Flecnode polynomial / ruled-surface geometry (G. Salmon)

Statement

Flecnode (definition). Let $p:\mathbb C^3\to\mathbb C$ (or $\mathbb R^3$) be a polynomial and $Z=\{p=0\}$ its zero-set surface. A point $w\in Z$ is a flecnode of $Z$ if there is a line $\ell = \{w+tv : t\in\mathbb C\}$ through $w$ such that $p$ restricted to $\ell$ vanishes to order $\ge 3$ at $t=0$ — i.e. $p(w)=0$, the directional derivative $\nabla p(w)\cdot v = 0$, *and* the second directional derivative $v^{\mathsf T}\mathrm{Hess}(p)(w)\,v=0$. (This is the surface analogue of a flex point of a plane curve: third-order, not merely first- or second-order, tangency.)

Flecnode polynomial (Guth, arXiv:1404.3412, Theorem 6). For any polynomial $p$ of degree $d$ in $\mathbb C^3$, elementary elimination theory (eliminating the direction $v$ from the three polynomial conditions above) produces a single polynomial $\mathrm{Flec}(p)(x,y,z)$, of degree at most $11d-24$, that vanishes at $w$ whenever $w$ is a flecnode of $Z=\{p=0\}$.

Cayley–Salmon theorem / Monge–Cayley–Salmon theorem. If $\mathrm{Flec}(p)$ vanishes identically on a Zariski-dense subset of the (irreducible, 2-dimensional) zero set of $p$, then that zero set is ruled: through every point of $Z$ there passes a line entirely contained in $Z$.

Corollary (line-counting form, Guth's Corollary 8). If an irreducible surface of degree $d$ in $\mathbb C^3$ contains more than $11d^2-24d$ complex lines, then the surface must be ruled. (Mechanism: every line lying in $Z=\{p=0\}$ automatically lies in $\{\mathrm{Flec}(p)=0\}$ too — a line contained in $Z$ trivially has infinite-order, hence third-order, contact with $Z$ at every one of its points — so once enough such lines are present, $\mathrm{Flec}(p)$ and $p$ must share a common factor / $\mathrm{Flec}(p)$ vanishes on all of $Z$, forcing the ruling conclusion via Bézout's theorem applied to a generic 2-dimensional slice.)

Facts

- Historical origin. Attributed to the Rev. George Salmon (*A Treatise on the Analytic Geometry of Three Dimensions*, 19th century) and Cayley; Tao's differential-geometric reproof credits an even earlier precursor to Monge, hence "Monge–Cayley–Salmon theorem." Distinct from — but related to — Salmon's more famous 1849 theorem (with Cayley) that every smooth cubic surface contains exactly 27 lines: that is a *finite*-line-count fact about smooth cubics, whereas the flecnode/ruling theorem is the *dichotomy* statement (too many lines forces infinitely many, i.e. a ruling) that applies to surfaces of any degree. - Construction is pure elimination theory. The three flecnode conditions ($p(w)=0$, first directional derivative $=0$, second directional derivative $=0$) are polynomial in $(w,v)$; resultant/elimination-theory computations remove the direction variable $v$, leaving a polynomial purely in $w=(x,y,z)$ — this is mechanically why a *single* algebraic condition (the flecnode polynomial) can certify "some line through $w$ has 3rd-order contact," without ever exhibiting the line itself. - The joints problem (Guth–Katz, arXiv:0908.3261, 'On the joints conjecture'). A *joint* is a point incident to three lines (from a given family of $N$ lines in $\mathbb R^3$) that are not coplanar. The flecnode polynomial is the tool used to handle points of *low* incidence multiplicity (specifically, points where only two of the family's lines meet) inside the Guth–Katz polynomial-partitioning argument: after a low-degree polynomial's zero set is shown to contain "too many" lines from the family, the Cayley–Salmon dichotomy forces that zero set to be ruled, which lets the argument reduce to counting on a 1-parameter family of lines rather than an arbitrary surface — closing the induction and yielding the (tight) bound of $O(N^{3/2})$ joints. - **The distinct-distances problem (Guth–Katz 2015, arXiv:1011.4105, resolving Erdős #89 — distinct distances in the plane up to a $\sqrt{\log n}$ factor). Via the Elekes–Sharir framework**, planar distinct-distance counting is reduced to a point-line incidence problem in the 3-dimensional group $SE(2)$ of rigid motions. Guth–Katz's cell decomposition (via the polynomial ham-sandwich theorem) handles points of *high* line-multiplicity; the flecnode polynomial + Cayley–Salmon ruling handles the *low*-multiplicity (two-line) points, showing that if too many two-line points survive, the relevant lines concentrate on a ruled surface, whose geometry (the surface parametrizing rigid motions taking one fixed line to another is itself governed by a low-degree equation) is then used directly to bound the incidence count. - **Why this matters for Erdős #604 — pinned distinct-distances at a single point (the pinned-distance strengthening of #89, still open):** the ruled-surface argument here is applied *aggregated over all base points at once* (via the SE(2) reduction), which is exactly why the technique's output does not obviously localize to a bound for a single pinned point — the unresolved technical gap documented in wiki/problems/604.md. - Generalization. A later paper, "Extactic divisors for webs and lines on projective surfaces" (arXiv:1509.05435), generalizes the flecnode-polynomial construction (Guth's degree-$(11d-24)$ elimination-theory object, which detects *third*-order line-contact) to an "extactic divisor" detecting higher-order contact and contact with more general 1-dimensional foliations/webs, not just lines — abstract-level association only; primary statements not independently re-verified this session (PDF unparseable by fetch tooling).

Technique

WHY it works (the mechanism). Tao's differential-geometric reproof of the Cayley–Salmon dichotomy makes the "why" geometric rather than purely algebraic: wherever flecnodes are Zariski-dense on the surface, one can locally extract a smooth *direction field* (the flex direction at each point) and integrate it (Picard existence theorem) into a family of curves lying on the surface. The third-order-vanishing condition defining a flecnode forces the first three derivatives of such an integral curve to be orthogonal to the surface's gradient — i.e. the curve has zero torsion, hence lies in a plane; the second-order part of the same condition then forces the curve's curvature to vanish within that plane too, so the curve is in fact a straight line. Density of flecnodes therefore densely — and then, by the closedness of "being a line," everywhere — rules the surface.

HOW to use it to prove things (recombination steps)

1. Set up an incidence problem about many lines in $\mathbb R^3$/$\mathbb C^3$ (or a problem reducible to one, e.g. via the Elekes–Sharir trick that turns a planar point-configuration question into a line-incidence question in the 3-dimensional group of rigid motions $SE(2)$). 2. Produce a low-degree surface containing many of the lines. Typically via a vanishing lemma / dimension-counting argument (e.g. "a nonzero polynomial of degree $O(N^{1/3})$ can be forced to vanish on $N$ given points/lines by a linear-algebra/pigeonhole count") — this is where the *number* of lines gets converted into a *degree bound* $d$ on a polynomial $p$. 3. Apply the degree threshold from Corollary 8: if the number of lines lying on $\{p=0\}$ exceeds $11d^2-24d$, invoke Cayley–Salmon to conclude the surface is ruled. 4. Exploit the ruling. A ruled surface is (locally, away from finitely many singular points) a *one-parameter* family of lines — this collapses a 2-dimensional incidence problem (lines on a surface) to a 1-dimensional one (lines in a family parametrized by a curve), which is tractable by direct algebraic/geometric analysis (as in Guth–Katz's joints argument) or feeds back into the ambient reduction (as in the SE(2) distinct-distances argument). 5. Combine with a complementary high-multiplicity argument (typically a cell decomposition via the polynomial ham-sandwich theorem, Polynomial method / slice rank (Croot-Lev-Pach, Ellenberg-Gijswijt)) to close the case where the ruling conclusion doesn't directly apply — the flecnode polynomial handles only the *low*-incidence-multiplicity regime; it is always used as one half of a dichotomy, never alone.

WHEN it applies: any incidence-geometry problem about a large family of *lines* (or curves reducible to lines by an algebraic change of coordinates, as SE(2)-rigid-motions become lines under Elekes–Sharir) in 3-dimensional space, where a low-degree polynomial can first be shown to contain "too many" of them. WHEN it does not directly transfer: problems that don't reduce to *lines-on-a-surface* (higher-dimensional or non-linear analogues need the extactic-divisor-style generalizations, arXiv:1509.05435, or entirely different machinery); and — per the open Erdős #604 — pinned distinct-distances at a single point gap — problems requiring the incidence bound to *localize* to a single distinguished point/line rather than being summed over the whole family, since the standard ruled-surface argument is inherently an aggregate statement.

Related

- Polynomial method / slice rank (Croot-Lev-Pach, Ellenberg-Gijswijt) — the umbrella technique (polynomial ham-sandwich cell decomposition + algebraic incidence bounds) that the flecnode polynomial is the complementary "low-multiplicity" tool for, in both the joints problem and Guth–Katz's distinct-distances proof. - Erdős #89 — distinct distances in the plane — Erdős distinct-distances problem in the plane; resolved up to a $\sqrt{\log n}$ factor by Guth–Katz 2015 using exactly this technique (via the Elekes–Sharir reduction). Already forward-links to `Flecnode polynomial / ruled-surface geometry (G. Salmon)`. - Erdős #604 — pinned distinct-distances at a single point — pinned-distance strengthening of #89, still open; the flecnode/ruled-surface argument's inherently *aggregated* (non-localizing) nature is documented there as the concrete technical obstruction to transferring the Guth–Katz method. - Concept referenced but not yet its own page: "Elekes–Sharir framework" (already forward-linked as `Elekes–Sharir(–Guth–Katz) reduction: distinct distances → point-line incidences in SE(2) from wiki/problems/89.md and wiki/problems/604.md) — the reduction that turns planar distinct-distance counting into a line-incidence problem this technique can attack; a natural next concept page. - Concept referenced but not yet its own page: "polynomial partitioning" (already forward-linked as concept/polynomial-partitioning` from wiki/problems/604.md) — the cell-decomposition machinery this technique pairs with. - The joints problem (Guth–Katz, arXiv:0908.3261; O(N^{3/2}) joints among N lines, tight) — the original application of the flecnode polynomial, not yet tracked as its own numbered Erdős problem in this wiki.