The zeros of a polynomial control more than its value at those points. They also constrain the critical points where its derivative vanishes. Gauss–Lucas says that every critical point lies in the convex hull of the zeros. A logarithmic derivative gives a particularly concrete proof: a critical point that is not itself a zero is a weighted average of all the distinct zeros, with strictly positive weights.
On a line the geometry is even more precise. Between consecutive distinct zeros there is exactly one additional critical point, while repeated zeros contribute their own derivative multiplicities. We will keep those two sources of critical points separate.
The logarithmic derivative
Let be a nonconstant complex polynomial, with distinct zeros and positive multiplicities :
Away from the zeros, product differentiation gives
At a zero , write with . Then
The bracket is nonzero at , so is a critical point exactly when , and its multiplicity as a zero of is . Formula (1) is not evaluated at a zero; the local factorization handles that case.
Concept lensAnalytic
Critical points as zeros of a gradient field
Away from the zeros, is harmonic, and its gradient, written as a complex number, is :
A critical point that is not a zero is therefore exactly a point where this field vanishes. Each term points directly away from with length . Outside the convex hull, a separating line gives every a positive component in one fixed direction, so the sum cannot vanish there.
Theorem 1.9.1
(Gauss–Lucas and a positive barycentric formula)
Every zero of lies in the convex hull of the zeros of . More precisely, if and , then
Every weight in this average is strictly positive. If the zeros are not all collinear, every critical point that is not a zero of lies in the interior of their planar convex hull.
Proof
(Convert reciprocal vectors to positive weights)
At a critical point with , (1) gives
Taking complex conjugates and rearranging yields (2). Its denominator is positive, so (2) expresses as a convex combination of the zeros with all coefficients positive. If a critical point is itself a zero, it is already in the hull. This proves Gauss–Lucas.
If the zeros are not collinear, their hull has nonempty planar interior. Suppose the strictly positive combination in (2) lay on its boundary. A supporting line through that point places all zeros in one closed half-plane, with at least one zero strictly on the inner side. Let be a real affine functional measuring signed distance into that half-plane and vanishing on the supporting line. Then for all and for at least one; the strictly positive weights in (2) force , a contradiction. Hence the combination lies in the interior. Repeated zeros that are themselves critical points need not be interior; for instance has a critical point at the vertex of a noncollinear zero hull. Such points were excluded from the strict conclusion.
Proof X-ray
Why positivity is the geometric step
The proof has two stages. The first is algebra: (1) holds only away from the zeros, which is why the case is split off and handled by the local factorization. The key move is , which turns a sum of reciprocal vectors into a sum of difference vectors with weights ; conjugating the vanishing sum gives a linear relation among the with positive coefficients. Positivity of the multiplicities is used exactly here. With a pole, a negative weight enters and the barycentric formula fails, as the rational example below shows.
The second stage, interiority, uses strict positivity of every weight. A supporting functional that is nonnegative at all zeros and positive at one cannot vanish at a strictly positive combination. Noncollinearity is used only to guarantee that such a zero off the supporting line exists.
Collinear zeros and interlacing
Theorem 1.9.2
(Interlacing with multiplicities)
Suppose the distinct zeros of lie on a line. After an affine change of coordinates, write them as real numbers with multiplicities . Then has:
- a zero of multiplicity at each with ; and
- exactly one simple zero in each open interval .
These account for all critical points, counted with multiplicity. When there are no intervals and the repeated-zero contribution is the whole list.
Proof
(The logarithmic derivative decreases between zeros)
After translating and rotating the line, the polynomial has the form in a real coordinate . Its logarithmic derivative is real on the real line away from the zeros:
On , at the left endpoint and at the right. Continuity and strict decrease give exactly one root there. Since and , differentiating at gives , so this zero of is simple.
The original zeros contribute derivative zeros, and the gaps contribute . Their total is , the degree of . No other critical points remain. An affine rotation and translation preserve the multiplicities and map the line back to its original position. For at least two distinct zeros, every critical point that is not a zero of lies in the relative interior of the segment joining the extreme zeros; repeated endpoint zeros themselves may be critical.
Examples and limits
Worked example 1.9.3
(A critical point inside a triangle)
For , the zeros are the three cube roots of unity. Its derivative has a double zero at the origin. The origin is the average of the three roots, and lies strictly inside their triangular hull. Multiplicity of a critical point does not imply that it is a zero of .
Worked example 1.9.4
(Repeated zeros and one gap root)
Let . The root has multiplicity two, so it contributes one zero of . The logarithmic derivative on is
It vanishes at , the unique gap critical point. Indeed . The repeated root lies at an endpoint of the hull; the other critical point lies strictly between the distinct zeros.
Worked example 1.9.5
(Critical points of )
Let , whose zeros are not collinear. Then , with discriminant . Hence
that is, approximately and . Both have positive real and imaginary parts with sum , so both lie strictly inside the triangle with vertices , as the theorem predicts for simple noncollinear zeros. Their average equals the average of the zeros, a general fact read off from the coefficients of and .
Counterexample mode
The polynomial hypothesis cannot be omitted
The entire function has exactly one zero, at , but
vanishes at , outside the singleton convex hull . Indeed away from its zero; the extra constant term comes from . Thus unlike a polynomial logarithmic derivative, it is not a sum of positive reciprocal contributions from its zeros alone.
A rational function can fail too. For ,
so the critical point lies outside the convex hull of its zeros, although the other critical point lies inside. Poles enter the logarithmic derivative with negative signs, and the positive-average proof no longer applies.
Common mistake
(Not every point of the hull is a critical point)
The convex-hull inclusion is a necessary geometric constraint, not a characterization of all points in the hull. For , the hull is the entire interval , but the only critical point is .
Common mistake
(Interlacing needs collinear zeros)
For zeros on a line there is exactly one critical point between consecutive zeros. Off a line there is no such rule: the zeros of are the vertices of a triangle, yet has no zero on any of its edges; its only critical point is the center, with multiplicity two.
Exercises
Quick check
(Find the critical points of , with multiplicity, and locate them relative to the zeros.)
Solution · Answer
The triple zero contributes a double critical point there. The logarithmic derivative vanishes at , the unique point in . Equivalently , and .
Quick check
(Prove that if all zeros of a nonconstant polynomial lie in the closed disk , then all its critical points lie there too.)
Solution · Answer
A closed disk is convex, so it contains the convex hull of the zeros. Gauss–Lucas puts every critical point of the nonconstant polynomial in that hull. The nonconstant hypothesis matters: a constant polynomial has , so every point is critical under the definition .
Quick check
(For , verify the barycentric identity (2) at its critical point .)
Solution · Answer
and , so (2) applies. Each root is at distance one from the origin and has multiplicity one, so (2) reduces to the average of the three cube roots of unity. Their sum is zero.
Quick check
(Explain why does not contradict Gauss–Lucas and compute its unique critical point.)
Solution · Answer
The theorem assumes a polynomial; this function is entire but not polynomial. Since , its unique critical point is , while its only zero is .