Read circle coverings, homotopy lifting and the integer fundamental group of the circle. We use elementary complex arithmetic, polynomial division and compact real analysis. Rouché comparison here is proved for polynomials only; no general holomorphic-function theorem is assumed.
Motivation
The absence of a zero inside a disk has a visible boundary consequence. Normalize a nonzero function to the unit circle and shrink the boundary radially across the disk. Its boundary loop must have winding zero. A polynomial of positive degree, however, winds like its leading monomial on a sufficiently large circle. Those two facts cannot coexist if the polynomial has no root.
Once roots exist, each linear factor contributes either one or zero to the winding on a circle, according to which side contains its root. Boundary comparison then counts roots even when their individual formulas are inaccessible. This develops a concrete application of fundamental groups while keeping the hypotheses and the information being counted explicit.
Definitions
Definition 14.1.1
(Winding number of a nonzero loop)
For a continuous loop , normalize to . The covering map gives a continuous lift . Define
Two lifts differ by one constant integer, so the definition is independent of their initial values. Products below are pointwise complex products of loops; they need not share a chosen value at their parameter origins.
For and , orient the circle positively by . A polynomial root has multiplicity if with . The number of roots inside a circle always counts multiplicities. A root on the circle is excluded whenever boundary winding of the polynomial is used.
After existence of roots is proved, repeated polynomial division factors a degree polynomial as its leading coefficient times linear factors, with repetitions. This is a consequence of the first theorem, not an input smuggled into its proof.
Theorem/proposition
Theorem 14.1.2
(Winding proves the fundamental theorem of algebra)
Every nonconstant complex polynomial has a complex root. Hence a polynomial of degree factors into linear factors over , counted with multiplicity. On a circle containing no root, the winding of equals its number of roots in the open disk.
Theorem 14.1.3
(Polynomial Rouché comparison)
Let be complex polynomials. If on every point of the circle one has
then neither polynomial vanishes on that circle and they have the same number of roots in its interior, with multiplicity. The strict inequality is a boundary condition at every point, not an estimate at finitely many samples.
Theorem 14.1.4
(Local root stability, including collisions)
Let have degree , with distinct roots and multiplicities . Choose pairwise disjoint closed disks centered there, with no root on their boundaries. Every sufficiently small perturbation of its coefficients has degree and exactly roots in the th open disk. All perturbed roots are in these disks. A global continuous labeling or a Lipschitz displacement estimate is not asserted.
Proof sketch or proof idea
Proof
(Winding rules, then a large circle)
First establish the winding rules directly from lifts. The sum of lifts for two normalized loops lifts their product, so winding is additive under pointwise multiplication. A constant nonzero loop has winding zero. The loop has lift and winding for every integer , positive or negative.
Winding is unchanged by a homotopy of nonzero loops, even when their common endpoint value moves. Normalize the homotopy and lift it using circle homotopy lifting. The difference between its two endpoint lifts is an integer depending continuously on the homotopy parameter, hence is constant. If is a continuous nonzero function on the closed disk, the loops form such a homotopy from its boundary loop to the constant . Thus that boundary has winding zero.
Now let , with and . Put . Choose with . For ,
The straight homotopy from to remains nonzero on that circle by the triangle inequality. Its winding is therefore . If had no root anywhere, itself on the closed radius- disk would give the preceding zero-winding homotopy, contradicting . This proves root existence without factorization or an integral theorem. Divide by the resulting factor and repeat with the smaller-degree polynomial to obtain all factors. Division also makes the usual multiplicities precise.
Proof
(Count factors, then compare by a straight homotopy)
For one factor on the circle centered at , write . If , continuously shrink the second term to zero: all loops stay nonzero and winding is one. If , shrink the circular term to zero instead: winding is zero. The equality case would put a root on the boundary and is excluded. Winding additivity for the finite factorization now gives the root-count formula, including repeated factors.
For Rouché comparison use , with . On the boundary,
Thus and have equal boundary winding. The root-count formula proves equality of the interior counts. This argument does not require each intermediate polynomial to retain a fixed degree; the entire claim concerns finite-circle winding and the endpoint polynomials. A polynomial identically zero at an intermediate parameter would violate the displayed boundary bound and therefore cannot occur.
Proof
(A compact margin on the boundary circles)
For stability, let be the smallest of on the finite collection of chosen boundary circles. Compactness and absence of zeros give . Put larger than the absolute value of every point on those circles. Write and . If
then on every circle. Here are coefficients and the earlier are roots. Also require , ensuring degree . Rouché gives roots in each disk. Their sum is , so factorization leaves no further root elsewhere. This proves the precise neighborhood-of-coefficients assertion, including multiple roots which may split.
Proof X-ray
Which direction prevents circular reasoning
Root existence comes first from large-circle winding versus a hypothetical nonzero disk extension. Only afterwards do linear factors justify counting roots by winding. Rouché then compares counts, and compact boundary margins give stability. Factoring a polynomial before proving existence would reverse the essential first step.
Each hypothesis has one precise place. The strict inequality is used only to keep every intermediate nonzero on the circle, so that the straight homotopy is a homotopy of nonzero loops. Compactness of the circles is used only to turn “ has no zero there” into a positive margin , which is what makes a single coefficient neighborhood work. The degree condition on the perturbation is what lets the local counts add up to every root of .
Finiteness has a concrete use here. Because there are only finitely many boundary circles, a single positive minimum margin can be taken for all of them at once; because there are only finitely many coefficients, controlling each one uniformly controls the whole boundary error. Saying merely that each coefficient is close, without a common tolerance, would not yet prove that all root clusters are stable simultaneously.
One may compute winding by an integral when is piecewise . On a small parameter interval choose a differentiable argument , obtained from a smooth local inverse of the circle parametrization. Differentiating gives . Cover the compact parameter interval by finitely many such pieces and add the increments. Thus
This formula is a consequence of local arguments and lifting. It was not needed to assume a general analytic argument principle.
Worked examples
Worked example 14.1.5
(Five roots, with only one near the origin)
For , on the leading term has modulus 32 and the remainder at most 7. Comparison with gives five roots in the radius 2 disk. On compare with : the error is at most , strictly below 3/2. Exactly one root lies in the smaller disk. Therefore four roots lie in the open annulus between these circles, with no boundary roots. No formula for an individual root was used.
Worked example 14.1.6
(A double root splits at square-root scale)
The polynomial has a double root at 0. Perturb to with real. Its roots are , so any chosen small disk eventually contains both, exactly as stability says. Their displacement divided by the coefficient error is , unbounded as decreases to 0. The proved coefficient-neighborhood statement therefore cannot be strengthened to a uniform Lipschitz bound at a multiple root.
Worked example 14.1.7
(Root labels can exchange around a loop)
For the family , , one root beginning at 1 is ; it ends at although the endpoint polynomial equals the initial one. Any other continuous choice beginning at 1 agrees with it: their quotient takes values in the discrete two-point set of signs and is constant on the interval. Thus along a parameter loop the two labels exchange. Local clustering of roots does not supply one global continuous choice on all nonzero parameter values.
Worked example 14.1.8
(An explicit outer bound from coefficients)
Write a degree polynomial as times and let be the maximum of the absolute values of these normalized lower coefficients. Every root satisfies . Indeed, for , the entire lower sum on the radius- circle is at most
Rouché puts all roots inside that circle. If a root had modulus greater than , choosing strictly between these two real numbers would contradict that count. Thus the closed bound follows even though each comparison used a strict inequality. For , gives the safe bound 8; it is deliberately sufficient rather than optimal.
This bound gives a starting search region without locating a single root. Smaller disjoint circles and separate comparison estimates can then resolve clusters. An outer bound, a boundary margin and a root label are three different kinds of information: the first constrains where a root may be, the second certifies a stable count, and the third requires additional local or global choices.
Common mistakes
Common mistake
(Boundary zeros and multiplicities)
A winding computation requires that its loop avoid zero. Boundary zeros cannot be counted by a formula whose normalization is undefined there. Likewise multiplicity counts are not counts of distinct locations.
Common mistake
(The comparison must hold on the whole circle)
The comparison estimate must hold along the whole circle. Coefficient bounds and the triangle inequality provide a verifiable sufficient condition; numerical sampling alone does not provide that condition. Choosing a comparison polynomial is useful only after checking which term dominates on the specific radius.
Counterexample mode
Equality in the comparison can move a root onto the circle
Claim: Rouché's conclusion survives if the strict inequality is weakened to on the circle. Take , and the unit circle. Then at every boundary point, so the weak inequality holds. But has one root inside, while the only root of is , on the circle itself, so has no root inside and its boundary winding is not even defined. The proof needed on the circle for every ; at and this fails. The strict inequality is the repair.
Common mistake
(Stability is about unordered root clusters)
Root stability locates an unordered collection with multiplicities. It says neither that multiple roots persist unsplit nor that one can choose a single continuous global label. The square-root splitting and root-label examples separate these stronger requests from the theorem actually proved.
The same comparison counts zeros of functions holomorphic near a closed disk, once zeros are known to be isolated with finite multiplicity.
A path to generalization
From polynomials to holomorphic functions
The same comparison counts zeros of functions holomorphic near a closed disk, once zeros are known to be isolated with finite multiplicity.
If are holomorphic on a connected neighborhood of the closed disk and on its boundary circle, then and have the same number of zeros inside, with multiplicity. The proof above transfers once one knows that such a function, not identically zero, has finitely many zeros in the disk and factors near each as with . Establishing those facts belongs to complex analysis.
Summary
Circle lifting supplies an integer invariant, its product law and its homotopy invariance. Large-circle domination proves root existence, factorization turns winding into a root count, and a nonzero straight homotopy proves polynomial Rouché comparison. Positive compact boundary margins then give a precise coefficient-perturbation theorem without global labeling claims.
Exercises
Quick check
(Show that all four roots of lie in the annulus .)
Solution · Answer
At radius 2, exceeds , giving four interior roots. At radius , compare with the constant 2: , giving no interior roots. Both strict estimates exclude boundary zeros; all four lie in the stated open annulus.
Quick check
(Compute the winding of .)
Solution · Answer
The circle contracts to the nonzero constant 2 by shrinking its circular term, so has winding 0. Cubing multiplies winding by 3, and has winding . Product additivity gives for .
Quick check
(Prove the converse extension statement: a nonzero continuous loop with winding zero extends to a continuous nonzero function on the closed disk.)
Solution · Answer
Write , where and the zero-winding lift has . At polar point define the value to be times . Endpoint agreement makes the angular definition single-valued. The amplitude is positive; at its value is independent of angle. Boundedness of and the factor make continuity at the center uniform in . At it is , proving the extension.
Quick check
(For with , show that each radius disk about a cube root of 1 contains exactly one root.)
Solution · Answer
Distinct cube roots have distance . On the radius circle about one root, factorization gives . Thus the constant perturbation satisfies strict Rouché comparison on each circle. Each disk contained exactly one simple root before perturbation, hence contains one afterwards; their three roots exhaust the degree.
Quick check
(Show that for there is no continuous with everywhere.)
Solution · Answer
On the unit circle the equation forces . The loop has some integer winding , and its th power has winding . But its th power is , of winding 1. The equation is impossible for . This is a global obstruction, not a failure to choose roots in a small neighborhood.