Counting before solving
A polynomial equation can be difficult to solve exactly while its roots are easy to count in a region. For instance, the dominant term on a circle may control how many roots lie inside, even when no convenient formula for the roots exists. Complex analysis turns that observation into a theorem by following the image of the boundary around the origin.
Let . Let be a positively oriented, piecewise simple closed curve, and let denote its bounded interior. The curve and its interior are assumed to lie in an open set where the functions under discussion have the stated regularity. Zeros and poles are counted with multiplicity. We use the residue theorem as a prerequisite; the zero-counting argument below explains how its local residues become a global count.
Logarithmic derivatives and the count
Definition 1.8.1
(Logarithmic derivative and winding number about )
Let be meromorphic and not identically zero on a connected open set. Its logarithmic derivative is the meromorphic function . If is a closed piecewise curve that does not pass through , its winding number about is
The name reflects the formula wherever a branch of exists, but itself is single-valued and needs no branch. Substituting shows that whenever has no zero or pole on .
Theorem 1.8.2
(Argument principle)
Let be meromorphic on a neighborhood of , with no zero or pole on and not identically zero. If and are the numbers of zeros and poles in , counted with multiplicity, then
Proof
(Residues of the logarithmic derivative)
Near a zero of order , write with holomorphic and . Then
and is holomorphic near , so the residue is . Near a pole of order , the same factorization with exponent gives residue . Away from zeros and poles, is holomorphic.
Work in the connected component of the neighborhood containing . There is not identically zero because it has no boundary zeros. Zeros cannot accumulate at a regular point by the identity theorem, nor at a pole where ; poles are isolated. Compactness of therefore gives finitely many zeros and poles in , none of them on . The positively oriented simple closed curve has winding number one about each interior point, so the residue theorem gives . The last equality in the statement is the substitution .
Proof X-ray
Where each hypothesis enters the residue count
The decisive move is that the logarithmic derivative turns a product into a sum: the multiplicative factor becomes the additive term , and the nonvanishing factor contributes only the holomorphic term . So the order of a zero or pole is read off as a residue, with the sign of the exponent.
The hypotheses are used at three separate points. No zero or pole on makes continuous on the path, so the integral exists; for on the unit circle the integrand is not integrable near . The condition that is not identically zero gives isolated zeros, and compactness turns isolation into a finite list, so the residue theorem has finitely many terms. Positive orientation and simplicity of give winding number exactly about every interior singularity, so each order is counted once and with the right sign. The proof never locates a zero; it only adds integers.
Along , the integral is the winding number of the closed image curve about zero. That curve can cross itself; its winding number still counts signed turns and need not equal the number of geometric loops visible in a picture. The absence of boundary zeros and poles is essential: otherwise the logarithmic derivative is singular on the path and this formula is not defined as stated.
Common mistake
(A zero integral does not mean there are no zeros)
For a meromorphic the integral counts , not . For on the integral is , although has a zero at inside the circle. Only for holomorphic , where , does the integral count zeros alone.
Rouché's theorem
Theorem 1.8.3
(Rouché's theorem)
Suppose and are holomorphic on a neighborhood of and
Then and have the same number of zeros in , counted with multiplicity.
Proof
(Deforming into without crossing zero)
For put . On the boundary,
Thus every avoids zero. The winding number is integer-valued and varies continuously with : on the compact set , the denominators stay uniformly away from zero, so the logarithmic-derivative integrals vary continuously. A continuous integer-valued function on an interval is constant. The argument principle, applied to the holomorphic functions and , identifies the winding numbers at and with the respective zero counts. This proves the claim.
The strict inequality is a boundary condition, not an assertion that is smaller inside. Its uniform margin follows from continuity and compactness of . If alone, can vanish on the boundary and the conclusion need not even be well-defined.
Counterexample mode
Why the boundary inequality must be strict
On the unit circle let and . Then everywhere. The function has one zero in the open disk, while has no zero there: its zero lies on the boundary. Equality cannot simply replace the strict hypothesis.
Worked examples
Worked example 1.8.4
(A quintic in an annulus)
Consider . On , the term has modulus , while . Rouché gives exactly one zero in . On , take ; now and , so there are five zeros in . The reverse triangle inequality also shows is nonzero on both circles. Thus four zeros, counted with multiplicity, lie in .
Worked example 1.8.5
(A small analytic perturbation)
Fix and . For on ,
The linear function has one zero in the disk, so also has one, counted with multiplicity. The bound controls the perturbation on the boundary; it does not give the exact location of that root.
Worked example 1.8.6
(A triple zero splits into three simple zeros)
Let , and . On we have , so has three zeros in , counted with multiplicity, because has the single zero of order . Directly, the zeros of are the three cube roots of . They are distinct because , and each has modulus . So the count agrees with multiplicity, while the number of distinct zeros changes from one to three. In the degenerate case the two functions coincide.
Common mistake
(Rouché preserves the count with multiplicity, not the number of distinct zeros)
The preceding example shows the difference: has one distinct zero and has three. A conclusion such as “ has a unique zero in ” is justified only when the comparison function has total multiplicity there.
Practice and solutions
Quick check
(Count the zeros of in and in .)
Choose a different dominant term on each circle.
Solution · Answer
On , , so there is one zero in . On , , so there are four zeros in . Neither circle contains a zero, by the strict inequalities.
Quick check
(For , give a sufficient bound on a complex constant so that has exactly one zero in the unit disk.)
Solution · Answer
It is sufficient that . On the unit circle , and has exactly one zero in , which is simple.
Quick check
(Let . Compute with positive orientation.)
Solution · Answer
There is a double zero at and a triple pole at , both inside the circle and none on it; the integral is .
Quick check
(Does Rouché's theorem require throughout ?)
Solution · Answer
No. The hypothesis is a strict comparison on the boundary only; the interior is handled by holomorphy and the argument principle. In the quintic example, the comparison of with fails at inside the disk, yet the theorem applies because the strict inequality holds on .