The Real Curve Zoo & Hilbert's 16th

How many ovals can a plane curve have, and how can they nest? A zoo of wild curves ending at an open problem

The Real Curve Zoo & Hilbert's 16th

Set a polynomial in two variables equal to zero and draw the real solutions. What you get is a curve — and depending on the polynomial, it might be a single sweeping arc, a figure-eight, a four-petalled rose, or a scattering of closed loops that geometers call ovals.

Two questions follow immediately, and they turn out to be wildly different in difficulty. The first: how many ovals can a curve of degree d have? Harnack answered that in 1876, while still a student. The second: how can those ovals be arranged? Which ones can sit inside which others? That question went on Hilbert’s famous list of problems in 1900 as the sixteenth, and it is still not answered.

This is the rarest kind of open problem: one where the objects are completely elementary, you can draw them by hand, and the frontier of human knowledge sits at degree eight.

Interactive: The Zoo

Eight classical curves, each drawn straight from its implicit equation. The Cassini ovals are the ones to linger on — slide the parameter and watch two separate ovals fuse into one, passing through the lemniscate at the exact moment of contact.

Lemniscate of Bernoullidegree 4
(x² + y²)² = 2a²(x² − y²)

The figure-eight Jacob Bernoulli published in 1694, describing it as a ribbon; the points where the product of the distances to two fixed foci stays constant.

Every one of these is the zero set of a polynomial in x and y — no trigonometry, no parametrisation, just an equation. The petals, cusps, loops, and ovals are all consequences of algebra. Cusps are where the curve is singular, which is why the astroid and deltoid need higher degree than their smooth-looking outlines suggest.

Harnack’s Bound, and Why Ovals Are the Right Thing to Count

A smooth real projective curve breaks into finitely many connected pieces. In the projective plane, a piece either bounds a disc — an oval — or it does not, and there can be at most one of the latter kind. So counting pieces is essentially counting ovals, and it is a genuine topological invariant of the curve.

Harnack’s theorem (1876): a smooth curve of degree d has at most (d−1)(d−2)/2 + 1 components. The expression is not arbitrary — it is the genus of a smooth plane curve of degree d, plus one. The complex curve is a surface with g handles, complex conjugation acts on it, and the real curve is the fixed set; a theorem about how an involution can act on a genus-g surface caps the number of fixed circles at g + 1.

Harnack also proved the bound is sharp at every degree, by constructing curves that reach it. Those are the M-curves — maximal curves. His method is the perturbation trick in the next demo: build a deliberately singular curve as a product of simpler ones, then nudge it so every singular point bursts open.

Interactive: Watching Ovals Be Born

Multiply a few ellipses together and you get a singular curve whose nodes sit at the crossing points. Add a small constant ε and every node resolves at once. Sweep ε through zero and watch the arrangement flip — the oval count is measured from the picture by flood-fill, so the number shown is always the honest one.

grey: the singular union  ·  magenta: the perturbed curve
9
ovals (measured)
11
Harnack’s bound for degree 6
ε < 0singularε > 0
Nodes opened one way

Every node has resolved into a small arc, and the sign of ε decides which of the two ways each one opens. That single choice is the difference between 9 ovals and 5 — same curve, same degree, wildly different topology.

This construction tops out at 9 ovals, short of Harnack’s 11. That is a genuine limit of perturbing by a constant ε: reaching the maximum needs Harnack’s finer method, where the perturbing term is itself a polynomial chosen to open the nodes independently. The number above is measured from the picture, so it never flatters the construction.

Harnack’s theorem (1876): a smooth real curve of degree d has at most (d−1)(d−2)/2 + 1 connected components, and the bound is always attained. Curves that reach it are called M-curves. The bound is the easy half of the story — the hard half, which is still open, is how those ovals are allowed to be arranged.

The Hard Half: Arrangement

Knowing a sextic can have 11 ovals says nothing about how they sit relative to one another. Ovals can nest — one enclosing another — and the pattern of nesting is what Hilbert asked about. For degree 6 with the maximal 11 ovals there turn out to be exactly three possibilities, and getting to that answer took nearly a century.

Harnack exhibited one arrangement in 1876. Hilbert found a second in 1891 and believed that was the end of it. He was wrong: Gudkov constructed a third in 1969 and proved the list complete. Gudkov also spotted a startling pattern in the data — for an M-curve of even degree d = 2k, the number of ovals of one nesting parity satisfies a congruence mod 8 — and that observation, later proved by Arnold and Rokhlin, opened up the whole modern approach via topology of four-manifolds.

The other great surprise was a demolition. In 1906 Ragsdale conjectured bounds on how ovals of each parity could be distributed. It stood for most of a century before Itenberg produced counterexamples in the 1990s, built with Viro’s patchworking — a technique for assembling curves out of combinatorial pieces that is a close cousin of the tropical geometry in this module’s amoebas lesson. The stick figures turned out to be strong enough to break a ninety-year-old conjecture.

Interactive: Three Ways to Arrange Eleven Ovals

The three maximal sextic schemes, and the seventy-eight years between Harnack’s first and Gudkov’s last. Below them, how far the classification has actually got — which is not very far.

1 oval containing 19 free
Harnack's schemeAxel Harnack, 1876
⟨9 ⊔ 1⟨1⟩⟩  —  11 ovals

Harnack proved the bound of 11 ovals for a sextic and built a curve attaining it, with nine ovals lying free and one more tucked inside an eleventh.

Same degree, same oval count, different worlds

All three of these are smooth curves of degree 6 with the maximum 11 ovals. Harnack settled how many ovals there can be in 1876. But how they may be nested is a different question entirely, and it is the one Hilbert put on his famous list in 1900 as the sixteenth problem. He believed only two arrangements were possible. Gudkov found the third in 1969 and proved that the list stops at three.

How far the classification has got
Degree 2
One oval, or nothing. Trivial.
classified
Degree 4
Up to 4 ovals, never nested more than once.
classified
Degree 6
Settled by Gudkov in 1969 — exactly three maximal schemes.
classified
Degree 7
Settled by Viro in 1979.
classified
Degree 8
Still open today. Many schemes remain undecided.
open
Degree 9 and up
Wide open.
open

The main tool for building curves with prescribed ovals is Viro’s patchworking (1979): glue curves together from combinatorial pieces, in a construction closely related to the tropical geometry in this module’s amoebas lesson. Patchworking also demolished the Ragsdale conjecture of 1906 — Itenberg used it in the 1990s to build counterexamples, more than eighty years after the conjecture was posed.

Key Takeaways

  • Harnack’s bound (1876): a smooth real curve of degree d has at most (d−1)(d−2)/2 + 1 components — the genus plus one — and the bound is attained at every degree.
  • M-curves are the maximal ones, built by perturbing a product of simpler curves so that every node bursts open into an arc.
  • The sign of the perturbation matters enormously: flipping ε changes which way each node opens, and with it the whole topology of the curve.
  • Hilbert’s 16th asks about arrangement, not count: which nestings of ovals are actually realisable by a curve of degree d?
  • Degree 6 has exactly three maximal schemes — Harnack’s, Hilbert’s, and the one Gudkov found in 1969, disproving Hilbert’s belief that there were only two.
  • Degree 8 is still open, and so is everything above it. Viro’s patchworking is the main construction tool, and it is what Itenberg used to demolish the Ragsdale conjecture.