + 12$:

Here $13 > 5$, so this split fails — the constant term dominates on $|z|=1$. Try instead $f(z) = 12$ (constant) and $g(z) = z^7 - 5z^3$:

Now $6 < 12$, so Rouché applies with this split: $p$ has the same number of zeros in $|z|<1$ as $f(z)=12$, a nonvanishing constant, namely $0$ roots in $|z| < 1$. Conclusion: all $7$ roots of $p(z) = z^7 - 5z^3+12$ lie in the annulus $1 \le |z| < 2$ — in fact strictly in $1 < |z| < 2$, since we've shown $p$ has no zeros on $|z|=1$ or $|z|=2$ (both circles were used as Rouché contours, which requires no zero lies exactly on them).

Example 2 (transcendental equation, argument principle directly). Consider $f(z) = z e^{z} - 1$ inside the disk $|z| \le 1$, on the boundary circle $\gamma(t) = e^{it}$, $t\in[0,2\pi]$. We are asked: how many solutions does $ze^z = 1$ have with $|z|\le 1$? Try Rouché with $g(z) = -1$ and "$f$" $= ze^z$ (renaming to avoid clash, call the full function $F = ze^z - 1$, split as $h(z) = ze^z$, $k(z) = -1$, $F = h + k$). We need $|k| < |h|$ on $|z|=1$, i.e. $|ze^z| > 1$ there. But on $|z|=1$, writing $z = e^{it}$, we get $|z e^z| = |e^{it}e^{\cos t + i\sin t}| = e^{\cos t}$, which ranges over $[e^{-1}, e]$ and dips below $1$ when $\cos t < 0$. So this split fails — $|h|$ is not uniformly bigger than $|k|=1$ on the whole circle. Instead we directly use the argument principle: compute

We track the winding of $w(t) = F(e^{it}) = e^{it}e^{e^{it}} - 1$ as $t: 0 \to 2\pi$. Rather than integrate directly, use a refined Rouché split valid on sub-arcs, or — more robust — evaluate at sample points and use the argument principle's integer-valued continuity (as in Rouché's proof) to interpolate. We instead give a clean rigorous route: apply Rouché on a slightly smaller circle where the estimate does work, then use continuity of $Z$ in the radius. Take $|z| = r$ for $r$ close to $1$ but consider instead splitting differently: $h(z) = ze^z$, $k(z) = -1$ fails at $t=\pi$ exactly ($z=-1$: $|h(-1)| = |-e^{-1}| = e^{-1} < 1$). So try the disk of radius $r_0$ with $e^{-r_0} \cdot 1 > 1$ impossible for any $r_0>0$ since $e^{-r_0}<1$ always — meaning $ze^z=1$ can have solutions arbitrarily close to where $|ze^z|$ dips low, so no single split works on the full circle $|z|=1$ by this method; we must integrate the winding number directly (which is the honest thing to do rather than force Rouché where it does not apply). Numerically trace $w(t) = e^{it}e^{\cos t + i \sin t} - 1 = e^{\cos t}\, e^{i(t+\sin t)} - 1$. Its argument-of-shifted-origin is controlled by $\theta(t) = t + \sin t$ (the phase of $e^{it}e^{e^{it}}$, before subtracting $1$) together with the modulus $\rho(t) = e^{\cos t}$, which stays in $[e^{-1},e]\subset(0.36,2.72)$, always positive, so $e^{it}e^{e^{it}}$ itself winds around $0$ exactly once as $t:0\to2\pi$ (since $\theta(t) = t+\sin t$ is monotonically... check: $\theta'(t) = 1+\cos t \ge 0$, vanishing only at $t=\pi$, so $\theta$ is non-decreasing on $[0,2\pi]$ and $\theta(2\pi)-\theta(0) = 2\pi$; hence $e^{it}e^{e^{it}}$ winds exactly once around $0$). Subtracting $1$ shifts the curve; since $\rho(t) \ge e^{-1} \approx 0.368$ is bounded well away from where a unit shift could remove the single loop around $0$ (the curve $\rho(t)e^{i\theta(t)}$ traces a loop of modulus between $0.368$ and $2.72$ winding once around $0$; subtracting the constant $1$, which has modulus less than the loop's minimum radius $0.368$? — no, $1 > 0.368$, so we must be careful). This shows genuinely that a rigorous conclusion here needs either numerical winding-number computation or a cleverer split — which is itself an instructive lesson: not every problem yields to a one-line Rouché split, and one sometimes must fall back on the full argument principle (numerically evaluating the contour integral, or using a computer algebra / numerical winding-number count) — precisely the more laborious but always-valid tool the argument principle provides beneath Rouché's shortcut. For this specific equation, it is classically known (from the theory of the Lambert $W$ function) that $ze^z=1$ has exactly one solution in $|z|<1$, namely $z = W(1) \approx 0.5671$ (the Omega constant), which one confirms lies in $|z|<1$; a careful numerical integration of $Z$ above (or a finer Rouché split on an asymmetric sub-region) confirms $Z=1$.